(Preview) OpenAI Declares a ‘Code Red,’ Alan Dye Leaves Apple for Meta, Questions on Tranium 3, Substack, and F1
Summary
- OpenAI’s “code red” is a rational response to a consumer lead Thompson thinks should already have been unassailable. ChatGPT still has 800 million weekly users and remains the AI product Sharp sees ordinary consumers consistently using and liking. But Thompson lacks conviction because OpenAI secured users without building the revenue flywheel that would make its position “impossible to compete with.”
- OpenAI missed its safest window for introducing ads, leaving monetization and retention newly in conflict. Thompson says it could have launched “the world’s crappiest ads” in early 2023, improved them while competition was weak, and habituated users gradually. Now ads would immediately degrade the experience while Gemini could remain the cleaner, ad-free alternative.
- Advertising matters because AI retains meaningful marginal costs that subscriptions and rate limits handle poorly. Thompson argues that ads can make unrestricted access economical: more usage creates both serving expense and monetizable inventory, as on YouTube. Charging users runs into price elasticity, while effective ad markets pass higher prices to advertisers.
- The code red also exposes organizational sprawl across APIs, hardware, chips and commerce. Thompson calls shared chats potentially “super useful” and capable of creating a network effect, yet says the desktop implementation is broken and “half-assed.” His prescription is blunt: “The main thing needs to be the main thing.”
- Google is the uniquely dangerous rival because it can fund Gemini from Search, although it is not yet behaving like an unconstrained predator. Better Gemini tiers and Nano Banana Pro require payment, while Google raised Gemini 3 API prices—evidence that public-company margins and serving costs still constrain it. Even so, Google has repeatedly arrived late and then dominated, making it Thompson’s recurring “nemesis.”
- A Google victory could be bearish well beyond OpenAI because it might destroy AI industry economics. Sharp compares Search-funded Gemini to a subsidized competitor able to run at a loss and erase everyone else’s profit pool; Thompson agrees the concern is reasonable. Thompson’s honest conclusion is “I don’t know,” while Sharp points to the next six to 12 months as a potentially revealing period for whether his consumer-aggregation thesis was wrong or OpenAI simply moved too slowly.
Deep dive
1. OpenAI’s code red signals that its consumer lead is no longer secure
Sam Altman’s reported “code red” redirects resources toward ChatGPT and delays initiatives including advertising because “we are at a critical time.” Sharp finds the framing hilarious but the impulse healthy: ChatGPT is the linchpin for OpenAI’s broader ambitions.
Thompson had intended to tell skeptics to relax, then stayed up all night “bleeding this one out” because he lacked conviction. His concern: “By all rights, they ought to have secured the consumer space,” yet acquiring users and producing a good product did not, by themselves, create a defensible revenue flywheel.
Sharp’s normie evidence pushes against outright bearishness: ChatGPT is the one AI product everyone he knows uses and likes, and OpenAI still has 800 million weekly users. Thompson nevertheless leaves the outcome open—“I don’t know”—while Sharp points to the next six to 12 months as the period in which the thesis may become clearer.
2. The missed advertising window turned a moat into a product risk
Thompson’s aggregator logic goes beyond audience scale. Google Search and Facebook/Meta combine consumers, advertisers and reinvestment: an ad product that works attracts more advertisers, whose spending provides money to improve the product, creating a flywheel extraordinarily difficult to challenge.
Listener pushback sharpened the timing problem: Facebook and YouTube leaned into ads after users were entrenched, whereas ChatGPT users can still move to Gemini or Anthropic. Thompson “completely” agrees—the safe moment for even “the world’s crappiest ads” was early 2023, when OpenAI had room to improve the ads and train users to experience the product with them.
Pausing ads is therefore positive now, even though the delay embodies the strategic failure. Initial ads would diminish ChatGPT precisely when Gemini is “arguably better” and could offer “the nice ad-free experience.” Thompson still thinks OpenAI will probably produce a better model soon, but it must first secure the product’s position before monetizing it.
3. Ads could solve AI’s access and marginal-cost problem
Thompson argues that subscriptions alone are a poor way to make AI broadly accessible. Sharp took two or three years to subscribe to Thompson’s work despite being a close friend and still refuses YouTube Premium—an anecdote for how reluctant even privileged, engaged users are to pay.
AI usage still carries tangible marginal cost, making rate limits an existing form of user-experience degradation. Thompson’s counterfactual is straightforward: a ChatGPT with ads but without abrupt rate limits might be better, just as ubiquitous YouTube would be worse if videos buffered half the time.
YouTube demonstrates the alignment: every additional video costs Google to serve, but it also creates another ad impression. Productive AI ads could similarly turn longer sessions into more inventory, making wider access financially sustainable rather than something the provider must suppress.
Thompson concedes that advertising can produce sensationalism and other pathologies, but contrasts its economics with subscriptions: raising user prices reduces demand, while ad prices rise through supply and demand when advertisers get results. Advertisers, rather than users, bear those increases. That is why he calls the model “the most powerful business possible.”
4. Product sloppiness suggests OpenAI has lost sight of the main thing
Shared chats are Thompson’s specimen: “super useful” and potentially a genuine network effect because switching would mean getting friends to use something else, not merely trying another model. Yet chats disappear or fail on desktop, turning a promising lock-in mechanism into evidence that OpenAI “half-assed it.”
With thousands of employees spread across APIs, hardware, e-commerce, chips and other initiatives, Thompson asks, “What are people working on?” If resource constraints prevent the best ChatGPT experience, the code red is positive—but it also makes him regret softening his criticism of the API distraction.
He grants that moving from research lab to consumer company was painful: “People didn’t sign up to be Facebook, but you are presented with a Facebook opportunity. You need to take it.” OpenAI appeared to accept that mandate briefly, then resumed trying to become “the company of the universe.”
The Johnny Ive hardware effort compounds the strategic confusion. Thompson argues OpenAI should remain Apple’s horizontal-services partner with “the best experience everywhere”; signaling that it may compete with Apple merely pushes a valuable distribution partner toward Google.
5. Google may be the shark, but it still has economic constraints
Google is not yet going “for the kill.” Free Gemini exists, but stronger tiers and Nano Banana Pro require payment, and Google raised Gemini 3 API pricing, probably because it is more expensive to serve. Thompson reads this as rational concern for margins, not unlimited subsidization.
The historical threat remains ominous: Google was not first in search, smartphones, browsers or email, yet came to dominate each. Borrowing Canal+’s Formula 1 ads, Thompson casts OpenAI as the apparent leader and Google as Max arriving like the shark from Jaws: “Google is Max in these commercials.”
Sharp’s darker scenario is that Google Search acts like a subsidy for loss-leading Gemini, destroying rivals’ ability to earn profits and potentially driving them out. Thompson answers categorically that this is a reasonable concern; a tweet frames Gemini winning as “super bearish for the market.”
Thompson closes with the Uber/Lyft problem of interpretation: a sound thesis can fail when execution or shocks preserve a rival. He recalls that Uber seemed poised to take the market until its 2017 scandals and Lyft’s massive infusions of money changed the situation. If OpenAI loses, he may never know whether consumer aggregation was wrong or management squandered the opportunity—but being wrong forces a valuable reexamination of “assumptions that you had about the world.”