Broadcom’s billboard is photographed on September 5, 2025 in San Jose, California, as it prepares to launch new optical chip technology to fend off Nvidia.
Brittany Hosea-Small | Reuters
broadcom announced in its September earnings call that it had signed a contract with a customer who ordered $10 billion in custom chips.
At the time, Broadcom did not say who it was, but on Thursday, CEO Hock Tan revealed that the mystery customer was AI Labs Anthropic, which had ordered the latest software. google Tensor processing unit.
“We received a $10 billion order to sell our latest TPU ironwood racks to Anthropic,” Tan said on Broadcom’s fourth-quarter earnings call on Thursday. He also said Anthropic placed an additional $11 billion in orders with Broadcom in the company’s most recent quarter.
Broadcom doesn’t typically disclose its large customers, but Tan’s comments in September attracted significant investor attention amid the AI infrastructure boom. Broadcom officials told CNBC in October that the mystery customer is not OpenAI, but has its own contracts to buy chips from chip makers.
Broadcom makes custom chips called ASICs, which some experts believe are more efficient for certain artificial intelligence algorithms than Broadcom’s chips, which dominate the market. Nvidia. Broadcom helps make Google’s TPUs, and last month the search company boasted that it had trained its cutting-edge Gemini 3 model entirely on its own TPUs.
The chipmaker calls its custom AI chips XPUs, and Tan said Thursday that his company is delivering not just chips but entire server racks to Anthropic, Broadcom’s fourth XPU customer.
Broadcom also announced Thursday that it has secured a fifth customer for its custom chip business. That customer placed $1 billion in orders during the fourth quarter, but again, Broadcom did not identify the customer.
“This is a real customer and will continue to grow,” Mr Tan said.
Google’s contract with humanity
In late October, Anthropic and Google announced a wide-ranging cloud partnership that highlighted how rapidly the race for AI infrastructure is moving.
The deal is worth tens of billions of dollars. The deal gives Anthropic access to as many as 1 million Google TPUs and is expected to bring well over a gigawatt of new AI computing capacity online by 2026.
Anthropic employs a multi-cloud, multi-chip strategy.
The startup distributes workloads across Google’s TPUs, Amazon’s Custom Trainium chip and Nvidia graphics processing unit. Anthropic’s various models are tailored to run on the most suitable platform for training, inference, and research.
For Google, Anthropic’s TPU bet is exactly the kind of validation that investors have been rewarded with, with Wall Street increasingly linking Alphabet’s stock price rise to waning demand for its AI chips.
Google Cloud CEO Thomas Kurian previously said Anthropic’s decision to dramatically increase its TPU usage reflects the “excellent price performance and efficiency” the company has experienced over several years.
After more than a decade of in-house development, Google is making its TPUs widely available to cloud customers as a service rather than selling the hardware outright, relying on its chips as a differentiator to meet the surge in demand for AI computing.
Analysts say TPUs are the most reliable replacement for Nvidia GPUs. Analysts argue that Google’s tightly customized ASICs and power-efficient design could be a meaningful driver of growth for the company’s cloud business, as power constraints, rather than chip supply, emerge as the primary bottleneck for AI.
