Broadcom has entered a long-term agreement with Google to develop and supply custom Tensor Processing Units, or TPUs, for future generations of Google’s AI chips, according to a filing Broadcom submitted to the U.S. Securities and Exchange Commission on April 6, 2026. The agreement also covers networking and other components for Google’s next-generation AI racks, with the supply arrangement running through 2031. Reuters reported the same day that the deal was confirmed in Broadcom’s filing and said Broadcom shares rose in extended trading after the announcement.
The announcement gives a clearer picture of how Google is building its AI infrastructure. Rather than relying only on third-party processors, Google continues to push its own TPU line, and Broadcom is a key partner in that effort. The SEC filing says Broadcom will develop and supply custom TPUs for Google’s future generations of TPUs, while also supplying networking and other parts used in Google’s AI racks. Reuters said demand for custom chips such as TPUs has increased as companies look for alternatives to Nvidia’s graphics processors for AI workloads.
Broadcom’s filing points to a multi-year framework that runs “through up to 2031,” which suggests the relationship is intended to support Google’s AI buildout over several product cycles. For a company like Google, that kind of arrangement helps secure hardware supply and planning visibility as AI infrastructure grows more demanding and for Broadcom, it strengthens its position in a market where the most valuable work is shifting toward custom silicon, rack-level networking and the wider systems that keep large AI clusters running.
The filing also shows that Broadcom’s role is broader than chip design alone, it includes supply assurance for networking components and other parts used in Google’s next-generation AI racks. That matters because modern AI systems are not defined only by the accelerator chip. They depend on the surrounding infrastructure: high-speed interconnects, server design, power delivery, and rack-level integration. In practical terms, Broadcom is helping support the plumbing around Google’s TPU deployment, not just the processors themselves.
Reuters said the deal reflects a wider industry trend, as AI spending rises, hyperscale cloud companies are trying to build custom hardware that can be more efficient, more tightly integrated with their software, and less dependent on general-purpose GPUs. Google has spent years developing TPUs for that purpose, and TPU sales have become an important part of Google Cloud’s revenue story. The new agreement suggests that Google plans to keep that strategy in place and expand it over a longer horizon.
[Related Post: ING Partners with Broadcom to Modernize Private Cloud Infrastructure with VMware Cloud Foundation 9.0]
Financial terms were not disclosed, so there is no public valuation attached to the Google agreement itself. Even so, the market read-through was immediate: Broadcom’s shares gained about 3% in extended trading after the news. That reaction suggests investors view the agreement as another sign that Broadcom has secured a durable place in the AI supply chain, particularly in custom silicon and the infrastructure that supports it.
The filing confirms that the two companies are not simply working on a one-off chip project. They are extending a manufacturing and supply relationship that reaches into Google’s next generation of TPU-based systems and into the networking layers that keep those systems operating at scale.
