Last Updated on April 10, 2026 by Alphabet Insider Staff
The deal covers custom TPU design, data center networking, and a major Anthropic compute commitment starting in 2027
Google and Broadcom just inked a long-term partnership to design and supply future generations of Google’s custom AI chips. The agreement runs through 2031. Filed with the SEC on April 6, the deal extends a collaboration dating back to 2016 and covers tensor processing units (TPUs) plus the networking hardware powering Google’s next-generation AI data center racks. Yet again, Alphabet continues to make good on its promise of betting on AI.
The two companies have been co-designing custom TPUs since 2016, and the seventh generation, called Ironwood, is now in production.
Why This Deal Matters
This is a major play for Alphabet. Teaming up with Broadcom on custom AI chips is a deliberate move to reduce dependence on any single GPU vendor. By deepening its custom silicon program with Broadcom, Google’s building an alternative to Nvidia that it controls end-to-end. Custom AI chips typically deliver better performance per watt than general-purpose GPUs, and because they’re designed in lockstep with Google’s software stack, they integrate more tightly with the frameworks Google uses for large-scale model training and inference. That combination is hard to replicate.
The deal also fits into Google’s broader AI spending push. The company has committed billions in capital expenditure toward AI infrastructure, and this multi-year chip agreement is a core part of that buildout. The arrangements include AI accelerators and networking hardware that support large-scale AI workloads, and the multi-year structure gives Broadcom clearer visibility into future AI-related demand and capacity planning.
The Anthropic Connection
Broadcom also signed a deal with Anthropic to provide the AI startup access to about 3.5 gigawatts of AI computing capacity drawing on Google’s AI processors, starting in 2027. For context, Anthropic was consuming around one gigawatt of compute in 2026, making this a substantial step up. One catch: Broadcom’s SEC filing notes that “the consumption of such expanded AI compute capacity by Anthropic is dependent on Anthropic’s continued commercial success,” so it’s not an unconditional guarantee.
What It Means for Broadcom
For Broadcom, this is a major long-term anchor customer relationship. It’s not selling commodity silicon into an open market. It’s the design and supply partner of record for one of the world’s largest AI operators, covering chips, networking, and full rack-level infrastructure. That depth of involvement cements its role as a top-tier custom chip and networking supplier for hyperscalers. CEO Hock Tan has said he has “line of sight to achieve AI revenue from chips, just chips, in excess of $100 billion in 2027,” a striking number given total company revenue was under $64 billion in fiscal 2025.
A Longer-Term Play for ChromeOS
The deal is focused on cloud infrastructure, but the ripple effects could eventually reach consumer hardware, like Chromebooks. As Google’s TPU capabilities mature, OEM partners like Acer, Lenovo, and HP could benefit from that performance trickle-down in higher-end Chromebooks. That includes on-device AI and more demanding workloads on ChromeOS.
Nothing’s confirmed yet, however, Google has a consistent track record of moving cloud-grade capabilities closer to the edge. That makes it a very reasonable trajectory for Chromebook fans to keep a watchful eye on.
TPU sales have become a crucial growth engine of Google’s cloud revenue as it seeks to demonstrate to investors that its AI investments are generating real returns. By locking in Broadcom through 2031, Google’s betting its homegrown chip program stays central to how it competes in AI, cloud, and beyond.
If the partnership scales as planned, Broadcom-built AI racks could quietly underpin much of Google’s AI-driven search, advertising, and cloud workload for the rest of the decade.
