^{Photo: South China Morning Post}

Broadcom has officially disclosed that its previously unnamed client responsible for a 10 billion dollar custom chip order earlier this year is Anthropic, one of the fastest-growing artificial intelligence labs in the world. The revelation came during Broadcom’s latest quarterly earnings call, where CEO Hock Tan detailed the extent of Anthropic’s recent procurement and confirmed that the AI company has since placed an additional 11 billion dollar order for custom hardware.

According to Tan, Anthropic’s initial purchase consisted of Google’s newest Ironwood tensor processing unit racks, built with Broadcom’s custom ASIC technology. These chips are central to Google’s AI infrastructure roadmap, and industry analysts have long viewed them as some of the strongest competitors to Nvidia’s market-dominant GPUs. Broadcom’s custom AI chip line, which it refers to as XPUs, is designed to deliver greater efficiency for specialized workloads such as large-scale model training and inference.

The identity of the customer has been a major point of speculation since September, when Broadcom first disclosed the 10 billion dollar agreement without naming the buyer. Prior to Thursday’s announcement, company officials had confirmed only that the mysterious client was not OpenAI. The new disclosure highlights how aggressively Anthropic has been scaling its compute resources as the demand for frontier models intensifies across the industry.

In addition to the Anthropic contract, Broadcom confirmed that it secured a fifth major custom chip client in the fourth quarter, with that customer placing a 1 billion dollar order. While the company did not identify the buyer, leadership emphasized that the deal represents a long-term growth opportunity for its AI hardware division.

Expanding the Google–Anthropic Cloud Partnership

Anthropic’s deepening investment in Broadcom-generated TPU infrastructure aligns with its broader partnership with Google Cloud. The two companies announced a wide-ranging collaboration in late October valued in the tens of billions of dollars. The agreement grants Anthropic access to up to one million Google TPUs and is projected to bring more than one gigawatt of new compute capacity online as early as 2026. This scale makes it one of the largest commitments to cloud-based AI infrastructure ever publicly disclosed.

Anthropic’s approach to compute infrastructure is intentionally diversified. The company uses Google’s TPUs for high-efficiency model training, Amazon’s Trainium chips for cost-effective scaling, and Nvidia GPUs for specific research and inference workloads. This multi-cloud, multi-chip strategy enables Anthropic to optimize performance, availability, and cost across its growing suite of AI models.

Google has welcomed Anthropic’s expanded reliance on its TPU architecture. Executives at Google Cloud have repeatedly emphasized the performance advantages and energy efficiency of its ASIC-based systems. Analysts increasingly point to the TPU platform as a crucial differentiator for Google as it competes with Amazon and Microsoft for leadership in AI cloud services.

TPUs have also gained credibility as the most viable alternative to Nvidia’s rapidly constrained GPU supply. With power availability becoming a major bottleneck for data center expansion, Google’s efficient chip designs and vertically integrated infrastructure could represent a structural advantage as global demand for AI compute continues to accelerate.

Market Implications and Strategic Significance

The confirmation that Anthropic is behind more than 21 billion dollars in Broadcom chip orders underscores the scale at which AI firms are expanding hardware investment. It also reinforces the recent shift in the semiconductor industry, where custom silicon and full-stack rack delivery are becoming central to competitive strategy. Broadcom supplying entire server racks, not just chips, represents a significant move up the value chain and positions the company as a critical supplier in the evolving AI infrastructure market.

For investors, the announcement further validates the rapidly growing economic impact of custom AI accelerators. As companies like Anthropic, OpenAI, and Google push into next-generation model development, demand for high-performance compute hardware is expected to rise dramatically over the next five years.

Broadcom’s continued traction in the AI hardware segment is now widely viewed as one of the company’s most important growth drivers. With multiple multibillion-dollar clients secured and more in negotiation, the company has solidified its position as a core supplier powering the next wave of AI innovation.