Broadcom Inc. (NASDAQ: AVGO) has announced two major, long-term artificial intelligence (AI) chip agreements, positioning itself at the center of the booming AI infrastructure market. The semiconductor giant has secured a partnership with Alphabet's Google to develop and supply custom AI chips and hardware through 2031. Concurrently, Broadcom expanded its partnership with AI startup Anthropic, agreeing to provide the company with access to approximately 3.5 gigawatts of computing capacity built on Google's Tensor Processing Unit (TPU) infrastructure, starting in 2027.
The deal with Google reinforces the tech giant's strategy of building vertically integrated AI systems to reduce reliance on external chipmakers like Nvidia. Broadcom will be a key development and supply partner for Google's next-generation AI infrastructure, including custom TPUs and supporting networking hardware. This agreement extends a nearly decade-long collaboration and locks in a multi-year roadmap for Google's AI hardware expansion.
The separate agreement with Anthropic underscores the scale of demand. The AI startup's revenue has reportedly more than tripled, climbing from around $9 billion at the end of 2025 to a run-rate exceeding $30 billion by the close of March 2026. Anthropic's Chief Financial Officer, Krishna Rao, stated the capacity is necessary "to serve the exponential growth we have seen in our customer base while also enabling Claude to define the frontier of AI development." The compute deal could cost Anthropic hundreds of billions of dollars, with industry estimates suggesting setting up a single gigawatt of power costs between $35 billion and $50 billion.
Market reaction was mixed initially, with Broadcom shares experiencing slight pressure in one session as investors weighed valuation concerns, but rising roughly 3% in after-hours trading following the announcement. Analysts view the contracts positively. D.A. Davidson analyst Gil Luria noted the deals "put the spotlight back on Broadcom as a major winner," with some market commentators suggesting significant share price upside.
The announcements highlight a broader industry trend of hyperscalers and AI companies seeking alternatives to traditional GPU-heavy architectures. Anthropic confirmed it runs its Claude network on several platforms, including AWS Trainium, Google TPUs, and Nvidia GPUs, arguing that using different chips enhances performance and resilience. Similarly, OpenAI has been seeking alternatives to Nvidia's GPUs, having reached separate agreements with AMD, though both Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman have publicly emphasized their ongoing close partnership.
