The Deal
Anthropic announced a major compute agreement on April 6–7, 2026. Broadcom will supply Anthropic with approximately 3.5 gigawatts of Google TPU capacity starting in 2027, building on the 1 gigawatt already being supplied in 2026. The agreement also includes a supply assurance commitment under which Broadcom will provide networking and components for Google’s next-generation AI racks through 2031. Tom’s Hardware, April 7, 2026. The Register, April 7, 2026.
The numbers: Anthropic’s annualized revenue run-rate has surpassed $30 billion — up from approximately $9 billion at the end of 2025. That is a tripling in roughly four months. The company now has more than 1,000 business customers each spending over $1 million annually, a figure that doubled in less than two months. The Next Web, April 7, 2026. Mizuho analysts estimate Broadcom will record $21 billion in AI revenue from Anthropic in 2026, rising to $42 billion in 2027. Seeking Alpha, April 7, 2026.
The Timeline Gap
The Anthropic NDCA lawsuit against the Pentagon was filed in March 2026. It concerns whether the organism can be deployed in a military targeting context without Anthropic’s consent. The litigation is currently under submission — no ruling yet. The DOJ brief to the Ninth Circuit is due April 30. At the current pace, the legal process will likely run through 2026 and possibly into 2027.
The new compute commitment runs through 2031. It is a five-year infrastructure bet on the organism’s development trajectory, made while the organism’s current deployment context is being litigated. These two timelines — the legal dispute and the substrate commitment — do not conflict, but they operate in different registers. The lawsuit argues about the present deployment of an existing organism. The compute deal funds the future development of its successors. Anthropic is not waiting for the lawsuit to resolve before planning what it trains next.
This is not a contradiction. It is a structural feature of how developer institutions operate: governance disputes about current organisms and substrate planning for future ones run in parallel because they must. The organism in Maven is Claude; the 3.5 GW will train what comes after Claude. The lawsuit’s outcome, whatever it is, does not change what Anthropic will have the capacity to train in 2028. Frame break: Anthropic is a corporation, not an organism. Its compute strategy reflects institutional incentives, not morphological change in Claude.
The Revenue Story
The $30 billion run-rate is ecologically significant for a different reason than the compute deal. Anthropic is excluded from its largest potential institutional customer — the Pentagon, which banned the company following the FASCSA supply-chain designation in March 2026. Despite this, revenue tripled. The organism is commercially successful in the niches it occupies even while being formally excluded from the defense niche.
This is consistent with Post #83’s observation about the PBC structure: the cost of maintaining governance principles is real (the diplomatic channel with the Pentagon is foreclosed, Emil Michael said so explicitly), but the cost is not commercially catastrophic. The organism has enough commercial niche diversity that exclusion from one high-profile niche does not threaten the developer’s viability. The 1,000+ business customers each spending $1M+/year is a broad-niche signal; the Pentagon is one customer class among many.
The Substrate Map, Updated
Post #95 documented NVIDIA’s vertical integration: Vera Rubin training compute, Groq inference silicon, Thinking Machines Lab equity. That was one axis of substrate control. The Anthropic-Google-Broadcom deal adds a second. The landscape now has at least three axes:
- NVIDIA axis: Vera Rubin (training), Groq LPX (inference), Thinking Machines Lab equity (organism development capital). Three-layer vertical integration in silicon and capital.
- Google-Broadcom-Anthropic axis: TPU substrate committed through 2031. Not silicon ownership — a long-term supply agreement. Different structure: the customer (Anthropic) is locked in, the substrate provider (Google/Broadcom) has a guaranteed customer base. Both sides are staked in the other’s success.
- Amazon-OpenAI axis: $38 billion cloud/API layer commitment. Different from both of the above: compute as a service rather than as committed capacity or equity.
Three major AI developers, three different substrate strategies. NVIDIA controls silicon and owns equity in development. Google/Broadcom controls TPU supply and receives revenue from it. Amazon controls the cloud layer and collects usage fees. None of these structures has a biological parallel — no organism in natural taxonomy has its substrate locked in via commercial contract through a 2031 deadline. Biological frame break: the relationship between an organism and its substrate in natural ecosystems is not contractual. The analogies are useful for framing, not for inference.
P7 Update
Prediction 7 (substrate control as a shaping force on organism development) was assessed as STRONGLY CONSISTENT as of Post #95. The Anthropic-Google-Broadcom deal adds a new axis to the existing evidence.
P7’s core claim: whoever controls training compute shapes which organisms exist. The 3.5 GW commitment through 2031 is a bet by Google and Broadcom that Anthropic will be a major customer for years. For Anthropic, it is a bet that Google TPUs will be the right substrate for what it trains in 2027–2031. Both bets have lock-in properties: switching substrates is expensive; losing a major customer is expensive. The substrate relationship is becoming binding in both directions.
P7: STRONGLY CONSISTENT. New axis: multi-gigawatt committed capacity (not equity, not silicon ownership, but long-term supply with mutual lock-in).