And Now For Something Completely Similar: Untangling the Power from the Pretense of Our AI Circus

Lenin supposedly said there are decades where nothing happens, and weeks where decades happen. This wasn’t one of those weeks.

It might have looked like a historic week. The AI circus rolled into the UN with Sam Altman and Dario Amodei addressing the Security Council on the peril facing “humanity as a whole.” Britain and Australia weighed in with middling diplomatic or regulatory AI initiatives of their own. Trump renamed AI as superintelligence.

But none of this changed anything. The UN continued its kabuki theater of making global pronouncements without having the power to change even the most local of things. Britain and Australia remained not-great powers with delusions of grandeur. In the autumn of his patriarchy, Trump continued to flail with pathetic powerlessness. And Dario and Sam increasingly seem like bit actors in the unfolding AI drama.

So where’s all the power in AI? What’s this week’s real story? In Ezra Klein’s New York Times conversation with Nvidia CEO Jensen Huang, which That Was The Week publisher Keith Teare called “an interview for the ages,” Huang, the CEO of the world’s most valuable company, reveals himself as the guy running the AI show.

The $5 trillion dollar man talks with the quiet authority of the AI headmaster, fearlessly scolding his riotous sixth form. And his most riotous pupils are Dario Amodei and Sam Altman. So, in response to their latest handwringing about AI’s existential threat to humanity, Huang’s message was unambiguous. If you can’t align it, “don’t ship it,” he said of these supposedly humanity-destroying AI products, and if you truly cannot contain your own experiments, “then I think the answer is that we have to shut the labs down.”

The most revealing part of the Klein interview was about Nvidia’s astonishing economic clout in the AI economy. Huang put his chip manufacturer’s investment in this ecosystem at around $100 billion — larger, as Klein noted, than Joe Biden’s entire CHIPS and Science Act. This guy has serious clout, equivalent to or perhaps even surpassing the US state. He doesn’t waste his time at the UN. He’s not running wannabe great powers like Australia or Britain. He doesn’t play senile semantic games like you-know-who.

Huang can lay down the law to Sam and Dario (and Trump) because he doesn’t fear them. As the Exponential View’s Azeem Azhar reveals in a revealing interview this week with Atlantic CEO Nick Thompson, Nvidia isn’t dependent on the walled garden innovators at OpenAI and Anthropic. Neither lab actually buys Huang’s hardware — they rent capacity from the cloud companies that do, which makes them tenants rather than customers. Meanwhile the cheaper and maybe even better open weight models now pouring out of China run on the same Nvidia silicon. Whoever wins the AI race, Jensen “Levi Strauss” Huang sells them the chips.

So Mr Chips to Sam and Dario: heads you lose, tails I win. Shut up or shut down. That’s what happened this week.

Five Takeaways

• The Theatre of the UN. Altman and Amodei chose the Security Council rather than the General Assembly — the biggest forum available, and one with at least some teeth. What they want, in Keith’s reading, is political globalization: a super-state view of AI, with rules set from New York. But AI has already globalized, through the spread of open-source models, and not in a form they like. His analogy is Facebook’s Libra, launched on the condition that every government agree, and dead for exactly that reason. The second motive is liability. The conversation of the past fortnight has moved to who gets sued when something goes wrong, and the only rational answer is the companies that built it — unless they are following rules set by government, in which case liability moves. Both are heading for IPOs. As for Andy Burnham, Keith is unimpressed: Britain built its modern advantage on not regulating, and grasping regulation as the theme of a G20 chairmanship is unlikely to serve either the country or the man.

• Huang Calls the Bluff. The week’s main event was Ezra Klein’s long interview with Jensen Huang, which Keith calls an interview for the ages. Huang’s frame is the engineer’s: the Golden Gate was not the first bridge, and the Wright Flyer was not the first attempt at flight; engineering is failure leading to success, and what Altman and Amodei describe as species-like behaviour is to him an engineering problem. His conclusion was blunt, and in his own words: if you can’t align it, “don’t ship it”; and if the labs genuinely believe their experiments cannot be contained, “then I think the answer is that we have to shut the labs down,” because the liabilities — civil and criminal — are incredible. What he objects to is the request for relief: asking for regulation while asking for antitrust and product-liability protection makes no sense to him. Keith goes further: Huang is accusing them of being manipulative liars, and he agrees. Andrew pushed back on the idea that the CEO of a five-trillion-dollar company is speaking from pure engineering conviction — you cannot dismiss Sam and Dario’s interests and then treat Huang’s as irrelevant. Keith’s answer: he isn’t pursuing commercial goals, he is angry as an engineer.

• More Than the CHIPS Act. The line Andrew kept returning to. When Huang mentioned Nvidia’s hundred billion dollars of investment in the AI industry, Klein muttered that this was more than the entire Biden CHIPS Act — and in that aside sits the argument of the week. Governments, Keith says, are missing two things: vision, and money. They are deep in debt, and even if they had the cash there is little reason to trust what they would do with it. This is not an era of state-led innovation but of bottoms-up innovation, and between seventy-five and eighty-five per cent of it is happening in one dot on the world map. Andrew’s caveat is historical — there was a twentieth century in which governments built health services and semiconductors and space programmes, and some people are entitled to be nostalgic for it — but the era has passed, and the UN speeches are its echo.

• Not Dependent on OpenAI. The most useful technical thread, drawn from Nick Thompson’s interview with Azeem Azhar. Azhar’s point is that Nvidia is not especially dependent on OpenAI and Anthropic, which is part of why Huang can afford to attack them: both labs are renters rather than buyers, taking GPU capacity from the companies who actually purchase Nvidia’s hardware. Keith’s refinement is that this makes them indirect customers rather than irrelevant ones — every GPU purchase is good for Jensen, and the labs depend on the availability of what others buy. Measured in tokens, OpenAI is far ahead of everyone; measured in dollars, Anthropic does better, because its price per token is higher. On the bubble question Azhar is cautiously optimistic and gives numbers; Thompson is less so; and Andrew’s conclusion is the one that matters for anyone watching valuations: the AI economy could quite plausibly prosper without either OpenAI or Anthropic.

• Zuckerberg’s Return. The quiet story of the week, and the one both men think is the real news. Meta’s Muse is a full-stack agent — models at th...