The White House Is Now Hoarding AI Tokens: How Government Compute Competition Reshapes the AI Trade

The hosts of The AI Daily Brief dropped a line this week that should reframe how every investor thinks about the AI capex cycle. “Part of the reason that they were opposing Anthropic’s plan to expand access to Mythos was that they knew that there was a token shortage and they didn’t want other people using up the tokens that they might want to use.”

The U.S. government is behaving like a hedge fund cornering a commodity market. The commodity is inference compute. The supplier roster you already know. What changes when the federal government stops being a customer-of-last-resort and starts being a customer-of-first-call is the entire scarcity math investors have been using to price NVIDIA, the hyperscalers, and the AMD comeback story. The podcast’s other prediction matters too. Novel taxation structures like token taxes are about to become a policy conversation, and Senator Elizabeth Warren’s recent Time op-ed titled “Why We Need to Tax AI” is the opening bid.

What an AI token shortage actually means

Tokens are the unit of work a model performs. A query is tokens in, tokens out, billed by the million. When Sundar Pichai disclosed that Gemini was processing more than 16 billion tokens per minute via the direct API, he was telling you the meter is spinning faster than the foundry can pour silicon. That figure is up roughly 60% quarter-over-quarter via the direct API alone, a growth rate that outpaces any rational expectation of near-term accelerator supply.

Moreover, Polymarket already treated the government access question as settled. The Anthropic Claude Mythos federal access market resolved at 0.999 YES across the April, May, and June deadlines. The crowd believes the White House got what it wanted. Anthropic publicly limited Mythos Preview to about 50 vetted organizations, which is what rationing looks like when nobody wants to call it rationing.

NVIDIA and AMD on the supply side of scarcity

If Washington is the new whale, the picks-and-shovels names get a longer runway. NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) reported Q1 FY27 revenue of $81.61 billion, up 85.2% year over year, with Data Center at $75.25 billion (+92%) and total supply-related commitments of $119 billion. Jensen Huang called it “the largest infrastructure expansion in human history.” Data Center Networking grew 199% year-over-year

Besides, AMD (NASDAQ:AMD) is the more interesting derivative trade. A Meta partnership covers up to 6 GW of Instinct GPUs, with the first 1 GW on custom MI450, and the Q2 guide of roughly $11.2 billion implies 46% year-over-year growth. Every token Washington walls off is a token an enterprise has to source elsewhere, and AMD’s Instinct roadmap is the most credible second source at scale.

The hyperscaler squeeze

Microsoft put up $30.88 billion of capex in a single quarter, up 84% year over year, with commercial RPO of $627 billion. Amazon spent $44.20 billion on capex while signing OpenAI to roughly 2 GW of Trainium capacity starting in 2027 and Anthropic to up to 5 GW. Plus, Alphabet carries a Google Cloud backlog of roughly $460 billion.

In addition, Meta raised its FY26 capex guide to $125 to $145 billion. These numbers assumed enterprise demand would absorb the supply. If federal procurement skims allocations the way large language model providers complain about, hyperscaler unit economics get pinched while suppliers keep raising prices. Microsoft’s commercial RPO + Google Cloud’s backlog already reflect demand the clouds cannot physically serve in the current cycle.

The token tax and what to watch next

Warren’s op-ed is the first credible signal that Washington has discovered AI as a revenue object. A per-token excise would land hardest on providers whose business is selling tokens by the API, and lightest on vertically integrated players who own the silicon. Anthropic has said “some version of Mythos will be here in the coming weeks.” Thus, the investable read is straightforward. Scarcity benefits those who make the scarce thing and those who locked in supply early. The advantage accrues to “enterprises who figure out how to manage this more quickly and more efficiently than others.”