Anthropic Is Left Out of the Pentagon's AI Deals and in Funding Talks at Up to $950 Billion
Updated: 6 days ago
Title: Anthropic Is Left Out of the Pentagon's AI Deals and in Funding Talks at Up to $950 Billion
Date: 15 May 2026
Type: Blog
Author: SAASiQ (contact@saasiq.ai)
Word count: 1566 words
Reading time: 6 min
Published: 15-05-2026
On 1 May the Pentagon signed agreements with eight companies to run their AI on its classified networks, and Anthropic was not one of them. It has been designated a supply-chain risk since March, after it refused to let the military use Claude for mass domestic surveillance or fully autonomous weapons. Twelve days later, on 13 May, the New York Times reported that Anthropic was in talks to raise between $30 billion and $50 billion at a valuation of up to $950 billion, which would put it above OpenAI.
Eight companies on the classified networks
The eight are SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, Amazon Web Services and Oracle. The Pentagon said on 1 May that they would deploy their models on its Impact Level 6 and Impact Level 7 networks 'for lawful operational use'. IL6 covers classified data held in the cloud, and IL7 is the most tightly controlled tier, used for top secret material.
OpenAI says its contract carries similar limits. When it announced its own Pentagon agreement on 27 February, Sam Altman said its two most important safety principles were prohibitions on domestic mass surveillance and human responsibility for the use of force, and that the Pentagon had put them into the contract. OpenAI has not published the contract, and the Electronic Frontier Foundation questioned in March how far its wording would hold.
How Anthropic came to be designated
The dispute was over Anthropic's acceptable use policy, which already ruled out mass surveillance and autonomous weapons. The Pentagon wanted use of Claude for 'all lawful purposes' and set a deadline of 5:01pm on Friday 27 February. When Anthropic did not agree, President Trump directed every federal agency to stop using its technology, with a six-month transition for some, and Defense Secretary Pete Hegseth said he was designating the company a supply-chain risk. He also said no military contractor could do any commercial business with Anthropic.
Two formal designations followed on 3 March, one under 10 U.S.C. § 3252, which covers Defense Department procurement, and one under the Federal Acquisition Supply Chain Security Act (41 U.S.C. § 4713), which reaches across federal buying. According to the D.C. Circuit's later order, the reason given was Anthropic's refusal to contractually authorise use of Claude for mass domestic surveillance or lethal autonomous warfare. The label is normally applied to companies thought to be linked to foreign adversaries.
Where the court cases stand
Anthropic brought two cases. In the Northern District of California, Judge Rita Lin granted a preliminary injunction on 26 March that blocked the presidential directive, the Hegseth directive on contractors and the § 3252 designation. She found Anthropic likely to succeed on First Amendment retaliation, due process and administrative law grounds, and wrote that nothing in the statute supports the 'Orwellian notion' that an American company may be branded a potential adversary for disagreeing with the government.
The FASCSA designation is being challenged separately in the D.C. Circuit. On 8 April a panel of three judges refused to pause it while the case is heard, saying Anthropic's interests seemed primarily financial and that a stay would force the military to keep dealing with an unwanted AI supplier during an ongoing conflict. So, inside the Pentagon, the designation stands and Claude is being removed on a 180-day timetable, while other agencies and private customers are unaffected, according to an analysis by the law firm Jones Walker.
Talks with the rest of government
On 17 April Dario Amodei met White House chief of staff Susie Wiles, with Treasury Secretary Scott Bessent joining, and Axios reported that both sides wanted to keep the Pentagon dispute apart from the way other departments deal with the company, and that the next steps would concern Claude Mythos Preview.
Anthropic announced Mythos Preview on 7 April and has not released it to the public because it can find and exploit software flaws. It is being used under Project Glasswing, with twelve launch partners including Apple, Google, JPMorganChase and Microsoft, and more than 40 other organisations that maintain critical software. Anthropic committed $100 million in usage credits and says the model found a 27-year-old flaw in OpenBSD and a 16-year-old one in FFmpeg.
On 5 May the Commerce Department's Center for AI Standards and Innovation announced that Microsoft, Google and xAI would give it access to new models before release for national security testing. That extends agreements it signed with OpenAI and Anthropic in 2024, and the centre said it had already completed more than 40 evaluations. Reuters reported on 11 May that the page describing the new agreements had been taken down, and Commerce did not explain why.
The funding talks
On 29 April CNBC reported talks at a valuation of about $900 billion, and Bloomberg put the possible raise at up to $50 billion at $850 billion to $900 billion. On 12 May Bloomberg reported that Anthropic was seeking at least $30 billion at a pre-money valuation above $900 billion, possibly closing by the end of May, and that it was looking at a listing as early as October. The New York Times figure of up to $950 billion came a day later, with no term sheet signed. Anthropic has not confirmed any of these numbers.
For comparison, Anthropic's Series G in February raised $30 billion at $380 billion post-money, and OpenAI was valued at $852 billion post-money when it closed its round on 31 March.
The revenue figures come from Anthropic itself. On 20 April it said run-rate revenue had passed $30 billion, up from about $9 billion at the end of 2025. At its Code with Claude conference on 6 May, Amodei said the company had planned for growth of 10 times a year and instead saw revenue and usage grow 80 times over in the first quarter on an annualised basis, which had caused 'difficulties with compute'.
The Amazon and Google investments both come with computing capacity. Amazon said on 20 April it would invest $5 billion now and up to $20 billion more, with Anthropic committing more than $100 billion to AWS over ten years for up to 5 gigawatts of capacity. Google agreed on 24 April to invest $10 billion at a $350 billion valuation and up to $30 billion more against performance targets, according to Bloomberg, with 5 gigawatts of capacity over five years. On 6 May Anthropic took all of Colossus 1, SpaceX's data centre in Memphis, which gives it more than 300 megawatts and over 220,000 Nvidia GPUs.
Microsoft, OpenAI and model choice
On 27 April Microsoft and OpenAI rewrote their agreement. Microsoft's licence to OpenAI's models and products runs to 2032 but is no longer exclusive, and OpenAI can now sell through any cloud provider. The next day AWS put OpenAI models into preview on Bedrock and announced Bedrock Managed Agents, built on OpenAI's models.
Microsoft made Anthropic a subprocessor for Microsoft 365 Copilot on 7 January. That setting is on by default for most commercial tenants but off by default in the EU, EFTA and UK. Microsoft told EU, EFTA and UK administrators on 3 April (message MC1269241) that from 4 May a separate setting would make Anthropic models available by default for Copilot in Excel and PowerPoint, with Word to follow in the summer, and that processing for those models takes place outside the EU Data Boundary. The notice told administrators to review the setting in the Microsoft 365 admin centre.
Services firms and Google's devices
Anthropic and OpenAI are both now funding consultancies to put their models to work. On 4 May Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs set up an enterprise AI services firm valued at $1.5 billion, with $300 million each from Anthropic, Blackstone and Hellman & Friedman. It is aimed at mid-sized companies, with Anthropic's applied AI engineers working alongside its staff. On 11 May OpenAI launched the OpenAI Deployment Company with more than $4 billion from 19 firms led by TPG, and agreed to buy the consultancy Tomoro, which brings about 150 deployment engineers.
Google announced Gemini Intelligence at The Android Show on 12 May, describing it as turning Android into a system that acts on what the user intends. It also previewed Googlebook, a laptop category built for Gemini Intelligence, with the first machines from Acer, ASUS, Dell, HP and Lenovo due this autumn.
For Oracle and Microsoft customers
Oracle is one of the eight companies on the Pentagon's classified networks. Its AI Agent Studio for Fusion Applications has supported models from OpenAI, Anthropic, Cohere, Google, Meta and xAI since October 2025, so a Fusion customer may already be running agents on Claude. For most commercial customers none of this changes what they can buy: the designation applies to federal procurement, and Judge Lin's injunction has left agencies outside the Pentagon free to use Claude for now. Defence contractors are the exception: Anthropic cannot be a prime or subcontractor on covered Pentagon systems, though Jones Walker reads the designation as not reaching general use of Claude Code to write software.
SAASiQ's view is that a model provider's acceptable use policy belongs in the contract file next to the licence terms, because this dispute shows it can decide which customers the provider is able to serve.
The D.C. Circuit hears argument on the FASCSA designation on 19 May, and Bloomberg reports that Anthropic's funding round could close by the end of the month.
SAASiQ - Intelligent Solutions for SaaS ©
