The Real AI Race Is for the US Government, and Everyone but Anthropic Is Selling for Pennies

There is a quieter AI race happening, and the prize is the US government. Through one buying program, federal agencies can now get the top AI tools for almost nothing: ChatGPT for a dollar, Google's Gemini for 47 cents, Musk's Grok for 42, Perplexity for a quarter. It is a land grab dressed up as a discount. And the one major lab missing from the party is Anthropic, which is busy fighting the same government instead of selling to it.

The vehicle is GSA's OneGov strategy, launched last year to treat the whole government as a single customer and squeeze deep discounts out of tech vendors. OpenAI went first, offering ChatGPT Enterprise for a dollar per agency for a year. Everyone followed. xAI priced Grok 4 at 42 cents per agency for 18 months, Google put Gemini for Government at 47 cents through 2026, and Perplexity undercut them all at 25 cents. More than a dozen companies have signed on.

Nobody is making money at these prices, and that is the point. These deals are loss leaders, free trials meant to get agencies hooked on a specific tool before the real, paid contracts get written. Whoever becomes the default inside federal workflows wins years of revenue, data, and standard-setting power later. For a few hundred thousand dollars in forgone fees now, you buy a shot at being the government's AI backbone for a decade. That is a trade every lab will make.

Which makes Anthropic's position stand out. The company did sign a one-dollar GSA deal, but it expires in August, and its relationship with this administration has gone the other way entirely. Trump ordered federal agencies off Anthropic's tools back in February, the Pentagon tagged it a supply chain risk, and last week the government export-restricted its two best models. So while rivals race to embed themselves in agencies, Anthropic is in court and in negotiations just to stay in the building.

There is a real concern hiding under the bargain prices. Routing federal AI through a handful of vendors, at giveaway rates, concentrates a lot of power, and advocacy groups have already pushed back hard, especially on putting Musk's Grok inside agencies given his proximity to the administration. Cheap and fast is great for adoption. It is less great if the government ends up dependent on a few private models it cannot easily walk away from. The same lock-in that makes these deals smart for vendors is the part watchdogs worry about.

So the story everyone is watching, the Anthropic ban, is one half of a bigger picture. The other half is every other major lab quietly wiring itself into the federal government for pocket change. One company is being pushed out while the rest pay to get in. That contrast is where AI and the state are actually being decided right now.