The Twenty Minute VCOpenAI Kills Sora & Hits $100M ARR on Ads | Oura Going Public & Whoop Raises at $10BN
At a glance
WHAT IT’S REALLY ABOUT
AI model leaks, OpenAI pivots, and markets debate revenue reality
- Anthropic’s accidental “Claude Mythos” leak is treated as both an operational embarrassment and a signal that more leaks will occur as agentic development accelerates and humans (or agents) ship faster with weaker security hygiene.
- The panel argues OpenAI’s decision to kill Sora reflects compute scarcity and a strategic retreat from expensive, low-monetization consumer video generation toward higher-ROI products like coding and ads.
- OpenAI’s early ad revenues (~$100M run-rate discussed) are framed as existential for a mass consumer business, with the view that meaningful success must be measured in tens of billions annually to justify scale valuations.
- A sharp selloff in cybersecurity stocks after the Anthropic news is characterized as largely knee-jerk, with the panel claiming AI agents should expand the threat surface and create a “golden age” of security demand rather than compress it.
- They debate how AI companies and “vibe coding” startups report ARR, highlighting gross vs. net reporting, token “reselling” and double/triple counting, and headline-driven tranche rounds that can mislead on true price and traction.
IDEAS WORTH REMEMBERING
5 ideasAgentic speed will amplify security mistakes, not eliminate them.
As development shifts to “ship constantly” and AI agents place code, configure infra, and publish assets, the same classes of errors (misconfigurations, exposed repos/DBs, staged assets accidentally published) occur at far higher velocity—making leaks more frequent even if per-action error rates fall.
Killing Sora is a compute-allocation decision as much as a product one.
Video generation is portrayed as extremely compute-intensive with low near-term revenue, so in a scarcity regime leadership reallocates compute toward higher-paying use cases (coding, enterprise, monetizable consumer surfaces), even if it publicly reverses a prior strategic narrative.
OpenAI’s consumer business likely needs ads at “Google/Facebook scale” to matter.
With consumer conversion rates discussed around ~5%, subscriptions alone may cap out; the panel’s view is that $100M is noise and OpenAI must prove it can build an ad engine that reaches tens of billions annually to justify consumer-side valuation expectations.
Market panic hit cyber stocks for the wrong reasons—agents should increase security spend.
They argue many sold-off names (e.g., identity/authentication or perimeter defenses) don’t directly overlap with LLM-based code scanning, and that broader agent adoption (root access, autonomous actions) expands the attack surface and drives more tooling, budgets, and M&A in security.
AI revenue optics are getting distorted by accounting choices and token “middlemen.”
Differences like gross vs. net reporting through cloud partners can inflate comparability, and when platforms resell underlying model tokens, the same economic activity can appear as ARR multiple times across the stack until profitability/Gross Margin discipline forces clearer accounting.
WORDS WORTH SAVING
5 quotesWe may be at the stage where we throw the humans under the bus, not the AI anymore, which I think at some level is pretty terrifying
— Rory O’Driscoll
You're seeing the economists, the accountants have wandered into the room and they said, "We have a scarce resource here. Let's optimize it. Let's devote this compute to the people who can pay the most for it"
— Rory O’Driscoll
I think it's saying that the ent- a big part of the whole strategic direction of the company was flawed.
— Jason Lemkin
In my, in my experience and opinion, this is one where it's just back ass backwards. Um, because if you're in the agentic world, this is the golden age of security.
— Jason Lemkin
You should conform your company around your customers and your model, not your VCs.
— Rory O’Driscoll
High quality AI-generated summary created from speaker-labeled transcript.
