Anthropic just signed its largest infrastructure deal yet. On April 6, the company announced a major expansion of its Google Cloud and Broadcom partnership, securing approximately 3.5 gigawatts of next-generation TPU capacity coming online from 2027, adding to the 1 gigawatt already in use in 2026. The majority of that compute will be U.S.-located.

The compute announcement was accompanied by revenue figures that put the scale in context. Anthropic’s revenue run rate has crossed $30 billion — up from $9 billion at the end of 2025. That is a tripling in roughly four months. More than 1,000 enterprise customers are now spending over $1 million per year with the company, a figure that has more than doubled since February.

Bloomberg confirmed the Broadcom portion of the deal. Google Cloud’s press corner carried the official announcement.

What This Deal Actually Signals

The headline numbers are remarkable enough on their own, but the more interesting signal is structural.

AI labs have been talking about the importance of compute for years. What this deal represents is a shift from speculative to contracted. Anthropic is not betting on GPU spot markets or hoping cloud availability aligns with demand. They are locking in 3.5 gigawatts of dedicated compute — roughly the generating capacity of three large power stations — for a multi-year period.

That kind of infrastructure commitment only makes sense if you have confident visibility into revenue growth. The $30 billion run rate and the 1,000-plus million-dollar enterprise accounts give that confidence.

It also signals something important about the competitive dynamics in enterprise AI. Anthropic, which operates Claude as its primary product, has historically been framed as the safety-focused research lab playing second fiddle to OpenAI in commercial deployment. The February-to-April doubling of million-dollar enterprise customers suggests that framing needs updating.

Why Enterprise Customers Are Choosing Claude

The $1M+ customer count matters as an indicator of where enterprise AI spend is going beyond pilots. At that spend level, companies are not experimenting — they are running Claude in production across meaningful workflows.

Several factors explain the acceleration. Claude has been gaining ground in code-heavy enterprise workflows, particularly since Claude Code’s expansion to the broader developer market. The multi-model reality of enterprise AI has also played to Anthropic’s advantage: companies running Microsoft Copilot, Gemini, and Claude simultaneously are discovering that Claude tends to perform best on complex reasoning and long-document tasks, and the $1M spend threshold often reflects that kind of complementary deployment.

The 3.5GW compute deal also addresses a concern that has quietly limited enterprise deals: capacity guarantees. With the TPU infrastructure locked in, Anthropic can now offer enterprise customers more confident commitments about throughput and latency at scale.

The Infrastructure Race Is Not Slowing Down

Spending on AI compute has not peaked. If anything, the Anthropic deal — alongside recent announcements from Microsoft (its own MAI model lineup), Google (Gemma 4 on its own hardware), and Nvidia’s accelerating chip roadmap — points to an infrastructure buildout that is entering its most capital-intensive phase yet.

For businesses watching from the sidelines, the practical takeaway is not about which company wins the infrastructure race. It is that the models available to enterprise customers are going to keep improving for years, the cost per token will keep declining as compute efficiency improves, and the enterprise AI window is not closing — it is widening.

What This Means for Business

The biggest risk for businesses right now is not that AI will be too expensive or too limited. It is that the infrastructure is accelerating faster than most organisations’ ability to deploy it thoughtfully.

Anthropic signing a 3.5GW deal and tripling revenue in four months is a signal that enterprise AI adoption has crossed from early adopter to mainstream. The companies showing up late to that transition will be catching up to competitors who have had AI agents handling real workflows for 12 months or more.

If your business is still at the “evaluating AI” stage, the infrastructure news from this week is a prompt to move faster on deployment, not slower. The tools are enterprise-ready. The question is whether your workflows are ready for them.

---

If this is the kind of problem agents can help with, the free Working With Claude field guide is the practical next step. Thirty-two pages, no fluff. Get the free guide.