Sam Altman: 'We've Not Had the iPhone Moment' — OpenAI's CEO Admits AI Adoption Is Running Late

Over the weekend of August 22–23, 2026, Sam Altman sat down with podcast host David Senra for a long-form conversation about building OpenAI. By Monday and Tuesday, the quotes that traveled furthest were not about model benchmarks or funding rounds — they were an admission of error. The CEO of the company most responsible for setting the AI hype cycle said, in effect, that everyone — himself included — got the timeline wrong.

“We’re still at the palm pilot stage of adoption,” Altman said. “We have all of the technological pieces, but we have not had the iPhone moment of like completely changing how someone interfaces with technology.”

Coming from almost anyone else in the industry, that would be a routine caveat. Coming from Altman, whose product announcements have repeatedly framed each release as a civilization-scale shift, it reads as a deliberate reset of expectations — and it landed hard enough that CNBC, Yahoo Finance, and 24/7 Wall St. all led with it within 48 hours.

What Altman actually said

The core of his argument is a distinction between capability and adoption. He is not claiming the models stopped improving. He is claiming that the economy around the models changes far more slowly than the models themselves do.

“The economy just has so much inertia,” Altman told Senra. “People keep doing the same things they’re doing. They keep buying from the same company. I think that means we’ve all been too ambitious on timelines, even with this incredible technology.”

He was specific about his own past forecasts, too. He said that when GPT-4 launched in 2023, he expected software businesses to be “up for grabs” almost immediately — that established vendors would be disrupted on a short horizon because their moats would dissolve in front of capable general-purpose models. Three years on, that reshuffling has been real but far slower than he projected. People keep buying from familiar companies and want to use tools in familiar ways; incumbency turned out to be more durable than model capability.

Notably, Altman framed the slower transition as not entirely bad — a gradual adjustment, he suggested, could be smoother for society than an abrupt one. It is a noticeably more patient posture than the “next six months of AI progress should equal the last two years” framing he was using just weeks earlier.

The detail that matters most: Sora and compute

The most operationally revealing moment of the interview had nothing to do with metaphors. Altman disclosed that OpenAI shelved Sora — its flagship video generation product — “because it was taking up too much compute.”

Read that carefully. A company that has signed some of the largest compute contracts in history, and that just showed off its first custom inference chip at Hot Chips, still found a consumer video product too expensive to keep feeding. He called compute the biggest bottleneck at OpenAI and the thing he personally spends most of his time on.

That detail cuts against a lazy interpretation of Altman’s comments — the idea that “slower adoption” means “less pressure on infrastructure.” The opposite is closer to the truth: even if enterprise habits change slowly, frontier labs are consuming every available GPU and watt anyway, competing with each other for the moment when adoption does accelerate. Scarcity at the layer of chips, power, and advanced packaging is the binding constraint either way.

The counter-signal from the supply chain

The market’s reaction to Altman’s candor was telling precisely because the supply-side data points the other way. Nvidia closed Monday at $208.48 with a market capitalization of roughly $5.05 trillion. Its most recent quarter showed $81.6 billion in revenue, up 85% year over year, with data center revenue of $75.2 billion (up 92%) and data center networking up 199%. Management cited $119 billion in supply-related commitments and guided next-quarter revenue to $91 billion.

Jensen Huang’s summary of the same economy Altman was describing: “Demand has gone parabolic. The reason is simple. Agentic AI has arrived.”

TSMC tells the same story from one layer further down: Q2 2026 revenue of $40.2 billion at a 67.7% gross margin, a raised 2026 capital budget of $60–64 billion, chairman C.C. Wei describing demand as “very strong” through 2029–2030, and packaging capacity so tight it is limiting customer growth. When Altman says Sora was ditched for eating too much compute, the scarcity ultimately traces back to wafer allocation decisions in Hsinchu.

Both things can be true at once. Enterprises are slow to change how they work — and the labs are buying every GPU they can source to be ready when they finally do.

Two speeds, one market

The commercial numbers sharpen the contrast. OpenAI reported second-quarter revenue of $6.7 billion, up 18% quarter over quarter, and the Wall Street Journal reported that Anthropic’s sales roughly doubled over the same period. Nasdaq Private Market data has valued Anthropic at $1.12 trillion and OpenAI at $882.6 billion — secondary-market marks, not completed transactions — with OpenAI’s potential IPO now reportedly pushed to next year while Anthropic could move as soon as this autumn.

That is the paradox Altman’s interview crystallized: provider revenue is compounding at rates most industries would call hypergrowth, while the downstream economic disruption everyone predicted in 2023 — software companies “up for grabs,” workflows rebuilt overnight — is arriving on a multi-year lag. Demand for AI products is not the same thing as the replacement of incumbent software, and the gap between the two is where a large share of current AI valuations quietly lives.

Altman also reiterated that OpenAI wants to be a platform company rather than compete with its own customers, while flagging two risks he thinks about: losing control as AI systems grow more powerful, and too much power concentrating in any single company or model — a tension he left unresolved, since a dominant platform is by construction where concentration risk accumulates.

Why it matters

The honest reading of the interview is not bearish or bullish but structural: the Palm Pilot–to–iPhone framing concedes that the interface moment — the product that makes AI feel unavoidable to a billion mainstream users — has not happened yet. OpenAI is reportedly betting on a screenless, voice-first device with Jony Ive’s studio to be exactly that moment, with a debut teased for late 2026. Until something like it lands, Altman’s own account suggests the industry should expect continued hypergrowth in model usage and infrastructure spend, and slower-than-advertised transformation everywhere else.

For builders, the practical takeaway is to stop pricing startup theses around imminent incumbent collapse. For investors, it is that “too ambitious on timelines” applies to adoption curves, not to compute demand — the constraint runs through Nvidia, TSMC, and the power grid for years regardless. And for everyone else, it is a rare thing: the industry’s chief promoter, voluntarily lowering his own forecast, on the record, with his name attached.