Broadcom’s Custom AI Chips Deal With Google and Anthropic Explained

• Anthropic is set to access 3.5GW of custom AI chips capacity through Broadcom and Google starting in 2027
• One line in Broadcom’s regulatory filing tells a more cautious story than the press release does

When Broadcom filed an 8-K with the US Securities and Exchange Commission on April 6, it did something routine in form but remarkable in substance. In a single page of regulatory language, the company confirmed two things that reframe the AI infrastructure race: a long-term agreement with Google to develop and supply custom AI chips for future generations of Tensor Processing Units, and an expanded collaboration with Anthropic that will give the Claude maker access to approximately 3.5 gigawatts of TPU-based compute capacity starting in 2027.

That is not a small number. For context, Broadcom CEO Hock Tan had noted on an earnings call just last month that Anthropic was consuming around one gigawatt of compute in 2026. The new commitment triples that before the year is even out.

The custom AI chips deal and what it actually means

Broadcom has been Google’s behind-the-scenes silicon partner since 2016, quietly designing TPUs while Google handled the commercial front. The new agreement formalises and extends that relationship significantly, covering not just future TPU generations but also networking and other components for Google’s next-generation AI racks through 2031.

That last detail matters. Broadcom is not just a chip designer in this arrangement. It is becoming a full-stack infrastructure partner, supplying the interconnects and components that tie these systems together at scale. Analysts at Mizuho, led by Vijay Rakesh, have estimated that Broadcom could pull in US$21 billion in AI revenue from the Anthropic relationship alone in 2026, rising to US$42 billion in 2027.

Whether those figures hold will depend on how quickly Anthropic’s growth compounds, but the trajectory is no longer speculative.

Anthropic’s growth numbers demand attention

Anthropic used its own blog post on April 6 to reveal something that most of the subsequent coverage treated as a footnote. Run-rate revenue has crossed US$30 billion, up from approximately US$9 billion at the end of 2025. When the company announced its Series G fundraising in February, it had over 500 business customers, each spending more than US$1 million annually.

Today, that number exceeds 1,000–a doubling in under two months. That is the commercial engine behind the infrastructure commitment. Krishna Rao, Anthropic’s CFO, framed it plainly: “We are making our most significant compute commitment to date to keep pace with our unprecedented growth.”

The company is also careful to signal that it is not betting the stack on any single platform. Claude is trained and deployed across AWS Trainium, Google TPUs, and NVIDIA GPUs–a multi-architecture approach that, as Anthropic puts it, allows the company to match workloads to the chips best suited for them.

Amazon remains its primary cloud and training partner, and Claude is the only frontier AI model currently available across all three major cloud platforms: AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry.

The line in the SEC filing that everyone should read

Here is where the story gets more interesting than the press releases suggest. Broadcom’s 8-K contains a sentence that sits in some tension with the scale of the announcement: “The consumption of such expanded AI compute capacity by Anthropic is dependent on Anthropic’s continued commercial success. In connection with this deployment, the parties are in discussions with certain operational and financial partners.”

That is a public company, in a regulatory document, signalling that the financial architecture behind a multi-gigawatt deployment is not yet fully settled. It does not undercut the deal, but it does mean the 3.5 gigawatts figure is conditional, not guaranteed. For enterprise buyers and infrastructure planners tracking this space, that distinction has real planning implications.

What shifts for the custom AI chips market

The broader significance here is what this triangular relationship–Broadcom building, Google enabling, Anthropic consuming–does to the competitive dynamics of AI silicon. Custom AI chips designed for specific workloads have long been positioned as a more cost-efficient alternative to general-purpose GPUs at scale.

Broadcom is now the clearest proof point that this thesis is playing out in practice, not just in theory. It has already signed a multi-year partnership with OpenAI, inked deals with a fifth undisclosed XPU customer, and now expanded its Google-Anthropic arrangement into something approaching infrastructure dominance.

The custom AI chips segment is no longer a challenger narrative. It is, structurally, the next phase of how AI compute gets built and delivered. Anthropic has committed to siting the vast majority of the new compute in the United States–framing the deal as an extension of its November 2025 pledge to invest US$50 billion in American computing infrastructure.

For hyperscalers and cloud-dependent enterprises in the Asia Pacific, that US-centric buildout is a data point worth tracking: as the most capable AI models increasingly run on infrastructure anchored in a single geography, questions around latency, resilience, and data sovereignty are only going to sharpen.

The compute race is no longer abstract. It is filed with the SEC, signed by Hock Tan, and coming online in 2027.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology eventsclick here for more information.