Anthropic Locks 3.5GW of TPU Capacity. Here Is What It Means for Every Builder on the Claude API.

The Signal Buried in the Infrastructure Deal

On April 6, 2026, Anthropic disclosed two facts simultaneously. The first was the headline: its revenue run rate had topped $30 billion, up from $9 billion at the end of 2025. The second was the structural news that operators should actually be modeling: Anthropic signed an agreement with Google and Broadcom for approximately 3.5 gigawatts of computing capacity built on Google's tensor processing units, with that capacity expected to come online starting in 2027.

These are not separate announcements. Demand is real, compute is locked, and Anthropic is betting its infrastructure strategy on that combination holding through the next product cycle. Builders on the Claude API should understand exactly what that bet implies.

How the 3.5GW Deal Actually Works

The structure involves three parties. Broadcom will supply Anthropic with roughly 3.5 gigawatts of Google TPU capacity starting in 2027 and separately commit to designing and supplying future generations of Google's TPUs through 2031. This is not a simple cloud contract. Anthropic is shifting toward long-term, utility-scale compute commitments that resemble energy procurement rather than traditional cloud infrastructure.

The new 3.5GW is additive. This capacity is in addition to the 1GW already coming online in 2026 under the Google Cloud agreement announced last October, and the filing states that Anthropic's use of the expanded capacity is contingent on its continued commercial performance. That contingency matters: Broadcom is not writing a blank check. Broadcom stated explicitly: "The consumption of such expanded AI compute capacity by Anthropic is dependent on Anthropic's continued commercial success."

The vast majority of the new compute will be situated in the United States and form part of Anthropic's November 2025 pledge to invest $50 billion in strengthening US compute infrastructure.

Why TPUs, Not GPUs

The hardware choice signals strategic intent. Anthropic trains and runs Claude on a diversified range of AI hardware, including AWS Trainium, Google TPUs, and NVIDIA GPUs, matching workloads to the chips best suited for them. But locking 3.5GW of TPU capacity is a directional bet.

Google TPUs deliver predictable linear throughput gains at scale—what matters more for production inference than raw peak FLOP counts. By committing to this volume through 2031, Anthropic is co-signing Google's hardware roadmap. If Google's next-generation TPU performance slips relative to NVIDIA H-series or AMD Instinct parts, Anthropic absorbs the cost.

For competitive context: OpenAI has separately committed to 6GW of AMD GPU capacity, with the first gigawatt expected in the second half of this year. Frontier labs are now competing on compute procurement as much as model capability. The arms race is about who can guarantee inference capacity at scale.

The Revenue Trajectory That Made This Possible

The $30B run-rate number warrants scrutiny beyond the headline. Anthropic's revenue trajectory ran as follows: $87 million run rate in January 2024, $1 billion by December 2024, $9 billion by end of 2025, $14 billion in February 2026, $19 billion in March, and $30 billion in April. Each increment accelerated.

Enterprise concentration explains why the compute commitment carries credibility. Anthropic now counts over 1,000 businesses spending more than $1 million annually, a figure that doubled from over 500 in February. Customers spending over $100,000 annually grew 7x in the past year; two years ago, only a dozen customers exceeded $1 million annualized spend.

Claude Code is the primary driver. The company's agentic AI coding tool launched publicly in mid-2025 and became the fastest-growing product in the company's history, hitting $1 billion in annualized revenue within six months of launch. A recent analysis estimated that 4% of all GitHub public commits worldwide were being authored by Claude Code, double the percentage from just one month prior.

One important caveat: these are annualized run-rates extrapolated from recent monthly performance, not GAAP revenue. The jump from $14 billion to $30 billion in roughly six weeks reflects both genuine demand acceleration and annualization mechanics. TechCrunch reported in late April that while Anthropic announced a $30 billion run-rate, the company's actual run rate was closer to $40 billion, according to a person with knowledge of the company's financials. Treat these as directional guidance, not audited figures.

What This Means for Operators Building on Claude

Capacity stability. The most immediate implication: Claude API availability is unlikely to hit supply constraints through 2027. Anthropic has secured infrastructure to serve its current trajectory, reducing tail risk that demand surges trigger throttling or price spikes at critical deployment moments.

Pricing dynamics. Locked compute costs support price predictability but work against aggressive rate cuts. Operators modeling continued cost-per-token declines like the 2023-2025 period should expect a slower curve.

Vendor concentration risk. Claude remains the only frontier AI model available to customers on all three of the world's largest cloud platforms: Amazon Web Services Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. This multi-cloud availability is a genuine enterprise differentiator. The downside: Anthropic is now deeply interlocked with Google's infrastructure roadmap. Any material TPU performance or availability issue becomes an Anthropic service issue.

For long-term API commitments, Anthropic's capital position is substantial. The company raised $30 billion in Series G funding led by GIC and Coatue, valuing Anthropic at $380 billion post-money. Anthropic projects positive free cash flow by 2027, while OpenAI has pushed its breakeven target to 2030. That structural gap matters when negotiating multi-year enterprise agreements.

What to Watch

1. Q3 2026 run-rate disclosure. The $30B figure was from April. Watch whether Anthropic confirms or revises at the next public disclosure point. A plateau would signal that the enterprise surge is normalizing.

2. TPU generation specifics. The 3.5GW commitment covers "next-generation" TPUs, but Google has not publicly detailed performance specs for the generation coming online in 2027. When those specs surface, model the inference cost-per-token implications.

3. OpenAI's 6GW AMD commitment timing. OpenAI committed to 6GW of AMD GPU capacity, with the first gigawatt expected in the second half of this year. On-time delivery is a direct read on AMD's ability to compete with NVIDIA and TPUs at scale.

4. IPO timeline. Bloomberg reported that the company is weighing an IPO as early as October 2026, with Goldman Sachs, JPMorgan, and Morgan Stanley already involved. A public Anthropic fundamentally changes the disclosure environment. Operators will finally see real revenue figures, not run-rate extrapolations.