The Structure of the AI Contract Redistribution After the Pentagon Dropped Anthropic — Seven Contractors, May 1, 2026

On May 1, 2026, the US Department of War (formerly the Department of Defense) announced that it had simultaneously signed AI usage contracts on classified networks with seven companies: NVIDIA, Microsoft, AWS, Reflection AI, SpaceX, OpenAI, and Google. The absence from the list is obvious — Anthropic. About two months after the Pentagon designated Anthropic a "supply chain risk" in late February, the framework for replacement contractors was complete.

This piece is a follow-up to the Dario trilogy article, analyzing what the Pentagon got after dropping Anthropic. The short answer: the Pentagon adopted a contract architecture that replaces Anthropic's Red Lines with the five-word formulation "lawful operational use." And among the seven chosen, the top two (NVIDIA and Microsoft) are simultaneously Anthropic's largest investors — a twisted structure made plain.

The seven companies — investors, competitors, and a newcomer

The Pentagon's seven, sorted by character:

Key features of this seven-company composition: two Anthropic investors (NVIDIA and Microsoft) are included; the primary cloud partner (AWS) is included; direct competitors (OpenAI, Google) are included; and a newcomer startup, Reflection AI A frontier AI startup founded in 2024 by Misha Laskin and Ioannis Antonoglou, both formerly of Google DeepMind. Reported in 2025 to have reached an $80B Series A valuation. Drew sudden attention by being included in the Pentagon's May 2026 seven-company contracts — read as a defense-focused strategy. Directly challenges Anthropic / OpenAI lineage frontier model development. As the Pentagon sought an Anthropic alternative, a large contract opportunity went to an emerging startup — symbolizing a temporary lowering of the barrier to entry for defense AI. , slipped into a major contract.

An interesting absence (whom the Pentagon did not choose): xAI (Elon Musk) appears only via SpaceX, not directly. Mid-tier frontier players like Mistral AI, Cohere, and Inflection were all excluded. The Pentagon appears to have screened on two axes: "US-based" and "existing hyperscaler or frontier capable of direct competition."

"Lawful Operational Use" — replacing Anthropic's Red Lines

The most important change in the contract is a substitution of language. The two Red Lines Anthropic had written into its July 2025 Pentagon contract (multiple awards totaling about $20B over five years):

• "Not permitted for use in mass domestic surveillance of US citizens"
• "Not permitted for use in fully autonomous weapons systems"

Across the new contracts with the seven, the Pentagon replaced these two sentences with the Lawful Operational Use clause A general-purpose clause adopted in the Pentagon's seven-company AI contracts on May 1, 2026. Replaces the two explicit Red Lines written into the Anthropic contract (refusal of mass domestic surveillance, refusal of fully autonomous weapons) with the five-word English phrase 'lawful operational use.' Unless explicitly prohibited by law, this clause gives the Pentagon wide leeway to use AI across military purposes including secret combat operations. Anthropic (in its February 2026 CBS interview) had criticized that 'law has not kept up with the technology' — the Pentagon used the same 'rule-of-law' concept in the opposite direction. phrase. As reported by TheNextWeb, the clause prohibits "explicitly unlawful use" only, without explicitly forbidding use in mass surveillance or autonomous weapons.

A key point Dario raised in the CBS News Exclusive interview becomes visible here. Dario criticized that "the technology has outpaced the law" and that "mass surveillance is technically legal, but before AI it was practically impossible." The Pentagon takes the same "lawfulness" concept and runs it in the opposite direction, applying a clause across all seven contracts that "secures wide leeway to use AI across military purposes — including secret combat operations — as long as it isn't explicitly prohibited by law."

In other words, the Pentagon did not handle the dispute with Anthropic internally as "Anthropic is being excessive" or "the others are fine." Instead it took Anthropic's stated concern — that the technology has overtaken the law — and inverted it, writing into all seven contracts the position that "absent explicit prohibition, everything goes." A clear counter from the Pentagon side to Anthropic's Red Lines.

The twisted standing of NVIDIA / Microsoft / AWS

The most interesting aspect is the relationship of the top three replacement contractors (NVIDIA / Microsoft / AWS) with Anthropic. As laid out in the November 2025 strategic-partnership piece, NVIDIA invested $10B in Anthropic plus 1 GW of GPUs; Microsoft invested $5B plus selling $30B of Azure compute; AWS remains Anthropic's primary cloud partner and a training partner on Trainium and Inferentia.

For these three, Anthropic and the Pentagon are both massive customers / investees — neither side wants to choose. The structure that lets them avoid choosing was completed by the May 1 announcement:

• NVIDIA — $10B plus 1 GW (Grace Blackwell plus Vera Rubin) via Anthropic; classified GPU sales via the Pentagon. Takes both
• Microsoft — $5B plus $30B in Azure compute via Anthropic; Azure Government Cloud plus the OpenAI partnership via the Pentagon. Takes both
• AWS — primary cloud partner plus Trainium training partner for Anthropic; GovCloud plus classified workloads via the Pentagon. Takes both

What matters here: Anthropic's Red Lines are fundamentally a two-party issue between the Pentagon and Anthropic, not a direct constraint on NVIDIA / Microsoft / AWS. "Anthropic doesn't sell Claude for mass-surveillance use, but Microsoft can sell GPT-5 to the Pentagon via OpenAI." "Anthropic doesn't restrict GPU usage for autonomous weapons, but NVIDIA can sell GPUs directly to the Pentagon." The Red Lines function with the contractual counterpart (Anthropic) and not upstream (NVIDIA / Microsoft / AWS).

From Anthropic's side, the result is that "our Red Lines hold" and "the investor relationships hold," but "our influence over the Pentagon's AI usage at the level of the whole loses force." In the CBS interview, Dario said, "we're a free market — if the Pentagon doesn't like it, they can use other providers," and that's what happened. The limit became visible: Anthropic's Red Lines don't carry weight against the Pentagon's AI policy as a whole.

Anthropic's valuation — $900B and $30B in revenue, even after being dropped

In parallel with the Pentagon announcement, Anthropic's own business numbers came out. According to TheNextWeb, Anthropic's valuation reached approximately $900B valuation The latest reported estimate of Anthropic's valuation as of May 1, 2026. More than doubled in roughly six months from $350B at the time of the November 2025 Microsoft / NVIDIA partnership. Continued to grow through the Pentagon clash, reaching a top-tier private-company valuation alongside OpenAI and SpaceX. This indicates that the loss of the Pentagon deal did not affect the business overall. Revenue run rate reached $30B, with rapid growth continuing. Detailed investor / equity structure has not been disclosed. ($900B, roughly 2.5x in six months from $350B in November), with a revenue run rate of about $30B. Losing the Pentagon engagement had no effect on the business overall, and the numbers themselves became a bullish signal.

The May reality confirmed Dario's "we'll survive — we're fine" assertion in the CBS News Exclusive. Losing one Pentagon engagement (about $200M over five years) is less than 1% of a $30B company-wide revenue run rate. With Claude riding the B2B distribution networks of Microsoft and NVIDIA, Anthropic's business has continued on the same trajectory with or without the Pentagon work.

Another important structural point: Microsoft and NVIDIA, who invested in Anthropic, obtained replacement revenue via direct Pentagon contracts, so pressure on Anthropic to concede to the Pentagon has dissipated. They can expect returns of $10B and $5B from Anthropic, while collecting separate revenue from the Pentagon. An equilibrium has formed in which everyone wins financially while Anthropic's Red Lines stay in place.

Editorial reading — the moment the encirclement of Anthropic is complete

Across the three months from February to May, the Pentagon completed a full encirclement against Anthropic's Red Lines. Three features of the structure:

(1) The "both-sides" structure routed through shared investors — NVIDIA and Microsoft retained their Anthropic stakes while acquiring direct Pentagon contracts. Apparently adversarial relationships are made to coexist, and every player wins financially. Anthropic keeps its Red Lines, the Pentagon gets the alternative without Red Lines, and NVIDIA and Microsoft collect from both.

(2) An explicit negation of the Red Lines via "Lawful Operational Use" — the contract design that replaces Anthropic's specific prohibitions with a general clause meaning "if it's lawful, it's all OK." The message frontally rejects the dispute Anthropic raised, and all seven companies accepted the same clause.

(3) Mobilizing the emerging startup Reflection AI — a temporary lowering of the entry barrier to frontier model development, granting a major contract opportunity to a newcomer that wants Pentagon work, to fill Anthropic's absence. A clear signal to Anthropic that "even if you say no, there are companies that will say yes."

Anthropic's response to the encirclement, as Dario laid out in the CBS interview, holds the line: "we'll survive, we'll challenge in court." The May figures — $900B valuation, $30B revenue run rate — back that up. Still, the influence Anthropic's Red Lines once carried over the Pentagon's AI policy as a whole was effectively reduced to zero by the May 1 seven-company announcement.

This outcome casts a different light on the dichotomy Dario drew at the WSJ Davos interview — "scientist-led vs entrepreneur-led AI companies." Anthropic, as scientist-led, holds its Red Lines. But the Pentagon can secure the AI capabilities it needs through the other six entrepreneur-led / general-hyperscaler companies. Scientist-led can defend its own ethics, but it cannot set the direction of the industry as a whole — a limit becomes visible.

For MEMEX, this piece sits as a May 2026 snapshot of the completed encirclement, raw material for tracking subsequent outcomes — Anthropic's challenges in court, Reflection AI's record, the outcomes of the Pentagon's AI operations. Recorded as a moment when the geopolitical structure of the US AI industry shifted.

Related Resources

• From Davos Optimism to Pentagon Reality — Three Consecutive Dario Amodei Interviews, January to February 2026 — the prehistory of the conflict
• The Structure of Anthropic's $350B Valuation — $10B from NVIDIA, $5B from Microsoft — the double-contract structure with the investors
• The US-China Summit, May 2026, Beijing — AI Chips and "3B" — parallel geopolitical policy in motion
• Project Glasswing Announcement — Anthropic — the prehistory of the Red Lines
• Dario Amodei profile