EVERY SPOKEN WORD
145 min read · 29,261 words- 0:00 – 8:56
Why AWS matters: Amazon’s profit engine hiding in plain sight
- DRDavid Rosenthal
People, turns out, loved the amazon.com episode. That was so awesome. Makes me a little nervous for this one.
- BGBen Gilbert
Oh, massively. By far and away, our biggest episode ever. Uh, is this how George Lucas felt when he was doing Empire Strikes Back? [laughing]
- DRDavid Rosenthal
[laughing] You did not just compare us to George Lucas, did you? [laughing] I swear we're humble.
- BGBen Gilbert
All right, let's do this.
- SPSpeaker
Who got the truth? Is it you? Is it you? Is it you? Who got the truth now? Is it you? Is it you? Is it you? Sit me down, say it straight, another story on the way. Who got the truth?
- BGBen Gilbert
Welcome to Season 11, Episode 3 of Acquired, the podcast about great technology companies and the stories and playbooks behind them. I'm Ben Gilbert, and I am the co-founder and managing director of Seattle-based Pioneer Square Labs, and our venture fund, PSL Ventures.
- DRDavid Rosenthal
And I'm David Rosenthal, and I am an angel investor based in San Francisco. Cold San Francisco, here in August. [chuckles]
- BGBen Gilbert
[chuckles] And we are your hosts. All right, David, let's say you run a lemonade stand. You sell me the highest quality lemonade you can for the lowest price, one dollar a cup, and when you add up all your costs, the variable ones, like the lemons, and the fixed ones, like the table that you rented, it costs about ninety-eight and a half cents to give me that lemonade. And you're happy with your turn to profit, I'm sure, but man, you are gonna have to sell a lot of lemonade.
- DRDavid Rosenthal
So you're telling me I'm amazon.com in the fourth quarter of two thousand and one, which is actually where we're gonna start our story.
- BGBen Gilbert
Perhaps. But you discover something interesting. By making all this lemonade, you get really good at the stuff it takes to run a lemonade business, the perfect cups, and ice, and lemons, everything, and it turns out, all that stuff that you just got good at, you can sell to other businesses. And guess what? You realize further that when you sell your services to other companies, when you charge them a dollar, it only costs you seventy cents to make it. So thirty percent margins instead of something like a percent and a half, you're gonna have to sell a lot less of those services than you ever did on lemonade to make the same amount of money.
- DRDavid Rosenthal
Well, then, if you told me that, I would dig into it even further, and I would realize that the existing companies that sold stands, and cups, and whatnot, they were actually making seventy percent margins on their stands and cups. [chuckles]
- BGBen Gilbert
[chuckles]
- DRDavid Rosenthal
And so I would be quite happy to take thirty percent margins and disrupt them, and still do better than my lemonade business.
- BGBen Gilbert
Well, listeners, of course, on our last episode, we talked about Amazon's retail business, and today, we are talking about Amazon Web Services, the cloud computing pioneer, and those margin percentages that I just used are the real ones for the retail business and for AWS. AWS's revenue is only about fifteen percent the size of Amazon's massive retail business, but their profits, or the operating income, to be specific, from AWS, are, in total, the same, if not more, than their e-commerce store. I think it's the case that every year since twenty fifteen, when they started breaking out AWS's financials, the total operating income from AWS has actually been bigger than [chuckles] the retail business.
- DRDavid Rosenthal
There may have been some quarters where it was off, but generally, that trend is accurate.
- BGBen Gilbert
Wild. So we're gonna talk about a completely different type of business today than we talked about last time. Sort of... There's a lot of similarities, and a lot more than you would sort of guess when looking at an online retailer that started as an online bookstore and a cloud computing pioneer. Well, speaking of e-commerce, we have huge news. You can finally, finally buy Acquired merch on the internet. That is available at acquired.fm/store, or click the link in the show notes. You can grab your favorite tee, crew neck, hoodie, tank, or even a onesie, since I know a lot of you out there are like David and have little ones at home. Now, for our presenting sponsor this episode, we have a company that we are very excited about, Fundrise. On our amazon.com episode, they shared some news about a fascinating new product they have called the Fundrise Innovation Fund that enables their customers not only to invest in real estate, but also now private, late-stage growth tech companies, which is, of course, a very interesting place to be investing right now, now that prices are very different than they were six and twelve months ago. We are back today with Ben Miller, CEO and co-founder of Fundrise.
- DRDavid Rosenthal
We know that you, Ben, personally love dissecting investing markets from first principles, and you've been noodling on how to disrupt VC for a while. Tell us why you think growth VC in the traditional sense is broken, and why what you're doing with retail investing in private companies is the future.
- SPSpeaker
When you wanna start a big business, you find a big problem. You find something that's broken. Twenty percent of the upside of a company's success goes to the people who work at the VC fund, twenty percent of carried interest. So who creates the value? People who build the company, people who invest in the company. So the team, in my view, creates the value. They put in two thousand five hundred hours a year. How many hours does a typical VC put into a company? A hundred? Fifty? So that's, like, two percent of what a typical employee does. So what happens when, like, a sector is, like, overcompensated, overpaid? It gets overcrowded, or it gets overfunded. It gets undisciplined. When you look at twenty twenty-one, you can see it in what's happening in venture, is because it's overcompensated, it's basically gotten messed up. So we basically created the Fundrise Innovation Fund with the idea of eliminating the carried interest, give it back to the team, give it back to the employees. So obviously, capital deserves a good return for taking the risk.... but twenty percent of that return shouldn't necessarily go to GPs who didn't actually put up the money, didn't actually do the work. In the long term, markets get more efficient, so probably those returns get split between the investors and the team members. But our mission is basically to lower basically the fees of the intermediaries and increase basically the benefits to the people who are taking the risk and doing the work. And that is really, I think, the future of venture. I mean, it's like a nightmare for the venture guys. The worst thing that could ever happen to them is to lose their twenty percent carried interest. It's the most disruptive thing we could do, if you think about us being a disruptor. But yet, the future of markets is markets become more efficient. The vast majority of public stock managers do not take a twenty percent carried interest. It's not justified by the efficiency of the market. And when the venture industry started thirty, forty years ago, the market was inefficient. There was very little money, and it was justified. But as the market's gotten much, much larger, huge, and all the data, every company, it's all available to anybody, it's just not justified anymore. So you can do the job and still have successful amount of profits without taking the twenty percent carried interest. I mean, the biggest public asset manager is BlackRock, and you don't see them taking a carried interest, right? They have, uh, AUM fee, but by having really good efficient systems and scale, they're a great business. And so it's structure and incentives dictate behavior. It's like what Charlie Munger says, and so we're trying to restructure the whole venture industry.
- BGBen Gilbert
Our thanks to Fundrise, the largest private investment platform in the world for retail investors. If you wanna join the over three hundred and fifty thousand individuals investing with Fundrise, you can click the link in the show notes. And if you're a founder and you wanna get in touch about having the Innovation Fund participate in your next funding round, email notvc, that's notvc@fundrise.com. After you finish this episode, come discuss it with David and I, and thirteen thousand other smart members of the Acquired community at acquired.fm/slack. And if you're dying for more Acquired in the meantime, go check out The LP Show by searching Acquired LP in any podcast player. The next episode is with David's partner in crime and Kindergarten Ventures, Nat Manning, talking about his company, Kettle, and how the business of reinsurance works. That, of course, is already live if you are a paying LP, which you can become at acquired.fm/lp. Now, without further ado, David, take us in. And listeners, as always, this show is not investment advice. David and I may, certainly do, have investments in the companies we discuss, and this show is for informational and entertainment purposes only.
- 8:56 – 13:29
Rewinding to 2001–2002: Amazon’s post–dot-com survival context and the scale problem
- DRDavid Rosenthal
Well, we left off the amazon.com episode in 2007 with the sort of Sony PlayStation-like coda of the Kindle story and the new chapter, one might say, that it seemed at the time to the outside world that Amazon was opening as a true technology company with the Kindle. I believe the quote from Eric Schmidt in The Everything Store was, "The book guys finally got technology." [laughing] And of course, as we talked about, Jeff Bezos always got technology. This was not a shift. And in particular, this was not anything new because of everything we are going to talk about on this whole separate episode today. So to do that, we need to rewind back, as I said above, to the end of 2001, kinda early 2002, the immediate post-dot-com bubble popping crash era. And Bezos and Amazon, as hard as it is now to remember, he was like an embattled CEO at this point. They had just gotten rid of COO Joe Galli. The board has brought in Coach Campbell. Amazon's fighting for its life against both eBay and Wall Street.
- BGBen Gilbert
Is it insane to think that the board was sort of in the place with Jeff Bezos, thinking, "We really need some adult supervision to be a, uh, scale CEO and help this guy out"? Frigging Jeff Bezos! Obviously, that did not pan out, and Bezos came valiantly riding back in and ran the business for another twenty years.
- DRDavid Rosenthal
Another twenty years, until handing the reins to somebody else, who we're going to spend a lot of time in just a little bit here talking about, of course, current Amazon CEO Andy Jassy.
- BGBen Gilbert
Yep.
- DRDavid Rosenthal
So I don't even know what the right word is to use to describe AWS. I was gonna say, I wrote behemoth in my notes.
- BGBen Gilbert
Pioneer, inventor.
- DRDavid Rosenthal
I don't think there's anything you can say that captures how big and how important AWS is. It is one of the biggest and most important businesses, technologies, products of the modern world.
- BGBen Gilbert
Yep, no doubt.
- DRDavid Rosenthal
I don't think it's controversial to say even much more so than amazon.com.
- BGBen Gilbert
Yeah. I mean, it's interesting. During the pandemic, you could argue that amazon.com was more important because everybody needed to sort of buy goods and get them at home.
- DRDavid Rosenthal
But everybody also needed to be on the internet, and the internet [chuckles] runs on AWS.
- BGBen Gilbert
Yeah.
- DRDavid Rosenthal
So today, we're gonna tell that story. It's funny, as we did the research... So there's no, like, everything store book dedicated to AWS. There are a lot of very disparate resources and stories out there, and there actually are quite a few conflicting and competing stories about what the true origin is of AWS. You might say it has a cloudy origin. See what we did there?
- BGBen Gilbert
Ayoo.
- DRDavid Rosenthal
Ayoo. [chuckles]
- BGBen Gilbert
It is true. As we were doing the research, you know, of course, David and I read both of Brad Stone's excellent books. I watched the PBS FRONTLINE documentary, which, of course, has a very specific angle that they're trying to take on the company. And you sort of read any of these Amazon analysis pieces-... they're like ninety-five percent about the retail business, and they'll talk about things like the relationship with employees and the big New York Times piece that came out in twenty fifteen. And they'll talk about the relationship with the warehouse workers, or, "Is this good for the world?" And everyone indexes on that, which is important and deserves all the attention it's got. But almost none of these spend a material amount of time on AWS, other than mostly an apocryphal founding story, which is not even really how it happened.
- DRDavid Rosenthal
So we identified, you're referring to one origin story of AWS. We identified not one, not two, not three, but four [chuckles] separate origin stories, and we're gonna tell them all here. I think there is something important to learn about what AWS is, and about Amazon, and about Amazon culture in all of these. So let's start with the first, and most obviously untrue one, [laughing]
- BGBen Gilbert
[laughing]
- DRDavid Rosenthal
which is ironically also the one that the layperson believes the most.
- BGBen Gilbert
Yes, because it's tempting. I mean, it's like an, "Oh!" Like, it's too convenient.
- 13:29 – 20:34
Origin story #1 (myth): “AWS sold Amazon’s excess server capacity”
- DRDavid Rosenthal
Yes, and that story is the excess capacity narrative. So the way this story goes is that right around this time, two thousand and one, two thousand and two, two thousand and three, Amazon.com, the retail business, like all retail businesses in America at least, is highly seasonal. They have huge spikes of traffic and demand in Q4 for the holiday shopping season, and that's when most of, maybe not most, but the largest share of any quarter revenue happens in Q4.
- BGBen Gilbert
So much so that for the first at least five years of the business, there was a rule in November and December that you could not commit new code to production.
- DRDavid Rosenthal
That's right.
- BGBen Gilbert
It was so all hands on deck that no new features were allowed unless it was a red flag bug fix.
- DRDavid Rosenthal
Oh, and we didn't talk about this in the Amazon.com episode, but for years and years and years, the executive team, and the business side of the company, and the engineers, everybody would go work in the warehouses in Q4.
- BGBen Gilbert
That or customer service. Oh, how times have changed. Can you imagine someone sitting down in Day 1 North or Doppler being told that they have to go pick and pack for a while?
- DRDavid Rosenthal
I think for a while they continued to do it even when it wasn't necessary, just as, like, a culture thing, but obviously, those days are gone now. So the urban legend is that because of this dynamic, Amazon had this brilliant realization around then, again, when they're trying to achieve profitability, that they had excess technical infrastructure capacity in their IT operations during quarters one through three. So they had to build out for the peak demand of Q4, all the traffic on the website, all the transactions happening, but the rest of the year, all that capacity was just sitting there, and so they decided, "Let's rent out that capacity to other developers." Brilliant, brilliant! We are gonna turn a large expense line in the business into a revenue line.
- BGBen Gilbert
Magic. And of course, this falls down in two enormous places. One is, if you've ever been inside a pre-cloud technology company, you know that-
- DRDavid Rosenthal
It doesn't work that way. [chuckles]
- BGBen Gilbert
Yeah, you can't just say, like, "Oh, cool, like, the servers aren't in use right now, and there's nothing highly customized about these servers, and they're not tightly coupled to our applications in any way. So we'll just make it so that anyone can very easily just run their applications on it, and there's enough security set up correctly so that anyone can just get access to our data center, and the network hardware sort of understands how to serve other tenants other than us." None of that existed, and none of that was true, so there's just no way you can be like, "Oh, yeah, other companies just started using our infrastructure, and it was pretty rip and replace."
- DRDavid Rosenthal
In a pre-cloud infrastructure world, you installed your software, your code base, on your servers [chuckles] that you owned. The Amazon.com code base was literally installed on a bunch of boxes that they owned. You couldn't just rent out the capacity.
- BGBen Gilbert
Until two thousand, they were servers from Digital Equipment Corporation, DEC. They were DEC Alpha servers. These were unbelievably high-margin servers that you, I believe, leased from the manufacturer. It was the same business model that IBM had forever and Oracle has or had forever, where you get this highly bundled hardware, software platform that you would use to run your applications, and they would make eighty percent gross margins on these things. There was this massive markup. They were monolithic, and to be honest, the thing that really changed all this was Linux. When Linux came out, and you could do the stuff that you used to need Unix workstations for on an open-source operating system, well, then everything changed, 'cause you can go buy a whole bunch of different hardware, put Linux on it, and then write your own applications. And so this laid the groundwork for maybe infrastructure doesn't have to be as insanely expensive, and all the profit pools from all of this infrastructure don't have to be captured by, say, a DEC or an Oracle or an IBM, and this would lay the groundwork for a lot of things to come, including, frankly, just saving Amazon as a company. I mean, in two thousand, they almost went out of business because they were so tight on cash, and they were spending so much on infrastructure, that this sort of move to the open-source ecosystem and doing a massive rewrite of all of Amazon.com to run on Linux and run on these... They did this big deal with HP, run on HP servers.
- DRDavid Rosenthal
Right, rather than DEC.
- BGBen Gilbert
That frankly saved the company from a cost perspective during that really tight time.
- DRDavid Rosenthal
But that is not virtualized cloud servers. [laughing]
- BGBen Gilbert
[laughing]
- DRDavid Rosenthal
That's not what we're talking about with AWS. Here's the other reason why this excess capacity myth is a myth.
- BGBen Gilbert
... how is Amazon supposed to serve their AWS customers if all of them are on excess capacity during Q4 at all?
- DRDavid Rosenthal
[chuckles] Right.
- BGBen Gilbert
Like, let's say I'm Netflix, and I just took a dependency, and all of my streaming is happening on AWS. Is Amazon just gonna tell me I can't do it during Q4 when they need the servers? It's ridiculous.
- DRDavid Rosenthal
[laughing] No holiday movies. Can't watch Die Hard at Christmas.
- BGBen Gilbert
Ugh. So it is a very convenient narrative when someone's trying to solve the puzzle of: how did this internet retailer turn into a real technology company? Oh, they had all these extra servers. Dispelled.
- DRDavid Rosenthal
So the best and final word on this that we have to put here, 'cause it literally is from part of the horse's mouth itself, comes from Werner Vogels, the, at the time, AWS CTO, now CTO of all of Amazon, who wrote flat out in a Quora post in 2011, quote, "The excess capacity story is a myth. It was never a matter of selling excess capacity. Actually, within two months after launch, AWS would have already burned through the excess amazon.com capacity. Amazon Web Services was always considered a business by itself, with the expectation that it could even grow as big as the amazon.com retail operation." Hmm. Maybe, maybe.
- BGBen Gilbert
The other interesting thing to point out is it doesn't give Amazon enough credit about their intentionality-
- DRDavid Rosenthal
Yes
- BGBen Gilbert
-and strategy.
- DRDavid Rosenthal
It short sells Amazon.
- BGBen Gilbert
Yeah. They had this extra capacity, this cost center that they were using. Well, two things: one, technology was never a cost center for Amazon. They never looked at it like, "Oh, we have an IT department." They always thought about themselves as a technology company, so it was always thinking about, "Okay, in eighteen months, Moore's Law is gonna make it so we have twice as much compute. What crazy, cool stuff can we do with that?" They always looked at technology as an investment, not a cost center. And the other thing, to your point, that it sells them short on is, as if this wasn't an intentional strategy. This was an incredibly intentional strategy and a brand-new business school case study-type laser focus on an emerging market that they had reason to believe that they could create.
- 20:34 – 31:45
Origin story #2: Tim O’Reilly, Web 2.0, and AWS as APIs for the Amazon catalog
- DRDavid Rosenthal
Okay, that's origin story number one. Origin story number two, we're gonna get into this a lot more, and, you know, I didn't even really realize before diving into this, the depth of innovation of what AWS was and what Amazon was doing and led them to it, is so beyond anything else that was happening at the time. This is a true fundamental innovation. So let's get into it. Remember from the Kindle Coda vignette, how, uh, it was one of those crazy stories about who was responsible for the inspiration for the Kindle, and it turned out it was Tesla founder, Martin Eberhard?
- BGBen Gilbert
Crazy! He had invented the first e-reader that wasn't quite viable yet and tried to sell it to Amazon and tried to get investment from Amazon. And Amazon said: "No, we'll wait till the world shifts a little bit, different technology. It's actually something we can own outright, rather than funding and potentially having competitors use, too." And of course, that would be a few years later, and Amazon would create the Kindle internally.
- DRDavid Rosenthal
So there is a similar sort of figure involved in inspiring the vision for AWS, and that is Tim O'Reilly. And for anybody of a certain age, you certainly remember the O'Reilly programming books, the O'Reilly conferences, and in particular, for me, I mean, they were the organization, and Tim, as the leader of the organization, championed the whole idea of Web 2.0.
- BGBen Gilbert
For sure. I mean, I remember first reading, I think, the PHP book that they put out, and then when Web 2.0, this sort of idea of, you know, I can consume on the web, but also I can post on the web, and that sort of led to social media, and one of the key enabling technologies in all that is AJAX. And I remember reading the O'Reilly AJAX book of, wow, I can use, what is it? Asynchronous JavaScript and XML to make dynamic web pages without needing to refresh. That was truly magical at the time.
- DRDavid Rosenthal
And there were a few core tenets that they sort of defined as what Web 2.0 meant. Part of it was in opposition to Web 1.0, which they considered static, and so Web 2.0 was dynamic, like you're saying, but that wasn't all of it. Another huge part of what they meant by Web 2.0 was what they called participatory culture and interoperability. And they meant that both users on websites could interact with the website. So you had Flickr, you would upload your photos, and you would interact and change the website, or Google Maps, of course, was such a canonical Web 2.0 project. But even more than users interoperating and interacting with Web 2.0 sites was other developers. Remember mashups, Ben?
- BGBen Gilbert
Like mashing up APIs?
- DRDavid Rosenthal
Yes. So Web 2.0 mashups were such all the rage, and Google Maps was, like, a core part of this. People would take the Google Maps API and build all sorts of other websites using Google Maps' data and content underneath it, or, you know, Flickr had an API. It was APIs, it was interoperability, it was anybody can access. It's democratizing what we've built.
- BGBen Gilbert
Totally. It's so funny to hear all the crypto people today talk about composability. I feel like the old man yelling from tree or, "Get off my lawn," person, but it is very clear that people did not experience the 2006 to 2010 era of the exact same promise. But instead of smart contracts or composability on blockchains, people were saying, "It's a RESTful API. It has CRUD operations to create or read or update or delete things on a service. So if you're authenticated, then you don't need to necessarily use a web UI. You can just use the API, and you can upload a photo programmatically, or you can fetch your entire list of tweets programmatically." It was like all the web, instead of being in these siloed applications, was magically free for data to sort of move about in a utopian way without anybody's capitalist intentions getting in the way and siloing the data all to themselves.
- DRDavid Rosenthal
... So in the early days of all this, I think it was early 2002, Tim O'Reilly flies up to Seattle and meets with Jeff Bezos. And the reason he wants to come see Jeff, they've had a sort of checkered history in the past. You know, O'Reilly has not always been the biggest fan of Amazon. He's a book publisher, obviously, so he has some feelings.
- BGBen Gilbert
[chuckles]
- DRDavid Rosenthal
But he wants to make the pitch to Jeff that Amazon should embrace Web 2.0 and transform amazon.com into a participatory website.
- BGBen Gilbert
And this is a great idea. Being a Web 2.0 company means that you can do business with other companies without needing, like, a BD agreement in place. You don't actually need a partnership agreement. You basically can just publish your API. You can say, "Pay as you go, and here's how you pay, and here's how you get an account, and we can shut down your account if we need." But you can get API access to do business with us programmatically through this application programming interface, and it's great. Maybe no one or two companies will ever even need to talk to each other, which means you can do business with thousands of companies out there, not just a few that your BD people cherry-pick.
- DRDavid Rosenthal
Yes, and Jeff totally gets it. He gets this in so many ways. Amazon.com has this obvious business use case for APIs and allowing other developers and other websites to access data and content from amazon.com, which is they have a giant affiliate program [chuckles] that's called the Amazon Associates program.
- BGBen Gilbert
And they've got a catalog of every uniquely identifiable product in the world, certainly in the media space, but at this time, growing into many other categories, too. So wouldn't it be nice to access that authoritative catalog to fetch an image and display that image on my website if I'm trying to tell people, "Hey, go buy this CD"?
- DRDavid Rosenthal
Display the CD right there and then share the revenue with Amazon.
- BGBen Gilbert
It's good for both of us if I can do that.
- DRDavid Rosenthal
So after this meeting, Jeff does two things: one, he completely embraces this idea, and Tim and O'Reilly, he invites Tim up regularly to Seattle, has him speak at All Hands within the company, evangelize this idea of Web 2.0 and APIs within Amazon. Two, he starts a new team within Amazon to do just what Tim is suggesting. They build APIs that let any website developer plug into the amazon.com product catalog, do everything you just said, Ben, and the stated goal and mission of this team is to make amazon.com APIs available to developers and, quote, "Let them surprise us with what they build." That same year, you know, this is Amazon, they move fast, they hold a conference for developers in 2002. A total of eight people attend the conference.
- BGBen Gilbert
[laughing]
- DRDavid Rosenthal
They announce to the world the launch of this new division within Amazon that is called Amazon Web Services.
- BGBen Gilbert
So to your point here, this is not an important thing in the world yet. Amazon having a developer conference with eight people there, I mean, you look at re:Invent now, and I think it has 100,000 people who watch the keynote. Very different world.
- DRDavid Rosenthal
Yes, very different world and very different product. This is called Amazon Web Services, but it is not [chuckles] cloud-based IT infrastructure. It's other developers using the amazon.com product catalog, and indeed, it lives, Amazon Web Services lives within the Amazon Associates program, [chuckles] and that is run by a guy named Colin Bryar, which is very, very fun because Colin goes on to do many things, including recently co-authoring the book Working Backwards, which is a great book we used for a source in both this episode and the previous episode on amazon.com. But for now, in 2002, Colin technically becomes the first head of AWS.
- BGBen Gilbert
Wow! And it was just within Amazon Associates at this point, 'cause the whole point in this origin story, the scope of the ambition of AWS, was to make available assets of amazon.com to our affiliates, to Amazon Associates, who want to basically fetch images and items from the catalog and have that information passed along when someone purchases something to share some revenue. That was the scope of the ambition based on where it lived in the organization.
- DRDavid Rosenthal
Now, all of that is absolutely true. There is no element of myth or falsehood [chuckles] to anything in this second origin story here, and now we'll sort of transition from number two to number three together. But what I think is so important about number two, even though it leads to AWS, that is the creation of AWS, but not the AWS we know and love, but it's this idea of Web 2.0 and APIs that really starts to take hold, at least in Jeff Bezos's mind.
- BGBen Gilbert
And we have not once in this story said the phrase cloud computing or the cloud. We've said web services, and I think people today have heard AWS so many times that they sort of forget that it's a little bit of a misnomer. It's still called Amazon Web Services, but the vast majority of what is happening when customers are paying the ludicrous amount of revenue to Amazon to access AWS is not web services. It is not these RESTful API endpoints that you use to fetch and post information.
- DRDavid Rosenthal
Fun side barb: Do you know the origin of the term cloud as applied to IT infrastructure?
- BGBen Gilbert
Oh, I do not!
- DRDavid Rosenthal
This is so cool. It started-... at General Magic.
- BGBen Gilbert
Really?
- DRDavid Rosenthal
Yeah, how crazy is that? The Apple spin-out that invented the iPhone 20 years before the iPhone. As part of what they were doing, they also wanted to have, you know, the internet sort of barely existed, so I don't think they thought of it as the internet, but a distributed, always accessible back-end IT infrastructure for all the services that were gonna be on the mobile device. And so they started calling what they built for that a cloud infrastructure that the devices could access.
- BGBen Gilbert
General Magic was a pioneer in so many ways. It's amazing.
- 31:45 – 55:25
Origin story #3: The monolith crisis and Bezos’s ‘hardened interfaces’ mandate
- DRDavid Rosenthal
Such a pioneer. Okay, so back now to Amazon. They've launched Amazon Web Services, Web 2.0, you know, blah, blah, blah. Like, that's cool, but that's not what anybody is really focused on at Amazon. They're focused on, there are a lot of problems within the company, and arguably, the biggest problem is that the code base of amazon.com that Michelle Kappen designed back in 1995 has been, you know, amazing. He made so many great technical decisions that we talked about on the amazon.com episode. He designed it for how websites were built in 1995, [chuckles] which was small teams, not at scale, and monolith software code bases. Everything we talked about in the beginning of the episode. All of amazon.com, at this point, when it's now, you know, a multi- multi-billion dollar company, is running on one monolithic software code base. [chuckles]
- BGBen Gilbert
Yeah, I do know, after talking to some folks who were early Amazon engineers, around the summertime, they would start looking at what is the server that would be available on the market going into Q4 that is the baddest-ass thing we could possibly buy, and they would just buy the most expensive, suped-up server they possibly could from Deck or whoever else, and they would just try to make it through Christmas.
- DRDavid Rosenthal
Yes, Amazon would do code freezes going into the holidays. And think about this, this is just so foreign to everything we think about with technology companies now and how things run, and it's all thanks to AWS. You had to do a code freeze heading into the holidays because as you were adding new features and new elements and new teams... And remember, Amazon, at this point, they've got A9, they're working on search. Lab 126 is just kind of starting up, getting going. They've got all of these teams, huge numbers of engineers and product managers, that are building features, adding features, needing to access various parts of the site. Anytime you add one of those to the monolith software code base, it could break everything. [chuckles]
- BGBen Gilbert
Yeah.
- DRDavid Rosenthal
And so you had to do a code freeze, and it gets to a point where, remember, Amazon, as a company now, is trying to focus on profitability, efficiency. It gets to the point where the company just literally grinds to a halt. There's a lot of good stuff in the Working Backwards book about this, about how hard it became to get anything done and built at Amazon because of this rat's nest of complexities involved on the technical and infrastructure side.
- BGBen Gilbert
And as we're articulating problems here that were happening, one of them is, of course, you're gonna tip the server over if you add any additional complexity. The other of which is Amazon is doing the Amazon thing, and they're trying to enter new businesses and new categories. They're trying to grow, and they're trying to grow because the way that they've designed the business, as we mentioned in the last episode, the cashflow.com idea, where they're spending supplier money to grow before they're paying suppliers. Basically, they're investing the float in growth, so they do have to keep growing because they have bills coming due. And so they're continuing to look for new categories to expand into. They're looking around, they're seeing competition everywhere, so they're just trying to get big fast. So you have the issue of, well, we don't wanna commit more code and tip the server over, which, of course, means you can't launch these new businesses, you can't continue to grow, and you can't bring on more customers, 'cause more customers is more traffic, which is also gonna tip the server over.
- DRDavid Rosenthal
Let's just take one incredibly illustrative example, the marketplace business. When Amazon figured that out, that was transformative. That was high-margin revenue. That was how they competed with eBay. Well, technically, to do that, they had to re-architect how the buy button worked on the website. Now, imagine with a monolithic software code base, what was involved in that?
- BGBen Gilbert
You just get so slow in your actual software development, and therefore, slow to ship, and therefore, slow to innovate because you're afraid of, uh-oh, what did this other team commit to the code base here? What does that assume? Can I trust the contract that this function had is still true, or did someone update this function in a way that was tightly coupled to the requirement that they had for their thing? And before you know it, the code is making a bunch of assumptions all over the place, and if you go try to change anything, it's all so brittle that you basically need to talk to people, a bunch of people, before you're ever editing code because you might break something.
- DRDavid Rosenthal
Yes, and this is not just Amazon. This is every internet company. And the first companies to get to this kind of scale were like Amazon. It was this time. There were no internet companies of this scale before, and everybody is realizing you run into this brick wall, just from a complexity standpoint, when you reach a certain scale. This is a huge problem. Jeff is so focused on this, and not only Jeff, his new assistant at this time is focused on this, his new technical assistant, his shadow, who is, at this point in time, Andy Jassy, who was the first.... A lot of listeners maybe don't know about this, but anybody familiar with Amazon or who worked at Amazon knows Jeff's shadow. That's a legendary role to have.
- BGBen Gilbert
Which was a Microsoft thing before. Bill Gates' TA was sort of the blueprint for this.
- DRDavid Rosenthal
Right, technical assistant. Exactly. So the reason that Jassy becomes Jeff's first shadow is Jassy, he was a Harvard MBA, he had been a product line launcher. He'd launched music for Amazon. He ended up in the marketing department after that. And then two thousand, two thousand and one dot-com crash, Amazon axes the whole marketing department. [laughing] 'Cause, you know, we're not doing ads anymore, we gotta get profitable. And Jassy was gonna get laid off with the whole department, but Jeff liked him, and so Jeff said, "I'm gonna save Andy."
- BGBen Gilbert
Wow!
- DRDavid Rosenthal
"He's not gonna get laid off. I'm gonna find something for him to do. While we're figuring this out, let's take this technical assistant idea from Microsoft. He can come be my shadow," and he creates the role for him.
- BGBen Gilbert
And Andy's background is not technical up until this point. He becomes the technical assistant. He's brilliant, but he came in as one of the MBAs who was a category launcher when they were figuring out music, and electronics, and all these different verticals that they were going into. I can't remember which one Andy launched, but he was the launcher for-
- DRDavid Rosenthal
He did music.
- BGBen Gilbert
Music, for one of those. And I think fairly recently, like within the last five years before this, he had considered a career in the sports industry.
- DRDavid Rosenthal
Oh, yeah, he wanted to be a sportscaster.
- BGBen Gilbert
Yeah. He's, like, a well-known sports nut, has his basement tricked out as a sports bar, and almost took that career path. So we're not talking about a distinguished engineer at Amazon who's taking this technical advisor role because they're this technical luminary. It's a really smart guy who's just a very malleable, facile person.
- DRDavid Rosenthal
Yeah, it was just an excuse to keep Andy in the company and give him a job. But this is now the biggest problem in the company that Jeff is focused on and that Andy's focused on, and this is where all these threads come together. I'm just kind of in awe thinking about this. If I were looking at this problem of my technical, you know, infrastructure has ground to a halt, we can't ship anything, communication is so hard in the company, the natural thing to do, and I think what most companies would do and did try to do at this point in time, is, "Okay, we gotta improve our communication. We need better coordination loops, more communication, tighter communication, more coordination between teams. We need to build out our engineering management discipline here. We need to build out our processes. We're gonna get really efficient to be able to solve this complexity challenge."
- BGBen Gilbert
And at Microsoft, when they encountered this problem a decade or two earlier, they invented the program management role. That was basically-
- DRDavid Rosenthal
Ah.
- BGBen Gilbert
-the responsibility is... It was twofold. It was, there are not enough unicorn people out there who are 10X developers and also unbelievable sort of communicators, and so we'll just hire communication mouthpieces for the 10X developers. We can recruit these four sigma IQ engineer-type people, typically terrible communicators, and so let's just attach a PM to every dev, or a PM to every-
- DRDavid Rosenthal
Ah
- BGBen Gilbert
... two to five devs, and that way, they'll have communication associated with what they're doing, and all the PMs can talk to each other-
- DRDavid Rosenthal
That's so funny
- BGBen Gilbert
... and they can figure out what's happening between these two teams, and then they both go write specs, and the engineers write their engineering documents, and then, boom, we're off to the races. And the PMs can just keep talking it out to make sure that we're all on the same page.
- DRDavid Rosenthal
Now, I don't know this. You may, 'cause you were one of these people.
- BGBen Gilbert
That was my job, yeah. [laughing]
- DRDavid Rosenthal
Great. Was the Microsoft PM program, and it was program management, not product management, was that the origin of the modern Silicon Valley product manager?
- BGBen Gilbert
Well, it is specifically the origin of program management.
- 55:25 – 58:16
From software architecture to cloud infrastructure: making IT itself an API
- DRDavid Rosenthal
... There's one more, small compared to the big ideas, but kind of inevitable as things were going, but one more leap that we should talk about that happens here. Everything we just described so far in AWS origin story number three is related to software engineering and the amazon.com-
- BGBen Gilbert
Mm
- DRDavid Rosenthal
... code base. But what AWS is, is abstracted hardware, IT infrastructure, and software, too. But the core, like S3, EC2, that's IT infrastructure. How do you get from transforming your software architecture to, "Oh, now I need cloud IT infrastructure?" Well, it's kind of the same problem. It's an inevitable outcome. When you transition your software architecture to this service-oriented architecture and no longer a monolithic code base, you know, IT used to centrally plan, like we were talking about. We can ship these features at these times, and we need a code freeze at that time, and we need X capacity, and we can forecast that, and we can look out into the future. Now, with this, you've got all these distributed teams doing God knows what without talking to anybody. IT can't centrally plan anymore. So what Amazon realizes is they need to do the same thing with IT that they did with software engineering, which is transform it also into an API-accessible pool of computing resources, versus, "I'm giving you this server, and that's what you got."
- BGBen Gilbert
And you're talking about just internally. If all these teams are hitting each other's APIs internally, then, yeah, there has to be some dynamic way that if a, a whole bunch more load starts coming in and you weren't told about it, you do have to be able to spin up the hardware to handle that.
- DRDavid Rosenthal
And it's brilliant of, like, "Well, let's just make that an API, too." We place an API call into IT, they have a pool of computing resources.
- BGBen Gilbert
[chuckles] But much harder than it sounds.
- DRDavid Rosenthal
Yes.
- BGBen Gilbert
"Oh, yeah, IT can just become an API." [laughing]
- DRDavid Rosenthal
[laughing] No, no, no, no. This is a multi-year journey for IT at Amazon, too. Of course, it wasn't like Bezos just sent the email that Steve Yegge described, and everything happened overnight.
- BGBen Gilbert
Yeah. Okay, so what you're telling me then is Bezos gets excited about this, Jassy starts working with him on it. They're basically translating this idea of the first little nugget that you planted is, we should make sure that all of the APIs that we're making available internally, we should sort of, like, design them in mind as if they could be externally consumed at some point. But you haven't yet told me, how does some commercial offering eventually become available, and what is the commercial offering to third-party customers?
- 58:16 – 1:05:11
Andy Jassy’s 2003 vision: staffing up and formalizing AWS as a new business
- DRDavid Rosenthal
All right. So we're now in mid-two thousand and three. This has been this huge transformative project within Amazon over the last eighteen months. Jassy's been working a lot on it as Jeff's TA, and Jeff's like, "Okay, Andy, it's time for you to go back out into the company. You're done being my TA. You gotta go become a leader of something within the company." It's almost like an echo of Jeff and David Shaw back at D. E. Shaw.
- BGBen Gilbert
Yeah.
- DRDavid Rosenthal
Like, they start thinking together, like, "Okay, Andy, what are you gonna do? What's a new thing you're gonna go lead within Amazon?" [chuckles]
- BGBen Gilbert
Yep.
- DRDavid Rosenthal
Andy's probably happy he didn't leave Amazon, I think. [chuckles]
- BGBen Gilbert
[chuckles]
- DRDavid Rosenthal
And they come together to this idea of, well, maybe there's an opportunity to take the API-based IT infrastructure that we're developing here and offer it to third parties. So the legend goes, Andy puts together a six-pager.
- BGBen Gilbert
And this is the official Amazon legend. You can read about stuff on aboutamazon.com.
- DRDavid Rosenthal
Yep, it's in official Amazon documents. You know, everything in Amazon happens in written narratives, in six-pagers. He writes the six-pager. Famously, he has to tinker with the margins and adjust them to fit everything onto six pages. He can't fit financial projections on there, so there's no financial projections in this document, and then he whiteboards them out on the spot in the meeting with the S team and the board, where he's proposing this big, grand vision to take over Amazon Web Services, relaunch it with this new vision of being cloud IT infrastructure. In the document, there is an ask, a proposal to hire fifty-seven new people to go pursue this initiative. Andy talks about he's so nervous going into the meeting. This is such a huge career moment. He's asking for fifty-seven people. Nobody asks-
- BGBen Gilbert
[chuckles]
- DRDavid Rosenthal
... for fifty-seven people. It's a ballsy move. He's risking everything, and Jeff loves it, the board loves it, the S team loves it. It gets approved, and I think all of that actually happened or happened in some way, shape, or form. Andy, literally, we mentioned Colin Bryar, who was running AWS until this point, Andy and Colin swap places. So Andy goes in, takes over AWS. Colin becomes Jeff's next shadow.
- BGBen Gilbert
Oh, I didn't realize that. Andy right away took over the publishing of images via the Amazon Associates API, that sort of fledgling AWS?
- DRDavid Rosenthal
Either that happened, or this is part of the Amazon corporate history, hand-waving, you know, "Oh, all of it just went smoothly." Colin becomes Jeff's shadow. He then goes on to run IMDb when Amazon acquires IMDb, and then later, he would leave Amazon, team up with Bill Carr, who ran Prime Video, and then they write Working Backwards. I didn't realize this. This is brilliant. They now have a consulting firm together as part of Working Backwards to help companies implement the Amazon culture. [chuckles]
- BGBen Gilbert
It's great.... perfect.
- DRDavid Rosenthal
Freaking brilliant. So back to Andy. He gets approval, he's gonna hire fifty-seven people. He recruits Adam Selipsky to come in, join the company. Adam, of course, would later leave AWS to become CEO of Tableau, and is now back at AWS, where he is now CEO of AWS.
- BGBen Gilbert
So I watched every single re:Invent keynote to prepare for this, which I will tell you, that is a lot of IT conference keynote watching, and the most recent one is Adam Selipsky. And it's like ten, eleven years of Jassy up there on stage, and you finally get a different voice, and it's a little bit jarring, especially when you're mainlining them all back-to-back when it's suddenly not Andy Jassy. But yeah, Adam is the guy now.
- DRDavid Rosenthal
Yeah. Do you know who else was in that first wave of external recruits who come in to join AWS?
- BGBen Gilbert
I do not.
- DRDavid Rosenthal
Jeff Lawson.
- BGBen Gilbert
Oh, no way! The CEO of Twilio.
- DRDavid Rosenthal
Yes.
- BGBen Gilbert
Wow.
- DRDavid Rosenthal
Which totally makes sense that Twilio would come out of AWS.
- BGBen Gilbert
Yes, of course.
- DRDavid Rosenthal
So yeah, I think all of this really happens. Andy does write this doc. He does take over AWS. He absolutely builds and leads AWS from what it was, which was very different, into what it is today. But I think there's a little more to the story, too.
- BGBen Gilbert
It's a convenient narrative, and it's also a little bit odd that this sort of idea could come from someone who wasn't in the muck.
- DRDavid Rosenthal
Yeah. It's actually a really good interview that Andy does with the Harvard i-Lab in 2013, that's on YouTube. They, they're talking about the origin of AWS. I think the topic is intrapreneurship at companies, which, ugh, my God, the most disgusting word of all time. [laughing]
- BGBen Gilbert
[laughing]
- DRDavid Rosenthal
But Andy, in the talk, he's like: "Well, you know, we had to decide, you know, as part of this vision document and the discussion around it, how do we launch this? Do we pick just one service, one kind of IT primitive, and launch with that? Or do we put together a whole bunch of things and launch them all together?" And he says what they ended up doing was they got a tiger team together of the ten best technical minds inside the company, and they deconstructed all the major web services, web applications of the day, amazon.com itself, Google, eBay, all the... He doesn't mention them by name, but I assume the other big web services of the time, big web applications, and then figured out what you would need to re-architect those services based on this new cloud IT primitive infrastructure. And so they come up with a list. They decide you need storage, you need compute, you need databases, and you need a content distribution network, like what Akamai was, to be able to recreate any internet service of scale.
- BGBen Gilbert
I love that you say, "was," like what Akamai was. [chuckles]
- 1:05:11 – 1:08:49
What actually launched (and when): S3 first, then EC2, then CloudFront and RDS
- BGBen Gilbert
Nope, and in fact, I can recall personally using Amazon S3 for something I was working on, and there was no EC2.
- DRDavid Rosenthal
Yes. So unlike myth number one about AWS origins, you couldn't just take excess amazon.com IT capacity and externalize it. They had to go build all this from scratch as external services. It takes a couple years to do that for everything, and in fairness, you know, maybe in defense of the official narrative, they do start working on all of these suite of services all at once, and it just takes a while to get them all built. That probably is true, but yeah, S3 [chuckles] is the first service to launch by itself in March 2006.
- BGBen Gilbert
And let's talk about it. It was an independently useful thing. So S3, Simple Storage Service, it's a place that is available on the internet. You don't have to think about where it is. It's in the cloud, and you can dump images there if you're an application developer, and then elsewhere from your application or other applications, or no applications. If you just want to access the image directly by URL, you can access, and it's not just images, it's anything that you want to store up there, and it's this wonderful, magical, amazing thing where I don't have to buy a server, I don't have to configure a server, I don't have to rack a server, I don't have to think about maintaining a server, and I only pay as I go, and it is insanely cheap.
- DRDavid Rosenthal
Yeah, S3 launches in March 2006. EC2 launches a few months later in August 2006, I think in beta in August 2006.
- BGBen Gilbert
And what is EC2?
- DRDavid Rosenthal
EC2 is Elastic Compute Cloud, which is the compute counterpart to the storage part of AWS in S3.
- BGBen Gilbert
And a simple way to think about EC2 is, if you were a web application developer at the time, like I was, and you were writing stuff, and you were sort of running it on localhost on your computer, and you had previously been deploying it to some server at a data center that you could telnet to and ping and see it had an IP address, well, you could basically spin up an EC2 instance and treat it kind of like that, except it didn't have persistent storage associated with it. You could think about it like a computer without a hard drive that happens to live in the cloud and is yours until you stop using it.
- DRDavid Rosenthal
It's your processor in the cloud. CloudFront, which is the content delivery network, the CDN, the Akamai part of the puzzle piece, that launches in 2008.... and the first major database offering, RDS, the Relational Database Service, doesn't actually launch until two thousand and nine.
- BGBen Gilbert
And importantly, RDS, it wasn't like you just start using RDS, and now you don't have to use any of the stuff you've been using before. RDS would run the database that you were already using. So I can't remember if it actually launched with Postgres, but assume you're normally self-hosting Postgres on your server, or you have a separate database server that you're used to running that runs Postgres. Well, now you use RDS, and it runs Postgres, and all your queries work, and you can treat it like it's your own database server.
- DRDavid Rosenthal
So that's sort of the most obvious crack in the official narrative of the AWS origin, which brings us to the fourth origin story of AWS, the, uh, dissenting narrative, [chuckles] if you will.
- BGBen Gilbert
At this point, the compass in story number one was, like, a hundred and eighty degrees off, and then in story number two, it got ninety degrees off. We're fine-tuning now. Story three is basically right, but probably doesn't just include the full set of people that could have been written into the narrative.
- DRDavid Rosenthal
And I think story number four is basically right, too, but three and four are kind of the same.
- BGBen Gilbert
Success has many fathers.
- 1:08:49 – 1:17:58
Origin story #4: Pinkham/Black and the South Africa ‘skunkworks’ that built EC2
- DRDavid Rosenthal
Yes. Concurrently and separately to everything we just said in story number three, Andy Jassy working on Jeff's TA on this big problem, writing the vision doc, the business plan, all that, hiring fifty-seven people. In two thousand and three, a network engineer at Amazon named Benjamin Black is working on the IT architecture transition that we talked about, and he's working with Chris Pinkham, who is his boss, who, in fact, oversees all of network engineering within IT at Amazon, and Chris reports to Rick Dalzell, the CIO of Amazon. So the two of them, Benjamin and Chris, they write a six-pager about how they actually are going to restructure the network engineering part of the IT architecture to the new model that the company is moving to. And at the end of this document, supposedly, they mention that with the architecture that they have in mind, Amazon might actually be able to use that same architecture to sell virtual compute servers as a service to third-party developers. And indeed, Amazon could do that.
- BGBen Gilbert
Now, here's where things get murky because that document definitely does exist, this idea that most of it is focused on, "Here's how we're gonna execute our plan, and also, we could sell that infrastructure as a service." Here's where Ben Black and his blog post on the subject, and then in future interviews he gives with Network World and others, are very insistent that they then showed this to Jeff Bezos. The proposal made its way to Jeff Bezos.
- DRDavid Rosenthal
Yep, I think first to Rick, and then to Jeff.
- BGBen Gilbert
And he green-lit their project.
- DRDavid Rosenthal
Yes, separately from the rest of AWS.
- BGBen Gilbert
And what I can't tell is, did this, before it got in front of Jeff, get merged into Andy's proposal, and it was sort of green-lit as one big thing, or were there actually two different concurrent efforts?
- DRDavid Rosenthal
We're gonna tell the story, and then I have some thoughts on all this. So Chris is actually from South Africa, and right around this same time, he and his family want to move back to South Africa, leave Seattle, move back to Cape Town in South Africa. So he goes to Rick, his boss, and says, "Hey, I'm actually gonna leave and move back, move the family back." And Rick is like: "Oh, no, no, no, no. We're in the middle of this huge architecture transition. This is a key moment in the company. You are a super valuable person at the company. What if we do the same thing we're doing in Palo Alto with A9 and lab one twenty-six? We'll set up a new Amazon subsidiary in Cape Town, South Africa, that you can lead, and then we can retain your talents, and we can figure out what that new subsidiary will do." So Chris is like: "Oh, okay, that sounds good." Chris and Rick start thinking about this, and they decide, "Well, we just had this idea, Benjamin and I, in that paper we wrote about selling virtualized servers to third parties. What if we work on that at the new subsidiary?" [chuckles] So they do. Benjamin doesn't come along, but Chris and a really, really great engineer named Chris Brown-
- BGBen Gilbert
And from what I can tell, this is where Ben Black's involvement ends, where he was part of pitching the idea but is not actually a part of building the thing that they're gonna build in South Africa.
- DRDavid Rosenthal
Yep. So Chris and Chris go off to South Africa. They start working independently on this compute server idea, and they do. That becomes EC2. It's that team in South Africa that builds EC2. Brad Stone writes in The Everything Store, quote, "EC2 was born in isolation, with Pinkham talking to his colleagues in Seattle only sporadically, at least for the first year. Pinkham later said that the solitude was beneficial, as it offered a comfortable distance from Amazon's intrusive CEO." Quote from Pinkham: "I spent most of my time trying to hide from Bezos," Pinkham says. "He was a fun guy to talk to, but you did not want to be his pet project. He would love it to distraction." [chuckles] Hilarious.
- BGBen Gilbert
You can start to see, even in these very public, reasonably nice quotes, that there's enough tension between Chris Pinkham and the Bezos/Jassy leadership, that even in the official Amazon things that they put out about the development in South Africa, like, Chris Pinkham's name is sort of nowhere to be found, even in this South Africa-specific blog post about the history of EC2. There's clearly chafing between Chris and the leadership.
- DRDavid Rosenthal
... Yep. In Andy Jassy's infamous one-star review [laughing] of The Everything Store on amazon.com, in one of the several passages where he talks about how Brad had it all wrong, here's a quote from Andy: "The vision document proposing the AWS business and outlining the initial set of services for AWS, including our compute service, EC2, was finished and presented to the executive team in September two thousand and three. I wrote the document and was lucky to have the help of several people in putting it together. This was about a year before Chris Pinkham moved to South Africa [chuckles] to build the initial version of EC2. Chris played an integral role in the definition, team building, and product building of EC2, despite leaving before EC2 was launched." So clearly, there's some bad blood here, but, uh, my thoughts. I want your thoughts, too. I just find this whole thing ridiculous because, A, of course, it doesn't matter.
- BGBen Gilbert
Right. Who cares?
- DRDavid Rosenthal
But the most ridiculous thing is that what I think actually happened here, which is there were multiple teams working on multiple related things within the company, that's how Amazon prides itself on running: decentralized innovation. That was the whole point of this whole freaking exercise, was decentralized, let teams innovate. What's Jeff's, uh, invent and wander, you know, is the kinda mantra of him and the company. I think that's what actually happened. The official version now of the AWS history of it was all centrally planned, it was all in that two thousand and three document. That just seems sort of silly to me and counterproductive.
- BGBen Gilbert
Yeah, I agree. The other thing that becomes clear is it's really not about the idea, it's about the execution, and I know this is a trope, so to make it a little bit more specific, it can be about the idea if you define the idea as the hundreds of micro ideas that comprise the main idea. But if you're saying the idea is something articulatable in a sentence, well, that's pretty much worthless. And maybe even in a vision doc, it's about the thousands of micro decisions you make while executing it and actually doing the work to execute it, that sort of ends up mattering. But, you know, history is written by the victors, so we're seeing some of that play out here. The other thing that's very clear is Andy Jassy is just a brilliant strategist and fantastic leader, and so of course, someone like him in the organization would end up actually running it. So I don't even know why there's dispute over, "Well, it was my idea." It's like, well, who cares? Who's gonna end up turning this thing into a world-changing business?
- DRDavid Rosenthal
You know, you had that great playbook theme and takeaway from the amazon.com episode that I think you posted as a clip on Twitter and LinkedIn that went so viral, and people who were original Amazon employees loved it, which was your idea that Amazon was a pathfinding algorithm.
- BGBen Gilbert
Yeah, it was brute-forcing its way through a maze to eventually find the correct way by just gathering data. Launch stuff, gather data, tear it down, start again.
- DRDavid Rosenthal
Yep. Go through the maze, hit a dead end, backtrack a step or two, go take another path, and I think that actually is also how AWS launched.
- BGBen Gilbert
Early on, but I wanna get to that in my playbook 'cause I think it actually contrasts that in some ways.
- DRDavid Rosenthal
Ooh, okay, fun. Fun, fun, fun.
- BGBen Gilbert
All right, so it launches. Last thing to highlight here is the importance of the primitives. I don't know how intentional it was in the moment, but it became something that later on would become hugely important to them, which is that they truly were unopinionated about this as a platform. They said, "We're going to launch with primitives. It's the most basic story. That's the most basic compute. It's the most basic way to host your databases. It's the most basic CDN, and we can't wait to see what developers build in an innovative way with our absolute bare bones architecture that would go on to be called infrastructure as a service, as sort of a category." And they-- Again, I do not know if it was an intentional thing or not when they were first launching it, but they did not say, "Let's try and build a new OS, a new programming paradigm." No, no, no, no, no. We're just gonna give you super basic building blocks, and you run with it.
- 1:17:58 – 1:23:47
The business-model breakthrough: credit card provisioning, $3 bills, and startup formation
- DRDavid Rosenthal
So all that's on the technical side. We've been spending a lot of time there. We've alluded to this, but let's talk about what a radical innovation this was on the business and market side. I've got a great quote here. So when S3 launched, probably at the same time that you were playing around with it, a truly world-class, fantastic engineer at Microsoft at the time by the name of James Hamilton, who's now an S-team member and SVP distinguished engineer at Amazon, because of what he saw with AWS, he wrote on his personal blog about trying out S3 when it launched with a personal project. So here's a quote from him: "What was even more disruptive was a credit card was all that was needed to provision storage. There was no required proposal for financial approval, there was no RFP, no vendor selection process, no vendor negotiation, and no data center space needed to be found. I could just sign up and start working. From deciding to write the app to it being up and running on the internet was measured in days, and after debugging and testing extensively, the end of the month rolled around, and I got my Visa bill. Of course, I knew abstractly that S3 was disruptively priced, but when I saw that my bill for the entire development and test of this application was three dollars and eight cents, [chuckles] it just seemed wrong. Once development was complete, I was still storing all the test data in S3. So the following month, I got a bill-... for zero dollars and seven cents.
- BGBen Gilbert
So no joke, David, every month I get a bill from AWS for, like, seventy-one cents, and I have no idea what old project it was for. But it's one of these things where it's like, it's priced so dynamically. If it was a big, successful project holding a lot of data, then, you know, it would be expensive. They actually have pretty good margins on S3 and on bandwidth and some of these things. But because it was an abandoned project for which I do not know what the email address to log into AWS is from whatever team I was working on, I kind of just don't care.
- DRDavid Rosenthal
Could you imagine back in 2006, let alone even probably today, Oracle or Microsoft-
- BGBen Gilbert
[chuckles]
- DRDavid Rosenthal
-or IBM or HP or you name it...
- BGBen Gilbert
They all have six, seven, and eight-figure contracts. There's no way that they're gonna invest in, "Hey, let's let people pay with a credit card and service this tiny little market."
- DRDavid Rosenthal
"And we'll charge you three dollars and eight cents." This was unbelievably world-changing, truly world-changing. This is how Dropbox, Instagram, Airbnb, Uber, Zynga, all of these companies get started.
- BGBen Gilbert
I remember being at all these startup weekends and all these hackathons where the audience, the family members who came, the, like, venture capitalists who came to be the judges, it was blowing the audience's mind how fast people could stand something up in forty-eight hours because suddenly you didn't have to spend five million dollars and three months figuring out what data center you were gonna put something in. You actually could just have an idea and get it out there within two days. This birthed that movement.
- DRDavid Rosenthal
We all lived it. Rover.com, that, you know, we were all in various ways, part of adjacent to, next to our great friend and mentor, Greg Gottesman. He was a VC. He wasn't technical, you know. It got built in a weekend.
- BGBen Gilbert
Yeah, Phil Kimmel.
- DRDavid Rosenthal
Phil, our buddy, built it.
- BGBen Gilbert
Yep.
- DRDavid Rosenthal
Amazon, of course, embraces this. In fall of 2007, they start the AWS Startup Challenge, and they host it first in September 2007. They didn't win, but do you know who was part a contestant in that very first AWS Startup Challenge?
- BGBen Gilbert
Is it like TeachStreet? Like, is it gonna be some Amazon inside baseball?
- DRDavid Rosenthal
Oh, even better. Justin.tv.
- BGBen Gilbert
No way!
- DRDavid Rosenthal
Which, of course, would pivot into Twitch.
- BGBen Gilbert
Which Amazon would then buy.
- DRDavid Rosenthal
Of course, Amazon would then buy.
- BGBen Gilbert
Wow, for, like, the better part of a billion dollars, right?
- DRDavid Rosenthal
Yep. I don't have a good sense of how Twitch is doing now. I assume Amazon got a good deal on that almost no matter what, but we got to find the right way to revisit that.
- BGBen Gilbert
For sure. But that is a great use case. Like, Justin.tv early, I mean, they were using a lot of bandwidth to stream video. They were using a lot of S3 to store. Like, it was a great use case. And man, did Amazon embrace this sort of thing. This is probably one of the biggest keys to success or sort of playbook themes for why AWS became successful. They realized how perfect this was for startups. They realized how hard it would be for large enterprises to just wholesale move over. They realized that was not gonna be the first beachhead market. But for startups who were building something from scratch, who could go on to become fifty-plus billion dollar companies, my God, let's get them on AWS. And the blitz was so impressive. I mean, I remember the first time I met Dave Chappell, who was doing developer evangelism for AWS Early with Jeff Barr and so many other folks there. It was just a breath of fresh air where every happy hour you went to, there were AWS people who were giving you tons of free credits, who were helping introduce you to other people for your startup. They all thought about themselves as active participants in the startup community. So it just became this obvious default that you would build on AWS because it felt so ingrained with how you make startups, is you start an AWS account for the thing that you're gonna build. There's a famous Andy Jassy refrain that you hear at basically every re:Invent, where he talks about, first,
- 1:23:47 – 1:33:13
From startups to enterprises: lift-and-shift, compliance, and the road to ‘mission critical’
- BGBen Gilbert
there were the enterprise cloud doubters who said: "Oh, maybe this is good for startups, but it's no good for line of business applications. It's no good for mission-critical applications," and, "Oh, maybe it'll be good for my test environment or my dev environment, but I won't be able to run enterprise-grade stuff there."
- DRDavid Rosenthal
I think his line is, at first it was that nobody thought you could run a real application. It was only, like, what James was building, like a personal-
- BGBen Gilbert
Ah.
- DRDavid Rosenthal
-test project. And then it was like: "Oh, well, you know, startups can run in AWS, but, like, real enterprises wouldn't do that."
- BGBen Gilbert
[chuckles]
- DRDavid Rosenthal
And then it was like: "Well, as a real enterprise, we can run non-differentiated, you know, non-mission-critical stuff in AWS, but we're not gonna put our mission-critical stuff in AWS. That's gonna be on-premise." And then it was like: "Oh, my God, take my money!"
- BGBen Gilbert
Right. So I think there's this interesting, obvious first beachhead of customers that are startups. But when you think about the enterprise adoption and how eventually now, you know, your bank's application is on AWS, and everything is moving to the cloud, or a hundred and twenty billion dollars a year of revenue has already moved to the cloud of at least Microsoft, Amazon or Google. So there's these sort of three prepositions of the cloud. There's people building on the cloud, which to me, that's lift and shift. And that's really, like, a phrase that the cloud industry uses for, hey, you were running, you know, some local databases, you had some local storage. You basically had your data center, and you just want to lift that up and shift it over and drop it in Amazon's data center, and you're not gonna take advantage of any cool stuff. You're just gonna now run your stuff in Amazon. So the benefit that you get of that is you only pay for what you use, you don't have to pay the big upfront costs, and you don't have to maintain it yourself, but otherwise, exactly the same thing.
- DRDavid Rosenthal
Jassy, actually, at the first re:Invent in 2012, as part of his presentation, he has a great-... slide on this, where he talks about the six reasons AWS wins versus traditional infrastructure for enterprises, and it's exactly what you said. It's one, you're trading CapEx for OpEx, which is great. You know, you can take all that expense in every income statement every year, as opposed to capitalizing it. Two, you're getting lower OpEx than you could on your own, thanks to AWS's economies of scale.
- BGBen Gilbert
They're getting better deals on their servers, so they're passing those along to you.
- DRDavid Rosenthal
Yep. Three, you don't have to guess on infrastructure capacity ahead of time. AWS is elastic. As you need more, it scales up. As you need less, it scales down, and that's actually four, it can scale down. [chuckles] When projects don't work, you're not stuck with legacy leftover infrastructure from things that don't work. Five, engineers can focus on writing code, not installing infrastructure. Focus on what makes your beer taste better. And then six was, you're instantly global on AWS, versus when you run your own on-prem data centers, you're, like, wherever your data centers are.
- BGBen Gilbert
Which sounds nice. It's not quite true. It's not one global availability zone. Actually, interesting point, that was the original premise. They thought they were gonna abstract that away, and you were gonna imagine sort of a global S3 data center, and when you deployed it, it just went to all of the data centers. And then they quickly realized, "We're gonna have so much traffic from so many customers that we're gonna consume WAN. We're gonna consume the Internet's bandwidth replicating unnecessarily." And so there is ... You do not run globally by default in every single... Anyway, yes. And so then there's this step two, which is building in the cloud, and that's taking advantage of using things like the relational database service, that RDS, that very early thing that they launched, which is, "Hey, this isn't just your exact same code and your exact same infrastructure, but in our data center, and build differently." You're actually taking advantage of a cloud-native service. And then there's building for the cloud, and that's the future, and that's things like Lambda and DynamoDB. And if you think about Lambda, for folks who have not done this or heard of the serverless movement, it's this idea that you don't even need to, like, reserve an EC2 instance or deploy code to it. You just write your code, and then when you wanna call it, a thing just spins up for a few milliseconds, runs your code, and spins down, and you were never aware of its IP address or where in the world it was. You just know that your code executed, and so that's really like building for the cloud. You're completely architecting your application differently to take advantage of this very different world of computing the cloud offers.
- DRDavid Rosenthal
If we rewind to origin stories number two and three of the big monolith software problem, and, like, that all the engineers and product teams in Amazon and every other internet company were spending all their time focused on not making their beer taste better, undifferentiated heavy IT lifting. In the beginning, really, what happened is probably development teams in those days were spending, like, seventy percent of their time on infrastructure and setup and thirty percent of their time on software development. And then AWS shifted it to, "Okay, spend seventy percent of your time on software development and thirty percent of your time on worrying about our APIs and your infrastructure." This for the cloud, you know, Lambda, everything is like, that's taking it down to zero. [chuckles]
- BGBen Gilbert
Right. That's the goal, at least. I think all this stuff sounds better in principle than it actually ends up in practice. But yeah, that's the idea. Now, AWS, in its earliest days, let's call it the first couple of years, was really startup-focused, new applications from whole cloth that wanna use our infrastructure as a service, primitive building blocks. And they very quickly realized, "Well, if we're doing infrastructure as a service, it also does enable this lift and shift thing. So as long as we work like hell to satisfy the compliance requirements, and availability requirements, and uptime, and all this stuff, replication requirements of enterprises-
- DRDavid Rosenthal
You get SOC 2 audited with Fanta.
- BGBen Gilbert
There you go. Perfect. So very quickly, AWS could serve these two markets of startups and the lift and shift enterprise. Now, another way you could have designed this is, instead of doing this infrastructure as a service and these primitives, you could say, "Let's think about the far future, the Lambdas of the world," and we're imagining now in 2006. "Why don't we just build that sort of stuff to start? Let's change the development paradigm. Let's build the platform of the future. That platform will live in the cloud. That platform is not Windows of the past or the App Store of the current day," where it was just sort of coming. "That platform of the cloud, why don't we start writing the brand-new paradigm today?" And there are a couple other big tech companies that took that approach at first that were completely wrong, and the unfortunate thing for Microsoft and for Google, who really started at this platform as a service layer, was you basically didn't get the startups because you didn't have a mature platform yet that people were excited to build on and understood how to build for. But you also didn't get the enterprises 'cause there was no ability to lift and shift. And so if you were creating a platform as a service in the late 2000s, you're really a decade early, and you're building for a market that doesn't yet exist.
- DRDavid Rosenthal
Okay, so let's talk about what happens, 'cause this is just... Man, Amazon ran the table on maybe the most important market of all time.
- BGBen Gilbert
For, like, the first five years, with nobody competing with them.
- DRDavid Rosenthal
It's incredible. So 2006 is when the first services launch. 2007, 2008, that's when these startups are getting started: Airbnb, Uber, you know, Instagram, and the like. You know, and they're becoming big, but they're not yet at the scale that they are today. 2009, Netflix becomes a customer.
- BGBen Gilbert
And how crazy is this? They had already built their own-- In the last, like, three years,... basically cloud internally in order to stream video, which was originally, I think, streamed through Silverlight. They had this big partnership with Microsoft.
- DRDavid Rosenthal
That's right! Oh, my God, that was so terrible.
- BGBen Gilbert
[chuckles] Yes. Yeah, I think you had to use IE to view it. It was bad. But they had just invested a bunch and then did an about-face and said, "Oh, we were wrong. Actually, we're gonna use AWS instead."
- DRDavid Rosenthal
We're moving all of it to AWS, and I believe Netflix is still, to this day, I think, one hundred percent on AWS.
- BGBen Gilbert
I don't know about one hundred percent, but yes, they're still an enormous customer. Reed Hastings was actually the very first guest interviewed on stage at the first re:Invent in twenty twelve.
- DRDavid Rosenthal
I think in that interview, if it's wasn't that one, it's another one around that time, he talks about... People say, "Reed, you compete with Amazon.com. Aren't you worried about being on AWS?" And he's like, "No, [chuckles] I'm not worried at all about being on AWS. It is legitimately the smart infrastructure decision for us to make."
- BGBen Gilbert
Which that was such a feather in Amazon's cap. They've had two big feathers in their cap. There's that one and the CIA one. Like, "It's secure enough for the CIA to use, so it should be secure enough for you," and that was a few years later. But the Netflix one, I mean, a lot of people were afraid to use AWS early on because they felt like they didn't wanna do business with Amazon if they were a retailer, or they didn't wanna do business with Amazon if they were in video or in any of these things that Amazon was competing on. And Reed getting up on stage and saying this matter-of-factly and so forcefully was him saying, "You can trust that AWS is different than Amazon."
- 1:33:13 – 1:45:51
Why incumbents missed it: Oracle/IBM margin traps, Microsoft’s platform misstep, Google’s culture fit
- DRDavid Rosenthal
Okay, so why is Reed and Netflix making this decision? Why then do a bunch of other customers do this? And Microsoft... Let's put Google to the side for a minute, but IBM, you know, Oracle, all these legacy technology companies, why are they asleep at the wheel here? It's a disruptive pricing model.
- BGBen Gilbert
And let's not lump them together, 'cause I actually think it's worth analyzing. Each company failed to claim this opportunity for unique reasons.
- DRDavid Rosenthal
Hmm. Okay.
- BGBen Gilbert
So the first couple, it's worth analyzing. I think what you're pointing out is these old server companies, so the IBMs and Oracle on the database side, that made these ridiculous gross margins, and they sold you this complete proprietary solution.
- DRDavid Rosenthal
Yeah, eighty percent gross margins.
- BGBen Gilbert
Totally, and they would sell that to you, and they would install it in your data center, and eventually, they would hand wave and call something cloud.
- DRDavid Rosenthal
Private cloud. Private cloud!
- BGBen Gilbert
They might do it in their data center, they might do it in yours, but it's effectively the same thing, and it's sold on a license basis that comes with auditing. Amazon has this ability to literally meter your usage and then charge you exactly what you need to be charged, whereas this old model of buying a bunch of Oracle licenses and deploying them on the servers in your data center, you just get these audits every once in a while that were like: "Okay, cool. Well, we sold you a license, and you bought this many licenses. We'll show up and make sure that you aren't misusing this thing." So they weren't gonna change that business model. I mean, it was a license to print money.
- DRDavid Rosenthal
Amazon targeted gross margins and operating margins for AWS in the twenty to forty percent range.
- BGBen Gilbert
Which felt like a ten X and a twenty X for them, but was unattractive to the traditional-
- DRDavid Rosenthal
Right! This is the perspective. Amazon.com is operating on, like, a two percent operating margin basis. For Amazon, they're like: "Oh, shoot, we get ten to twenty X our margin basis with this new business? Awesome," but that's still less than half of the margin that the old school guys are getting.
- BGBen Gilbert
And the old school guys are certainly fat and happy on their operations, whereas Amazon knows how to run everything they've ever run as this unbelievable lean machine, 'cause they're so COGS sensitive on everything.
- DRDavid Rosenthal
So here's another thing, though. You mentioned, call it the Oracles or the IBMs, whoever, they'd come install this software on computers for you or in their data centers, call it private cloud, whatever. They'd install Oracle Database version nineteen, and then two years later, you're paying your maintenance costs. You're gonna pay an upgrade cost to go to Oracle Database version twenty, and then you're gonna go a couple years later to version twenty-one, and you're gonna pay a bunch of money every time you migrate.
- BGBen Gilbert
Right. Why would you give up this annuity that you have?
- DRDavid Rosenthal
Right. Well, cloud infrastructure, it's always up to date. There is no version. Whatever you're using, you're using the latest stuff because it's always... And then even more than that, Amazon gets to constantly iterate versus doing these Windows XP, every four years, we're gonna ship a big update. No, no, it's just constantly changing.
- BGBen Gilbert
Yep. Okay, so that's super old guard, the IBMs and Oracles, which is very funny when you watch all these keynotes. I wonder if anyone's ever watched them all mainlined like I did, 'cause I have this unique perspective, seeing them all so close to each other. They used to, on stage, refer to IBM and Oracle in a tongue-in-cheek way. They would refer to, like, a New York company, and it would be, like, IBM's logo, but it would say, like, New York company. And, like, Oracle is, you know... They would go as far as to say, like, San Francisco company, and then they might make a reference to, like, a super yacht or, like, sailing or something to, like, really drive the point home. Around twenty sixteen, seventeen, they totally did an about-face, and they just start directly attacking them, and they start directly attacking Microsoft, too, because I think Microsoft went from, in the early days, someone where Amazon looked at them more as a partner, like, "We're happy to run Microsoft stuff on your AWS instances," and now that Azure's actually been a extremely viable competitor and made a big, big comeback-
- DRDavid Rosenthal
They're the best competitor to AWS, by far.
- BGBen Gilbert
Amazon now loves attacking SQL Server licenses and stuff like that, that Microsoft, of course, comes in and audits, just like the old guard four. So let's look at Microsoft, though. Let's think back to the mid-2000s, 'cause this really should have been their business to take. They should have figured this out, but there were essentially two problems going on at Microsoft. One is that the Windows group just had too much power.... and between them and the Windows Server people and the SQL Server people, the goal of those groups was to get customers to do more with this idea that people thought was gonna be big for a while, of PCs taking over the data center and PC operating systems becoming the data center operating system. And really, the goal was sell more Windows Server licenses, and that was a great business. So anything that looked too much like that within Microsoft got gobbled up in an internal power struggle, 'cause it could look like it would cannibalize that thing.
- DRDavid Rosenthal
This was probably happening when you were there, right?
- BGBen Gilbert
Yes. Uh, it was sort of over by the time I was there. 2012 is when I arrived. But they did eventually realize that they had to make a big bet on Azure, and totally separate from Windows Server. And so this, we should give Ballmer credit, 'cause he did see this. So they replaced the leader of that organization at the time, of Windows Server and tools business, with Satya, who would eventually, of course, become CEO and then really double down on the cloud strategy. But they realized, okay, Azure needs to be a thing that's kept separate and has CEO sponsorship, and can sort of escape the Windows Server thing. But their second problem is what we were talking about earlier. They launched this thing called Azure Cloud Services, which they've now basically deprecated, which was a platform as a service approach. Microsoft had the golden goose. They had all the IT relationships. What they should have done is gone to everyone that was using Windows Server and say, "Great news. We have primitives in a data center that you can lift and shift to, much like how Azure works today. You can trust us. You already pay us. We'll make this a part of your enterprise agreement." But Microsoft got clever, and they thought, "You know what? The Win32 runtime, the .NET platform, we're a great platform company. Developers wanna build for the things that we make, so let's make the next generation set of APIs and platforms for building great cloud applications." And they just totally did not recognize the magical thing they had in front of them, which was all the customers and all the distribution, who, over the next five years, would slowly dribble out and start their new stuff on AWS while Microsoft was still figuring out its strategy. They got caught in that middle of people building brand-new apps, didn't know how to build for their platforms, and they didn't want the lock-in. That's still a big thing in cloud: "Oh, don't get locked in. You wanna be multi-cloud." And they didn't make it easy for their existing customers to lift and shift. So Microsoft, while they're in a great place now and have figured out an interesting strategy, and we can talk about kind of the bear and bull later, they just had five years of watching pitches go by.
- DRDavid Rosenthal
Yeah. Oh, it was such a whiff. Okay, we talked about Oracle, and I wanna come back to Oracle in a minute. We talked about Microsoft. What about Google?
- BGBen Gilbert
So Google's the third place. Amazon's got thirty-five, maybe forty percent. Microsoft's got twenty to twenty-two percent. Google, somewhere around ten percent.
- DRDavid Rosenthal
Which that Microsoft having twenty-two percent-
- BGBen Gilbert
Very impressive.
- DRDavid Rosenthal
That's an enormous win.
- BGBen Gilbert
Totally. Here's my sorta take on Google. They accidentally became a business. They launched as a project, and then they figured out this business that became unbelievably cash generative immediately. The nature of their business being search and feeding all the data directly in to make the results better, is that they instantly became a consumer-sponsored monopoly. Totally legally done. Competition's just a click away, but they're the best experience, so they just have these unbelievable reinforcing effects of becoming a monopoly. So they're a super high gross margin monopoly in the biggest market in the world, which is people wanting to use the internet, and they're the front door to the internet. So their entire existence, it's not that it's been easy, 'cause it's been a computer science challenge.
- DRDavid Rosenthal
It's been very academic.
- BGBen Gilbert
And it's... They've never had to go into a hard business. I don't know what Google's advertising margins are, but that business probably runs at, I guess it depends if you put sales above the line or below the line, but eighty percent gross margins. A thirty percent gross margin business is not particularly attractive to them, nor are they good at sales. I know they're getting better, but the narrative at the time was they made this G Suite thing, which at the time was called Google Apps, but no one would buy it, so they ended up giving basically all of it away for free to consumers forever, Google Docs and Gmail and everything. It was the best thing to use, and they couldn't figure out any way to sell it to enterprises. So they didn't have the competency of enterprise sales the way that Microsoft did. They didn't have the ability like Amazon to operate in these really hard businesses, eking out every last dollar, and so it just kinda looked unattractive. Meanwhile, they actually had the best technology for it. They actually operated these big data centers and this really novel way of networking all the computers together in order to pull off search, and they were sort of inventing machine learning before machine learning. So a huge value prop of the cloud now is all your data's in the cloud, and that way you can use a bunch of stuff that Google invented, TensorFlow, [chuckles] Kubernetes, to run your stuff in the cloud. This also was kind of theirs to win, but they didn't have the sales and marketing muscle, and they didn't, I don't think, have the iron gut that Amazon had to go do something kinda grinded out and hard.
- DRDavid Rosenthal
Well, I think they also made the mistake that you were originally talking about. I thought you were talking about Google, and then you said it was Microsoft, too, of, like, building too far in the future. I think Google made that mistake, too.
- BGBen Gilbert
Yep, that's totally true. I mean, the first foray was Google App Engine, which was in no way infrastructure as a service. It was not primitives. It was, I think you can write in Python or Java, and it was a specific API surface for GAE, and you can make App Engine apps, and it was all abstracted away from you. It was kind of the same Microsoft thing, if we're gonna get really clever and build you a platform of the future. But Google, per the Steve Yegge rant, is not at all a platform company, and so they didn't really know how to build it, they didn't know how to sell it, they didn't know how to identify a market for it, they didn't know how to support developers in it at the time, and so that sort of fell on its face. And what is GCP? Google Cloud Platform is now a very viable player in this race, but that's not where they started.
- 1:45:51 – 1:58:26
Databases, lock-in, and the utility thesis: backlog, stickiness, and market-size unconstrained
- DRDavid Rosenthal
Yep. Okay, let's talk about Oracle. One of the things that I think to most people, was to me before doing the research here, is vastly underappreciated about AWS is, you know, people think about EC2 and S3, and it's like infrastructure as a service, and compute, storage, and, you know, networking, and all that. True, Amazon doesn't report this, but if you google estimates of what the most popular AWS services are, the most used ones, EC2, S3, they're juggernauts, but numbers three, four, and five are all databases. AWS is a huge database business. They have taken so much share from Oracle, [chuckles] and while it's all related, it's infrastructure, it's also a different kinda business from infrastructure. Famously, AWS Redshift, like, why is it called Redshift?
- BGBen Gilbert
Ooh, yes. So for people who don't know this, there's an official Amazon talk track, and then there is a real talk track. So the official Amazon talk track, do you know this one, David?
- DRDavid Rosenthal
Oh, that it's like a Doppler effect or something like that?
- BGBen Gilbert
Yeah, it's physics-related. I think Amazon actually used Doppler as a code name for Alexa. And of course, one of their buildings in Seattle is that, you know, when the sound waves like get bunched up or spread out, like when a siren goes by, it's the Doppler effect, and redshift is the light equivalent. It's like a star moving away from you.
- DRDavid Rosenthal
But there's another part of the story here. [chuckles]
- BGBen Gilbert
Shift away from big red.
- DRDavid Rosenthal
Yeah, which is Oracle. [chuckles] So yeah, the database market is freaking huge. There's two properties of the database market that people just don't think about but are incredible. One, the global market size for database software is one hundred billion dollars, and it is growing at ten percent per year. Because everything you do with computing, you need to store it in a database. You need databases, and you can't get away from them. It's big, and it's growing fast. Two, database software may be the stickiest software of all time.
- BGBen Gilbert
Especially at the scale that people are producing data now. It's actually worth contextualizing this a little bit. So there's all these stats all the time, which are something like last year, more data was produced and stored than in the entire decade before and in the entire century before that, and that's not the exact stat, but there's eleven different variants of it, which we all sort of intuitively know because we're storing data on our phones. But when you have two things exponentially growing, it's hard to intuit the difference between those two things. And so we sort of know this about data. We also know this about the internet. Like, when you talk about dial-up back in the day, and then when, uh, people got their first cable or T-One line, and meanwhile, I'm here podcasting, and, David, I am seeing you in gigabit down directly into my computer, and it's unbelievable. So you think, wow, these two things have the same phenomena, except that they're actually moving at very different rates. The internet has not gotten faster at the rate that data storage has increased. So this is most illustrated in some of the AWS re:Invent talks. They're like, "Hey, a lot of you wanna shift to the cloud, but you have a petabyte of data, or some of you have an exabyte of data in your data center. So what do we do about that?" And they first released this thing that was a hundred-terabyte super secure thing they would ship to your office, called the Snowball, and you'd plug it in, it would automatically get all your data. It had a Kindle on it, so it would actually display a custom shipping thing, and you could track it all the way back, and it would arrive in the Amazon data center, and they would auto... It, it was like tamper-proof, bulletproof. It was an amazing thing. And they've released a few other generations of them now. There's even some with compute on them for field applications. And then the curves kept going. The internet kept getting a little bit faster, but our data storage kept getting a lot more significant. And there's some stat that Andy gives on stage in a keynote in twenty sixteen, seventeen, somewhere in there, where they announce Amazon Snowmobile, and he's like: Hey... 'Cause all of us are sitting here on computers that have a terabyte or two terabytes or four-terabyte hard drive, you're like, a hundred terabytes is not that meaningful.... And so then they're like, "We will send a snowmobile to your data center," which is a semi-truck full of snowballs, effectively, so that you can get the data to us. And even with this solution, this never underestimate the bandwidth of a semi-truck moving down the highway, this type of solution, it can still take six months to migrate all of your data into the cloud, whereas it would've taken you years and years and years and years, I don't know, the better part of a century, to actually upload it over the wide area network, over the internet. And so that, I think, illustrates pretty heavily your point about once you decide to put all of your enterprise data into a database hosted in some specific vendor's cloud, there's pretty meaningful lock-in there. There are very practical concerns with moving.
- DRDavid Rosenthal
Oh, I can do you one better on an example. Amazon.com used Oracle databases-
- BGBen Gilbert
Yep
- DRDavid Rosenthal
... when it was started. Amazon.com did not finish their [chuckles] migration off of Oracle databases and onto AWS products until 2019. [laughing]
- BGBen Gilbert
Oh, my God!
- DRDavid Rosenthal
Thirteen years after AWS launched.
- BGBen Gilbert
That is insane.
- DRDavid Rosenthal
It took that long for Amazon itself to migrate off of Oracle.
- BGBen Gilbert
Meanwhile, by that point, Amazon had eight different database solutions for other companies to use and had invented three of them. There's open source ones they host for you, but they also created DynamoDB, and they invented new database technologies that are compatible with other relational databases, but way faster, way more performant, and it's still hard to migrate within the company.
- DRDavid Rosenthal
Amazing. You know, you just play that forward and you're like, wow, okay, A, there's still so much revenue that's gonna shift to AWS, and B, it's gonna be so sticky. So sticky! One of the most amazing stats that, again, one of our friends pointed out to us that, I, I tweeted about this, and I posted it on LinkedIn, it's just crazy. AWS today is on an eighty billion dollar revenue run rate, eight zero. That is not the most crazy, impressive, defensible thing about AWS. If you go look in the financials, in the 10-Q, the latest 10-Q from Amazon, they have to report the AWS revenue backlog.
- BGBen Gilbert
Basically, revenue that's contracted but not recognized yet.
- DRDavid Rosenthal
These are contracts, you know, mostly with big enterprises, of revenue they've signed deals for, but that is not yet recognized. It's gonna be recognized in future quarters. [chuckles] That backlog of committed contractual [chuckles] signed revenue is over one hundred billion dollars. I don't even know what to say about that.
- BGBen Gilbert
There's a lot more storage and compute not on the cloud than currently in the cloud.
- DRDavid Rosenthal
So Amazon could shut down all sales efforts, stop growing, literally turn off the lights in terms of new business today, and they still have a hundred billion dollars more business that is contractually coming their way. [chuckles]
- BGBen Gilbert
It's insane.
- DRDavid Rosenthal
Crazy. Just crazy.
- BGBen Gilbert
So David, you mentioned they're on a, what is it, seventy, eighty billion dollar run rate right now?
- DRDavid Rosenthal
Eighty. Eight zero.
- BGBen Gilbert
Well, in 2014, Jeff Bezos wrote a memo, the annual memo that comes out, a letter to shareholders, and said that, quote, "I believe that AWS is market size unconstrained." [laughing]
- DRDavid Rosenthal
[laughing]
- BGBen Gilbert
That was the point at which it was a year before they broke out AWS's financials, and I think it was a six billion dollar run rate business.
- DRDavid Rosenthal
When the, quote-unquote, "AWS IPO," which I think Ben Thompson coined that term-
- BGBen Gilbert
Yeah
- 1:58:26 – 2:49:32
Strategy, risks, and misses: ML adjacency, Snowflake ‘whiff,’ service sprawl, and multi-cloud reality
- BGBen Gilbert
Yep. It's pretty interesting to look at all the ways they're pressing their advantages, too. Twenty fifteen, that year, they broke out finances. They also bought Annapurna Labs, this Israeli company, and they started custom-designing chips, which we've seen in both their training chips. They've done custom... I think they're called Trainium, and then they have inference chips, which are also some crazy name, like Inferon or Inf-- I can't remember exactly, but they have custom machine learning chips.
- DRDavid Rosenthal
Do you know who makes those chips?
- BGBen Gilbert
TSMC.
- DRDavid Rosenthal
Of course. Big TSMC customer.
- BGBen Gilbert
Huh. Yeah, the other thing is that in many ways, it's the embrace, extend strategy that Microsoft ran, where first they have RDS, and they're like: "You can run anything in RDS," and then they start doing things like launching Amazon Aurora, which is a direct attack at Oracle and a proprietary database software that they own and control. And they're like: "But it's so fast, and it's so performant. It's compatible. Oh, and by the way, we generate much better margins on it." It's all these things that they used to attack Oracle for, and they're like: "Well, look, now that we have all the customers, why don't we do some proprietary databases, too? And we can generate more margins on those." And there are ways that they generate huge margins, like bandwidth. AWS makes ninety-plus percent gross margins on their bandwidth charges. There are many ways where, yes, cloud is still objectively better than the old way that the licenses were structured, the old way of storing on-prem, the old way of hiring all your own IT people. But also, Amazon is starting to feel themselves on the lead that they've generated and run some of the same playbook. The other thing, so then the question becomes like: Okay, well, why machine learning? 'Cause it's so clear that compute is this massive pillar of the business. Databases has sort of been stood up as not quite as important, but definitely more important from a stickiness perspective. Every year, they announce some new database thing when they're on stage. Machine learning, they've announced SageMaker. They've broken out the keynotes, so now there's a custom ML keynote. They have a whole bunch of cloud-hosted ML offerings. They run TensorFlow, which is funny, 'cause that's a thing that Google created. They have their own container service. They also have their own elastic Kubernetes service, so they sort of have to serve customers, 'cause customers want Kubernetes, but they're trying to get you to use their own custom ECS Amazon container service. And what's becoming clearer to me is the machine l- learning capabilities that Amazon has need to be good, but they actually don't need to be as good as Google's, because here's sort of the strategy with machine learning: You're gonna use whatever ML is available with where your data is, 'cause running machine learning near your data is the most important thing. So once you've picked Amazon to be your storage vendor, and you've sent semi trucks full of your data into [chuckles] their data centers, you're not shopping around for, "Oh, where should I run my ML?" You're gonna run your ML on AWS, and so they can't fall crazy behind here. But I think this is one way that even though Google should be best positioned to have better ML offerings than anyone else, it kind of doesn't matter if they're not the place where customers are storing their data.
- DRDavid Rosenthal
Okay, well, one last element of the story before we transition to analysis. I don't think we can call this a coda, 'cause they failed. [chuckles] It's not a coda, 'cause they didn't do it.
- BGBen Gilbert
It's a work in progress.
- DRDavid Rosenthal
A work in progress, I think justifiably so. This has been an AWS love fest. We've heaped so much praise on them. It's like they've done everything right. It's amazing. There's one thing they missed. Ben, do you wanna tell us about it?
- BGBen Gilbert
Data warehouses. How is Snowflake its own fifty billion dollar company?
- DRDavid Rosenthal
Unbelievable.
- BGBen Gilbert
It stores data in AWS and other public clouds, and it is its own fifty billion dollar company.... And what Amazon would tell you is, "We have Redshift, and it's one of the fastest-growing Amazon services ever, and it's doing really well." But you know the databases team at Amazon, that whole org has to be very, very unhappy that Snowflake managed to, I mean, run the gauntlet on the data warehouse market.
- DRDavid Rosenthal
It's crazy that AWS did not do this [chuckles] .
- BGBen Gilbert
It's probably AWS's biggest failure, and the question is, why? And I think there's a few areas. One is just big company stuff. I think before launching something, when you're at Amazon's scale, and now that they are the trusted partner of all these IT departments, you've got these security things, operational things, SLA guarantees that they're fully committed to. And I think it hamstrings your ability to really streamline a product, be opinionated, and get something to market that's both fast and intuitive and built for the user. I think Redshift requires a lot of customization, whereas Snowflake is awesome for developers out of the box. And it's funny that the playbook that Snowflake ran is pretty similar to the playbook that AWS ran when they were just S3 and EC2 serving individual developers. So there's a little bit of like they're a victim of their own success on this front. The other one is, Ben Thompson pointed this out in a piece that we'll link to in the show notes, it's right there in the name. They're fighting Oracle. They're fighting the last battle with Redshift. It's, "Hey, take your Oracle-style data warehouse and basically do that in the cloud," rather than lots and lots of Snowflake customers never would have become Oracle customers. It was a different customer segment with a different set of needs. I mean, it's just a fantastic product, and that's not really who Amazon was serving. And there's new leadership there now, and they're getting the house in order, and I think they recognize this, but this was a whiff.
- DRDavid Rosenthal
Probably not a whiff on the order of-
- BGBen Gilbert
Microsoft and Google whiffing on cloud. Yeah. It's an order of magnitude or two smaller.
- DRDavid Rosenthal
Yes. So AWS, we're gonna do analysis now, do grading. There's no way this isn't gonna be a very high grade, but, like, if there's a black mark, this is it.
- BGBen Gilbert
The other thing where they're sort of a victim of their own success is the Amazon two pizza team thing led them to launch all these different services. Rather than having a cohesive product strategy, AWS has kind of been alphabet soup. And I haven't logged into the AWS dashboard in a while, but it used to just be so overwhelming, so many amorphous logos that all kind of feel like the same thing, where it's hard to disambiguate between two things. And I think Amazon realizes this because their keynotes now seem to be much more about pitching these vertical solutions, like, "Here's this thing for this industry, here's a vertical solution, here's case studies of other people in your industry," rather than first presenting you with, "We have four hundred and seventy-six services!" And I think that in the keynotes, they've also really dialed back on what used to be the drumbeat of the keynote, which is, "We launched what we consider to be seventy-four significant features this year, and we're excited to tell you all about them." I think that won for a long time, and now it's created so much confusion for customers that that's actually, like, the bull case for a Google, who is sort of a newer entrant, who's coming in with a more cohesive product strategy and can help customers really understand what they should be doing, rather than being like, "Hey, there's no guardrails. Good luck." And AWS keeps launching even more new services now to provide those guardrails and say, "Well, if you use whatever, whatever manager, then you can't get yourself into too much trouble." And it's like, "Oh, cool, a thirteenth standards body." [chuckles] They definitely have a little bit of that cleanup effort going on now, but, hey, they got market leadership, and they make far more revenue and far more operating income than anyone else, so it's hard to argue with. All right, listeners, well, this is a great time to share our next sponsor with you, something extremely unusual. This segment is brought to you by our great friends at NZS Capital.
- DRDavid Rosenthal
They're just the best.
- BGBen Gilbert
They are. We first chatted with Britton and Brad and Joe and John with the idea about doing a paid sponsorship. They jumped on it, but they gave us one very particular caveat because they're such learners. And they said that the only thing we could ask listeners to do as a call to action is to read one of their white papers, think about it, and offer feedback and further refine the thesis. Well, this time, in contrast to our last two episodes, we are going to talk about a different white paper, and for those of you who joined us for the talkback, that was awesome to get to talk to all of you with the NZS guys on the call.
- DRDavid Rosenthal
And that was so great about complexity and complexity investing, sort of the founding thesis of the firm. Today, though, we are talking about their white paper, Redefining the Margin of Safety, and we'll have a link in the show notes for folks to read it. But it's a super cool concept. So margin of safety, for folks who listened to our crazy ten-hour, three-part Berkshire Hathaway series last year, you'll remember that as an original Ben Graham concept, and then, of course, Warren and Charlie popularized it at Berkshire. And the idea is actually similar to NZS's core thesis, which is that you can't really predict the future [chuckles] as an investor or anybody else, and thus, the idea of having margin of safety in your investments, in the Graham concept of it, is around valuation. That gives you a wide leeway to be wrong.
- BGBen Gilbert
Yes. In a very wide possible of scenarios, rather than a very narrow set of scenarios, you're gonna be okay and still may make a profitable investment. And of course, Graham's a little simplistic viewpoint on this is, if it's a high valuation relative to the business's fundamentals, it's a low margin of safety, and a low valuation gives you a high margin of safety.... Now, NZS is different. They believe valuation alone is insufficient to actually think through what the true margin of safety is.
- DRDavid Rosenthal
So they think about margin of safety really as focusing in today's fast-changing world on the adaptability of the management team of a company. A management team that's highly adaptable is going to give you, as an investor in the company, a much higher degree of margin of safety if you believe that the pace of change out there is accelerating and things are gonna be changing. Valuation doesn't tell you much, but a management team's ability to adapt to changing conditions does. So that's sort of part one of their thesis here. Part two, which is really cool, is that there's also a time element here. So not just management's capability to adapt, but what is the business cycle of the company that they're running? If it's a super, super fast business cycle, well, you could have the best management team in the world at adaptability, they're just not gonna have time to react, and then the likelihood that within a narrow window of time they'll make the right choices is low. So an example here is Groupon or Zynga. You know, it wasn't maybe necessarily that the management teams didn't know how to adapt to changing conditions, it's just that the business cycle for those companies was so fast they didn't have time to.
- BGBen Gilbert
They actually use this analogy in the paper where they're saying, "Well, imagine that you're driving only two feet away from another car and you're going quite fast. If they tap the brakes, you're screwed no matter what, how good of a driver you are." But of course, if you're at a safe distance and you give yourself time to react, then there could be a lot more margin of safety, no matter what the price of the investment. So a little bit about NZS before we close here. They've got over a billion dollars under management. They're a global equity fund. They're long only. They manage money for institutional and accredited investors. They're just some of our very favorite thinkers. They started the firm in twenty nineteen and worked together for many, many years before that. If you find this stuff interesting, you should go read the whole white paper by clicking the link in the show notes. It is well worth a read, and we say that just as fans of their ideas. We are not clients of the firm. And finally, they do really, really, really want your feedback on this. So if you've got a bone to pick, if you want to understand it further and read the whole white paper, if you totally agree, click the link in the show notes, go check out the white paper. You can email any of them. Their emails are on the white paper itself. Or even better, we're gonna do another Zoom call with them in October to talk about these ideas together. So there is a second link in the show notes to add that to your calendar, and we would love to see you there. Our thanks to NZS Capital.
- DRDavid Rosenthal
Well, should we transition to analysis?
- BGBen Gilbert
Yeah, let's start with power, 'cause I think AWS is actually one of the best case studies in power of all time. So longtime listeners of the show know power is what enables a business to achieve persistent differential returns, or put another way, to become more profitable than their closest competitor, and do so sustainably, so they can sort of build enterprise value and be sustainably more profitable than their nearest competitor. David, when AWS broke out their financials, they were at, did you say nineteen percent, eighteen, nineteen percent operating margins?
- DRDavid Rosenthal
Nineteen percent.
- BGBen Gilbert
Now they're at thirty percent. They've gotten more profitable when the landscape got more competitive.
- DRDavid Rosenthal
Yeah, how did that happen?
- BGBen Gilbert
So there's something going on [chuckles] there. So there's a couple things. Moore's law is in their favor of all their cogs getting materially cheaper over time, and if I had to guess, I think they're not discounting for customers as fast as they're realizing both economies of scale and legitimately just costs coming down from Moore's law. But I think actually what's going on here is it comes all the way full circle that Amazon is offering platform-as-a-service offerings at this point and telling customers: "Hey, you could just keep using our primitive building blocks, but actually, what you should be doing to take advantage of the full power of the cloud to run these Lambda functions or to take advantage of these proprietary databases that are way faster, is to pay us a little bit more margin and take advantage of cloud-native things." And it's like, ah, the old tricks are the new tricks. Now you're the incumbent, and you're finding a way to do margin expansion, and the mindset a decade ago, or almost two decades ago at this point, of, "Oh, we need to create the platform of the future," was right, but you needed to do infrastructure as a service, as a stepping stone to get there. And it turns out that Amazon did that and generated their eighteen percent operating margins, and now here they are in a more of a platform-as-a-service world with customer lock-in, generating thirty percent margins.
- DRDavid Rosenthal
It's funny, I'm looking at the list of seven powers here, which are, for folks who are new, counterpositioning, scale economies, switching costs, network economies, process power, branding, and cornered resource. And I'm like, check, check, check-
Episode duration: 2:49:32
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Transcript of episode APvj15_YCqk
