A Multi-Year Commitment to Custom AI Silicon
Broadcom and Alphabet have entered into long-term agreements extending through 2031 that formalize and deepen their collaboration on custom artificial intelligence hardware. Under this partnership, Broadcom will develop and supply tensor processing units (TPUs) for Google, while a separate supply assurance agreement ensures Broadcom will also provide networking components and related hardware used in Google's next-generation AI racks through at least the end of the decade.
This deal is significant because it goes well beyond chip design. It formalizes Broadcom's role not only in Alphabet's custom AI silicon but also in the surrounding rack-level infrastructure — particularly networking and other components that are increasingly becoming central bottlenecks for scaling AI clusters. As companies race to build ever-larger AI training and inference systems, the ability to move data efficiently between processors is just as critical as the processors themselves.
Anthropic Joins the Equation
The agreements also encompass an expanded collaboration involving Anthropic, the AI safety company. Starting in 2027, Anthropic will gain access to approximately 3.5 gigawatts of next-generation TPU-based AI compute capacity through Broadcom's infrastructure. However, Broadcom has been explicit about a key contingency: Anthropic's consumption of this expanded compute capacity is dependent on its overall continued commercial success. Financial terms of the arrangement were not disclosed.
This three-way dynamic is noteworthy. It illustrates how the AI infrastructure ecosystem is evolving into layered partnerships where chip designers, cloud providers, and AI model developers are binding themselves together through multi-year commitments to secure capacity and supply.
The Broader Context: Demand for Alternatives to Nvidia
This partnership unfolds against a backdrop of rising demand for custom AI silicon chips. While Nvidia's GPUs remain the dominant hardware for AI workloads, major cloud customers are increasingly seeking alternatives — whether for cost, performance optimization, or supply diversification. Alphabet's TPUs represent one of the most mature efforts in this direction, and TPU sales have become an increasingly important driver of Google Cloud's growth story.
The appetite for custom silicon reflects a strategic reality: hyperscalers that depend entirely on a single supplier for their most critical infrastructure carry concentration risk. By investing heavily in proprietary chip architectures developed with partners like Broadcom, companies like Alphabet can tailor hardware to their specific workloads while reducing dependence on any one vendor.
What This Signals for AI Infrastructure Investment
Broadcom's disclosure, made via a securities filing, sends a clear signal that surging demand for generative AI infrastructure remains robust. The market responded accordingly, with Broadcom positioned to open higher on the news. For a company already in the elite trillion-dollar-plus valuation club — yet arguably underappreciated relative to its central role in the AI supply chain — this kind of catalyst reinforces just how deeply embedded it is in the buildout of next-generation compute.
The broader takeaway is straightforward: the AI infrastructure boom is not a short-term phenomenon. When two of the largest technology companies in the world formalize hardware and networking commitments stretching seven years into the future, it reflects a level of conviction that generative AI workloads will continue scaling dramatically. The companies positioning themselves at the intersection of custom silicon, networking, and cloud infrastructure stand to be among the primary beneficiaries of that sustained investment cycle.