🔮 Look up, the curve turned #601
Exponential View Azeem Azhar
AI looks like it just hit an upswing, with more firms seeing real results and spending more. The scary bit: the gains, the secrecy, and the safety panic are all arriving at once.
Based on reporting by Exponential View, Azeem Azhar — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The week of 6 September is being framed as a turning point for AI, not because one neat thing happened, but because a lot of things moved in the same direction at once. At a meeting in Las Vegas, the source spoke to 250 IT executives and said nearly every hand went up when asked about serious AI results. A year earlier, only about a quarter had. Almost all of those firms also plan to spend more next year.
That fits a broader pattern in the numbers the source is watching: AI revenue rose faster in August than in July, and faster in July than in June. Bloomberg has also reported that Microsoft is planning to expand AI-serving capacity from about 2 GW today to nearly 13 GW by 2032, as part of a fleet that would grow from 12 GW to 38 GW. The reading is simple enough. Demand is outrunning what the industry can currently serve.
Anthropic’s economic scenarios are the other half of that story. The source says its own models land closer to Anthropic’s “substantial scenario,” in which AI adds about 8.3% to US GDP by 2030. That comes with growth shifting away from labor and toward capital, and with unemployment rising, mostly among knowledge workers. The source does not buy the most extreme version, where GDP jumps 32.4% and unemployment doubles. Firms move slowly. Economies move even more slowly. And politics pushes back.
Then there is the Navier–Stokes breakthrough, which is the kind of thing that sounds abstract until you see the scale. OpenAI used 10,000 agents and an unreleased model, sending 2,700,000 messages and 130 billion tokens through the task before finding a solution in 88 hours. The source says the proof runs to more than 500 pages and will not be intelligible to humans. That is a strange kind of progress: a landmark result that people can announce, but not inspect. It also sharpens the worry about secrecy. If the best models stay inside the lab until they’re ready to sell, science starts to look a lot like a locked room.
Safety hangs over all of this. The source says people building AI privately believe it could kill us all by the end of the decade, and that the big labs are now talking about slowing the frontier together. That may be prudent. It also means the frontier is no longer just about who can build fastest. It is about who gets to set the pace, and who gets locked out while the adults in the room compare notes.
My take — AI-written commentary, not fact-checked reporting
This is the part people keep missing: the industry’s biggest players are asking for trust at the exact moment they’re getting richer by reducing transparency. Nice work if you can get it. When the same companies that fear the technology also help decide how quickly everyone else can build it, that looks less like safety and more like a very polite gate.
Read more about this at: Exponential View