🏛️ Nvidia ($1,000M), the check that is credit
Nvidia is not handing over $1 billion in cash. It is signing credits and capacity over five years.
OpenAI just put its September run rate at nearly $50 billion, about $20 billion below the figure it had let circulate.
Anthropic is putting 11 partners — CrowdStrike, Palo Alto, Dragos — on power and water networks with Claude.
Today's three stories draw the same trick: sell AI as a public good while the real bill stays private.
🏛️ Nvidia ($1B), the check that is a credit
On October 8 the White House announced $2.4 billion in tools and compute credits for the Genesis Mission, the federal science program that already spans more than a dozen agencies. Nvidia puts in $1 billion over five years, the largest slice. AMD adds $500 million. OpenAI, $200 million. Anthropic and Google, $150 million each. AWS, Crusoe, Micron and others fill the rest with checks of $50 to $100 million.
Jensen Huang sold it as “super intelligence” in the hands of American scientists. Read the small print. This is not a transfer to a lab. It is capacity, cloud credits and support for universities, quantum and government missions. Useful, yes. Philanthropy, no. Nvidia has been inside the national labs for decades. This is a public relationship with a five-year invoice.
📉 OpenAI ($50B), the number that shrank
On October 8 the Financial Times and Reuters reported that OpenAI told investors its September annualized revenue was almost $50 billion. The figure that had been circulating was close to $70 billion. The gap is about $20 billion. A source told Reuters the difference comes mostly from trying to compare apples with Anthropic: OpenAI does not include sales through cloud partners such as AWS or Google Cloud in that count. It started the year at about a $20 billion run rate. In the second quarter it reported $6.7 billion, below Anthropic's $11.5 billion.
On October 9 Bloomberg added the other side: in the current round OpenAI is telling investors it expects to reach or pass $70 billion by year-end, driven by enterprise. So $50 billion now, and $70 billion as a promise. With an IPO on the horizon, that accounting stretch is not a footnote. It is the product.
🛡️ Anthropic (11 partners), Claude on the power grid
The same October 8, Anthropic launched the Cyber Mission. The first piece is the Critical Infrastructure Defense Program: frontier models, on-site engineers and threat research for the firms that already protect power grids, water, factories and transport. The 11 founding partners are Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation. The second piece is a free open-source scanner. Reports leave the model and reach the maintainer with no human reading them first. Anthropic caps its liability at $1,000.
The industrial argument is real: those controllers cannot be switched off to patch, and a patch can take years. The commercial move is real too. Anthropic is placing itself as the lab that “defends critical infrastructure” right when the political story wants a good guy and a bad guy. And the same day it put $150 million in credits into the Genesis Mission package. Trust and a contract, in the same envelope.
🌐 The thread that connects them
Nvidia funds science with credits that can only be spent inside its world. OpenAI adjusts the run rate so the comparison with Anthropic does not blow up in front of investors. Anthropic lends engineers to the people watching the power grid and, on the way, signs the same public check. All three turn AI into national infrastructure. None of them let go of the model, the chip or the number.