Anthropic locks in multi-gigawatt TPU deal as revenue triples

Running out of compute is an everyday problem in AI, but very few players can sign off on consuming the output of multiple nuclear plants at once.

Anthropic announced an expanded multi-gigawatt compute agreement with Google and Broadcom for next-generation TPU capacity, set to come online starting in 2027. While Anthropic simply referred to the deal as multiple gigawatts, a recent Broadcom SEC filing spilled the exact number: a massive 3.5 gigawatts.

This buildout builds directly on the 1-gigawatt agreement the companies struck in October 2025, and extends Anthropic's pledge from November 2025 to pour $50 billion into American computing infrastructure.

Explosive growth forces an infrastructure dash

The driving force behind this massive compute landgrab is straightforward: enterprise demand for Claude is surging faster than the company can provision servers.

Anthropic revealed that its run-rate revenue has crossed $30 billion, up dramatically from roughly $9 billion at the close of 2025. That is more than a 3x leap in just a few months.

Big-spending enterprise customers are stacking up at record speed. When the lab announced its Series G round in February, it counted around 500 enterprise customers spending over $1 million annually. Less than two months later, that figure doubled to more than 1,000.

Even external headwinds barely slowed the momentum. As TechCrunch reported, enterprise adoption surged despite the U.S. Department of Defense previously labeling Anthropic a supply-chain risk. The outlet also noted that Anthropic recently closed a $30 billion Series G round, pegging its valuation at $380 billion.

Hedging every bet across chips and clouds

Beyond sheer scale, the deal highlights Anthropic's deliberate strategy to avoid getting locked into a single chipmaker or cloud ecosystem.

While rivals frequently build around a single dominant GPU vendor, Anthropic spreads Claude across a mosaic of hardware, training and serving workloads across AWS Trainium, Google TPUs, and NVIDIA GPUs.

Amazon remains Anthropic's primary cloud and training partner through ongoing work on Project Rainier. Yet by inking billions in TPU deals with Google and Broadcom, Anthropic ensures it has dedicated silicon lines tuned specifically for its architectures.

The same neutrality applies to its distribution. Claude stands out as the only frontier AI model available across all three hyperscalers: AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry.

Why it matters

For enterprise customers and developers, this deal offers real reassurance. Over the past few quarters, Claude's sudden popularity occasionally caused capacity bottlenecks. Knowing that 3.5 gigawatts of dedicated compute will begin rolling out in 2027 signals that Anthropic has the staying power to support long-term enterprise migrations.

For the broader AI ecosystem, this partnership accelerates the shift away from single-vendor silicon dominance. We believe Anthropic's deep commitment to TPUs co-developed with Broadcom proves that custom ASICs are no longer a backup plan for frontier labs; they are primary infrastructure. It cements Broadcom's spot at the center of the AI hardware boom.

That said, the stakes could hardly be higher. Committing to gigawatts of compute years in advance requires Anthropic to sustain its dizzying revenue trajectory without stumbling. Any delivery delay or shift in model efficiency before 2027 could leave it balancing astronomical fixed overheads. The frontier race is officially a heavy-industry game now.