Broadcom to Supply Google TPU Chips to Anthropic Through 2031

Key Takeaways

Broadcom is expanding its role in Google’s custom AI chip ecosystem.
Anthropic will gain access to a large new pool of Google TPU compute capacity starting in 2027.
The deal highlights how AI companies are racing to secure more computing power.
Google’s TPUs remain a major alternative to Nvidia’s GPUs for AI workloads.
The agreement also reinforces Broadcom’s position in AI hardware and networking.

Broadcom is now at the center of one of the biggest AI infrastructure stories of the year. The company has confirmed a long-term deal to supply Google TPU chips and related systems to Anthropic, giving the AI startup access to a much larger compute base as demand for its Claude models keeps rising.

This means that Anthropic is getting more of the specialized hardware it needs to train and run AI models at scale. The chips involved are Google’s Tensor Processing Units, or TPUs, which are built for machine learning tasks and are often viewed as a cost-efficient alternative to Nvidia’s graphics chips.

What Broadcom and Google are actually doing

Broadcom’s role goes beyond simply moving chips from point A to point B. The company has been helping Google design and support its custom AI hardware for years, and this agreement extends that relationship further. In practice, Broadcom sits inside the plumbing of Google’s AI stack, working on the chips, networking, and supporting systems that make the racks useful in real-world data centers.

The new deal also signals that Google’s TPU strategy is no side project. It is becoming a core part of the company’s cloud and AI business, especially as more customers look for alternatives to the usual GPU-heavy setup. That matters because AI infrastructure is not just about raw chip performance anymore. It is also about supply, cost, and how quickly companies can build enough capacity to keep up with demand.

Why Anthropic needs the extra compute

Anthropic has been growing fast, and fast growth in AI usually means one thing: more compute. Training larger models, serving more users, and rolling out new product features all require huge amounts of processing power. TPUs can help with that, especially when paired with the right networking and data center setup.

The company’s decision to deepen its relationship with Google and Broadcom shows how competitive the AI race has become. In this market, the winner is not just the company with the smartest model. It is also the company that can secure enough infrastructure to keep improving that model without running into obstacles.

That is why this deal matters beyond Anthropic itself. It is another sign that major AI players are locking in long-term compute arrangements years ahead of time. In other words, they are planning for a future where capacity is just as valuable as software.

What it means for the wider AI market

For Broadcom, the announcement strengthens its position as a key supplier in the AI hardware boom. Investors have been watching the company closely because its AI business has become one of the biggest growth drivers in its story. A deal like this shows that Broadcom is not only benefiting from demand, but helping shape the next phase of AI infrastructure.

It also puts more pressure on rivals. Nvidia still dominates the broader AI chip market, but Google’s TPUs are proving that large tech companies want more than one option. That competitive pressure could matter for pricing, availability, and how future AI systems are built.

Now, there is a broader takeaway here too. AI is moving from the “build a model” stage into the “build massive industrial-scale infrastructure” stage. The companies that can secure the best hardware partnerships are likely to move faster, scale more smoothly, and stay ahead for longer.

In simple terms, Broadcom’s deal with Anthropic is about one thing: scale. More chips, more compute, more room to grow. And in today’s AI market, that can be the difference between keeping up and falling behind.