The U.S. state of California announces partnership with Anthropic: State agencies can use Claude at half price.

California Governor Gavin Newsom announced a partnership with Anthropic, enabling all California state agencies and local governments to use Claude with a 50% discount, along with Anthropic training and technical support.
(Previously: Grabbing power in the name of “safety”? Unpacking Anthropic’s contradictory mindset: only if we dominate, the world is safe)
(Background: Is the countdown to the end of the AI high-pricing era nearing? Five structural reasons why tokens must fall in price)

Two paths in the same country that are utterly opposite. While the federal government affixed Anthropic with the label of “supply chain risk,” California, on the 29th, announced a partnership with Anthropic—allowing state agencies across the entire state to use Claude at half price. This is not only a procurement contract, but also a political statement.

The discount amounts to 50%. The agreement covers all California state agencies and local governments. Anthropic also provides training and technical support. The governor’s office said Claude will help state officials draft documents and analyze information. Newsom’s position is direct: “AI should not replace the work of government personnel; it should help our employees move faster, solve problems more effectively, and deliver better outcomes for Californians.”

The arithmetic behind the half-price deal

In recent years, both enterprises and government agencies have been facing the same math problem: the enterprise subscription costs for AI tools are high, and the gap between import speed and budgets is widening. California’s choice to use collective bargaining to secure discounts is, in essence, demonstrating a new model for government procurement.

This is also a continuation of the direction set by the AI executive order that Newsom signed in March. The core logic of that order is: accelerate government use of AI, but do so within a framework of “stronger safety standards.” At the time, Newsom said: “While the rest of Washington designs policies and signs contracts in the shadow of misuse, we focus on doing it the right way.”

The agreement also sets another record: it is the first AI productivity tool that is available statewide in California’s procurement system. Put simply, in the past each agency operated on its own and bought its own tools; now it becomes a state-level license shared across all units—which is unprecedented in the history of U.S. government AI procurement.

Who is betting on the future

From the perspective of capital and politics, although Anthropic lost the door to the Department of Defense at the federal level, it gained an entry point into California’s state-wide government.

The market scale and symbolic significance represented by the two are completely different, and so are their directions. The former is an unlimited space for military applications; the latter is a controllable arena for civilian government services. Anthropic’s choice is also directly reflected in its insistence on the conditions for technical deployment.

Even if the federal government’s procurement budget and scale remain far larger than any single state government for the foreseeable future, if Anthropic’s strategy works—making “AI procurement with safety standards” a template that more states and more democratic countries will follow—then the loss of being excluded by the Pentagon today may not be the final outcome.