Anthropic has just sent a strong signal to the AI market: the race is no longer being fought only in models, but in megawatts. The company announced a new agreement with Google and Broadcom to secure multiple gigawatts of next-generation TPU-based capacity, infrastructure expected to come online starting in 2027 and power Claude’s continued push at the frontier. At first glance, this can look like backend news — a story about data centers and industrial contracts. In reality, it is one of the most important developments right now because it makes clear that, in 2026, AI advantage increasingly depends on who can lock in energy, chips, and cloud capacity before everyone else.

According to Anthropic, this is its biggest compute commitment so far. The company says its annualized revenue has now surpassed $30 billion, up from roughly $9 billion at the end of 2025, and that it now has more than 1,000 enterprise customers each spending over $1 million per year. That kind of growth helps explain why Anthropic is not merely tuning inference or squeezing more out of existing hardware. It is reserving the infrastructure it expects to need one or two years from now. Put differently, Anthropic is buying future training and deployment time before scarcity tightens again.

The interesting part of the announcement is how the roles are split. Google brings its TPUs to the table, Broadcom remains a critical piece in designing and scaling those chips, and Anthropic provides enough demand to justify a huge expansion. CNBC added a concrete figure that helps size the move: the expanded deal would give Anthropic access to around 3.5 gigawatts of capacity based on Google processors. That no longer sounds like a promising lab; it sounds like industrial-scale infrastructure. In practice, the conversation shifts from “which model is best” to “which company can sustain an entire compute factory without running short.”

TechCrunch also highlighted an important point: this expansion is not coming out of nowhere. It builds on the agreement Anthropic announced in 2025 to use more Google Cloud TPUs. Now the message is more ambitious. This is not just about gaining access to hardware that can complement or compete with NVIDIA. It is about building a strategic mix across AWS Trainium, Google TPUs, and NVIDIA GPUs. Anthropic explicitly emphasizes that diversity: training and running Claude across multiple platforms so each workload can be matched to the chip best suited for it. That flexibility matters because it reduces dependence on a single vendor while also strengthening bargaining power at a stage when every point of efficiency matters.

There is also a geopolitical and industrial reading here. Anthropic says the vast majority of this new capacity will be located in the United States, as an extension of its pledge to invest $50 billion in American compute infrastructure. That fits a broader trend: the top labs want more powerful models, but also more stable supply chains, more domestic capacity, and less exposure to regulatory shocks or global bottlenecks. Frontier AI is no longer just software; it is industrial policy, energy, semiconductors, and technological sovereignty.

Another important layer is what this says about Google. For a long time, the market viewed Google as a cloud provider, a model competitor, and a chip designer — but not always as the major outside winner from the boom in third-party AI labs. This deal changes that perception a bit. If Anthropic scales Claude on Google TPUs while Broadcom keeps building muscle for that platform, Google is not only competing with Gemini; it is also monetizing the model war around it. That makes its TPUs a more strategic asset against NVIDIA’s long-standing dominance.

For the broader public, the takeaway is simple: the next major AI battle will not just be about who has the friendliest chatbot or the highest benchmark. It will be about who can guarantee enough compute to keep training, serve millions of queries, and satisfy enterprise customers without costs spiraling out of control. Anthropic just showed that it understands that reality — and that it is willing to sign enormous deals so it does not fall behind. At this stage, the future of AI is being written with prompts and watts alike.