California Embraces Claude at Half Price, Months After the Pentagon Blacklisted Anthropic

California Governor Gavin Newsom announced a partnership on June 29 that lets every state agency, city, and county buy Claude at half price through the new SITeS procurement portal, plus free workforce training and technical support. It’s the largest state-level AI deployment in U.S. history. The DMV is already using it for customer service, and the health services department runs internal Medicaid workflows through it. Here’s the part that makes this interesting: the same company was labeled a “supply chain risk” by the Pentagon back in February, barring every defense contractor from using it. One government layer blacklists a vendor, another embraces it at half price. That gap is the actual story here.

My read is that this isn’t about California being braver or the Pentagon being more cautious. Federal and state governments never shared a definition of “AI risk” to begin with. The defense establishment worries about autonomous weapons and surveillance authorization; city hall worries about DMV wait times and Medicaid processing speed. Those risk models barely overlap. If you work in procurement for another state and you’re seeing the opposite signal, agencies actively avoiding Anthropic because of the federal label, I’d genuinely like to hear what’s actually happening on the ground.

From Blacklist to Half-Price Deal in Four Months

Start in February. Trump ordered all federal agencies to stop using Anthropic’s technology on February 27, and Defense Secretary Hegseth designated Anthropic a “supply chain risk,” the first time the U.S. has ever applied that classification to a domestic company. The trigger was a standoff over the Pentagon wanting unrestricted access to Claude for any lawful military use, while Anthropic held the line on refusing to let its models power fully autonomous weapons or domestic mass surveillance. Negotiations collapsed. OpenAI picked up a $200 million ceiling contract that had been Anthropic’s within hours. Anthropic sued over the designation and lost its bid for a temporary block at the appeals court on April 8; the case is still ongoing.

Three months later, California went the opposite direction entirely. The governor’s office announcement calls it “first-of-its-kind”: every state agency, city, and county gets Claude at half price through the California Department of Technology’s new SITeS portal, with transparent pricing. Newsom’s statement reads almost defensively: “AI should not replace the human work of government; it should help our workers move faster, solve problems more effectively.” That phrasing sounds like it’s pre-empting the obvious question about outsourcing government jobs to AI.

TechCrunch’s reporting put a direct question to state CIO Chris Given: did the federal supply-chain-risk designation come up during negotiations? His answer: it “just didn’t come up.” That line says more than the contract itself. State procurement and federal defense procurement run on almost entirely separate tracks. A company can be a defense pariah and a preferred state vendor at the same time.

What the Numbers Actually Say

Start with scale. California’s executive branch employs roughly 220,000 to 250,000 people (the broader BLS state-government figure, which includes entities like the University of California system, runs closer to 560,000), and that’s before counting city and county staff the discount also covers. Claude Enterprise’s base seat fee runs $20 per user per month, with actual usage typically pushing that to $25-$200 depending on intensity. The 50% discount cuts the seat fee in half; usage-based costs aren’t discounted the same way.

Run a conservative floor: even if only 10% of state employees, about 22,000 people, use Claude daily at an effective $30/month once usage is factored in, that’s $7.92 million a year. That’s just the state layer, before cities, counties, and heavier power users are counted, so a realistic annualized revenue range lands somewhere between $20 million and $40 million. Compare that to the $200 million ceiling contract Anthropic lost to OpenAI in February: California’s deal looks smaller in raw dollars, but it’s recurring subscription revenue with room to replicate across other states, cities, and counties, while the federal contract is a one-time award that’s now tied up in litigation.

The sharper question is a first-principles one: if a company’s technology is too risky for the Pentagon to use for lawful military purposes, why does the same technology need no equivalent security review to process Medicaid applications or DMV customer service? Maybe both sides are genuinely assessing different risk profiles. Or maybe the federal blacklist was really about negotiating leverage over usage restrictions, not a trust score on the technology itself. Those two explanations point to very different futures, one where the federal-state split widens permanently, and one where this is just a temporary political friction that eventually smooths over.

Metrics Worth Watching

First, whether other states follow. If New York, Texas, or Illinois announce comparable state-level Claude or other LLM procurement deals within the next three months, this stops looking like a California one-off and starts looking like states building AI procurement logic that’s deliberately decoupled from federal policy.

Second, actual usage data from California’s deal. Whether DMV wait times measurably drop, and by how much, and whether the health services department can show a verifiable efficiency gain in Medicaid processing, are numbers worth checking for in the next six months, and they’ll say more than any press release.

Third, the progress of Anthropic’s lawsuit against the Pentagon. Any substantive movement in the federal appeal by year-end, regardless of outcome, will force a reassessment of whether “supply chain risk” was really a technical concern or a pressure tactic.

Fourth, whether this deal becomes material for a congressional hearing. A company blacklisted by the Department of Defense simultaneously serving the government of a state with nearly 40 million residents is the kind of contradiction lawmakers or reporters eventually pick up, and whether the federal government tries to force states to apply defense-grade vetting to AI vendors is worth tracking.

If this was useful, subscribe to the newsletter for weekly AI PM insights and GenAI case studies.

---

Related reading:

• Anthropic Restores Claude Fable 5 After 19-Day Export Ban: The Real Cost of One Classifier
• US Orders Anthropic to Pull Fable 5 and Mythos 5: A Narrow Jailbreak That Took Down Its Most Powerful Models
• Anthropic Files Confidential S-1: The $965 Billion AI IPO That Could Reset the Market