The Twenty Minute VCMamoon Hamid: AI - Where Value Accrues, Startups vs Incumbents & Scaling Laws | E1217
CHAPTERS
- 0:00 – 3:00
AI supercycle and the new reality of incumbent spending
Mamoon frames the current moment as an AI-driven supercycle, comparable to the early internet era but larger in magnitude. Harry presses on what’s different this time: massive incumbent CapEx to build frontier models and what that means for venture investors.
- •AI as a once-in-a-generation technology supercycle
- •Comparison to the late-90s internet boom
- •Incumbents (Google/Microsoft/Amazon/Meta/Oracle) spending at frontier scale
- •Why VC opportunity still exists despite platform dominance
- 3:00 – 4:12
Where AI value accrues: vertical copilots that multiply scarce labor
Mamoon explains his core thesis for investing: focus on application-layer AI that supercharges highly paid, scarce workers. He highlights investing in products aimed at doctors, lawyers, and developers where time savings and quality gains are most valuable.
- •Framework: target highest-compensated/scarcest roles (doctors, lawyers, developers)
- •Applications that improve productivity and reduce labor constraints
- •Examples: Harvey, Ambience, Codium
- •Why specific pain-point software is attractive in AI
- 4:12 – 5:28
Differentiation in crowded AI apps: quality, tuning, and founder depth
Harry challenges the ‘10 startups per category’ reality. Mamoon argues differentiation comes from technical depth, model tuning, and output quality—especially in high-stakes domains like healthcare—paired with true domain expertise.
- •Competitive density is real in AI applications
- •Model output quality must be near-perfect in regulated/high-stakes workflows
- •Technical founders and deep ML expertise matter
- •Domain-expert cofounders create durable advantage
- 5:28 – 7:50
Is AI investing different from SaaS? Pricing mania and the “YOLO bucket”
Mamoon contends AI investing fundamentals are unchanged: invest in generational companies with great founders in the right markets at the right time. They discuss today’s extreme pre-product pricing and how firms selectively break rules without making it the norm.
- •AI doesn’t change core venture fundamentals
- •Entropy is high; change is rapid
- •Pre-product mega-rounds distort ownership math
- •Rule-breaking should be exceptional (a dedicated “YOLO bucket”)
- 7:50 – 9:15
Sugar-high revenue vs sustainable growth: from seat pricing to paid labor outcomes
They unpack why AI companies can scale revenue unusually fast: pricing shifts from seats to labor/value delivered. Mamoon explains how higher per-user pricing and outcome-based models can create rapid early revenue that may or may not be durable.
- •AI products increasingly sell labor/capability, not just software seats
- •Seat pricing moving from ~$30 to $300–$500/month in some categories
- •Fast early revenue can be structural, not merely hype
- •Need to evaluate durability beyond initial adoption spikes
- 9:15 – 13:12
Build vs buy in the AI era: Klarna’s internal tools, POCs, and history repeating
Harry asks whether companies will replace SaaS by building internal AI systems. Mamoon argues many will experiment via proof-of-concepts, but most will still choose specialized vendors—similar to the internet era’s build-it-yourself wave followed by standard platforms.
- •Klarna as a prominent ‘replace SaaS internally’ example
- •Internal systems can be costly to build and maintain
- •CIO-driven POCs are widespread; many won’t productionize internally
- •Analogy to 1990s internet consulting/build waves (Razorfish/Sapient)
- 13:12 – 17:19
Foundation models vs applications, middleware skepticism, and token economics debates
The conversation turns to whether foundation model providers will subsume apps and where investors may be overconcentrated. Mamoon is cautious on “middle layer” hype, defends vertical applications, and discusses commoditization, price drops, and the path to durable margins.
- •Concern: overinvestment in middleware between models and apps
- •Why apps likely persist: foundation model teams can’t build everything
- •Token pricing has fallen dramatically; commoditization pressures
- •Compute/infrastructure can be great businesses; LLM margins must evolve
- 17:19 – 20:51
Scaling laws and the $600B AI CapEx question: GDP, labor share, and demand creation
Harry raises Sequoia’s ‘$600B AI question’ about CapEx vs revenue. Mamoon argues the addressable pie is enormous because AI targets labor, not just software budgets, and even modest shifts in tech’s GDP share imply trillions in incremental spend.
- •Token costs down ~200x in early innings; more deflation likely
- •Debate: do scaling laws continue and how much model improvement remains
- •World GDP framing: ~$100T with ~50–60% labor component
- •AI monetization tied to labor substitution/augmentation and shortages
- 20:51 – 22:08
Kleiner Perkins positioning: boutique craft with scaled capital via follow-ons
Harry probes whether Kleiner is a boutique or a capital accumulator. Mamoon positions KP as a small, craft-driven team that believes venture doesn’t scale with headcount, while still running enough capital to double down on the best early-stage winners.
- •Early-stage fund and growth fund structure
- •Small team model; belief that VC craft doesn’t scale linearly
- •Growth fund used heavily to double down on existing winners
- •Examples of doubling down: Rippling, Glean, Figma
- 22:08 – 24:08
Reserves management and capital concentration: 60/40 rule and signaling realities
Mamoon shares practical guidance on reserves: invest more than half in the first check, then hold meaningful follow-on capacity. They discuss signaling, doing ‘something’ in follow-ons to avoid negative signals, and pay-to-play dynamics in difficult rounds.
- •Typical approach: ~60% initial check, ~40% reserved for follow-ons
- •Reserves as both art and science; reallocation as outcomes change
- •Board signaling makes ‘zero follow-on’ fraught; often do a small amount
- •Pay-to-play scenarios and why they change the decision calculus
- 24:08 – 27:21
When markets say no: Box bridges, founder dilution lessons, and why breakouts plateau
Mamoon recounts Box nearly running out of cash and requiring multiple bridge rounds during the financial crisis, emphasizing conviction in founders during market dislocations. They also cover founder dilution consequences and why successful companies later stall—often by failing to self-disrupt.
- •Box case study: multiple bridges in 2008–09 amid investor fear
- •Doubling down during dislocation can create exceptional entry points
- •Founder dilution can become extreme; importance of re-upping incentives
- •Breakouts plateau when innovation slows and incumbents protect turf
- 27:21 – 35:06
Liquidity realities: selling decisions, slow M&A, and reopening the IPO window
They explore how to think about selling—using ‘local maxima’ in perceived market value—and how constrained exits are today. Mamoon discusses why M&A has slowed (regulatory and corporate risk aversion) and why he expects a stronger IPO market after election uncertainty clears.
- •Selling at ‘local maxima’ when market perception peaks
- •Yammer→Microsoft example and reinvesting into the next wave (Slack)
- •M&A sluggish; big companies are gun-shy amid regulatory scrutiny
- •IPO market timing: waiting post-election; need big debuts to reopen
- 35:06 – 38:56
How Mamoon picked winners: Figma and Slack through engagement signals and market creation
Mamoon explains what he saw in Figma: a product finally enabled by browser tech with unusually high usage intensity even at small scale. He applies a similar lens to Slack—engagement and universality over early revenue multiples—while emphasizing love for market-creating products.
- •Figma diligence: usage intensity (workday-level engagement) as early proof
- •WebGL/browser capability as an enabling inflection
- •Slack investment: ignored revenue multiples; focused on daily multi-hour use
- •Preference for market creators (Slack, Figma, Glean) over crowded me-too plays
- 38:56 – 47:53
Founder ‘type’, fundraising discipline, and what VCs should stop doing
Mamoon outlines founder archetypes he prefers: product-obsessed first-timers creating new markets and ambitious repeat founders. He argues against ‘raise the most at the highest price’ thinking, criticizes VC echo chambers and data platform overconfidence, and explains why non-consensus decision-making matters.
- •Preferred founders: young/product-obsessed market creators; strong repeat founders
- •Disliked approach: top-down market mapping without true obsession
- •Fundraising: optimize for partners and long-term outcomes, not max valuation
- •VC pitfalls: echo chambers, false arbitrage, overreliance on “proprietary data”
- 47:53 – 1:03:41
Decision-making without votes, contentious deals, biggest loss, and board craft + quickfire
They cover Kleiner’s decision process: discussion-driven, no formal voting, with rare veto power. Mamoon names Figma as a contentious internal debate, discusses deployment pace and stage neuroplasticity, reflects on his biggest loss (Tally), and closes with board best practices and personal quickfire topics including faith.
- •No voting structures for early-stage; conviction-tested through debate
- •Figma as a contentious deal due to competition and pricing vs traction
- •Biggest loss: Tally (~$30M); consumer lending sensitivity to rates
- •Best boards: CEO-led structure, deep dives on 1–2 critical issues; quickfire incl. faith and geopolitics
