Alphabet deepens AI ties
Alphabet has extended TPU and networking supply agreements with Broadcom through 2031 and broadened cloud compute access for Anthropic, moves that lock in hardware and capacity for large AI workloads. Market commentary frames the deals as an attempt to secure long‑duration infrastructure advantages for Google Cloud as hyperscalers compete on supply chains as well as software (simplywall.st).
Alphabet just turned an artificial intelligence supply chain into a 5-year reservation book. Broadcom disclosed on April 6 that Google extended custom Tensor Processing Unit chip and networking agreements through 2031, while Anthropic expanded its access to Google-based compute starting in 2027. (sec.gov) The Broadcom filing says Google will use Broadcom for future generations of its custom Tensor Processing Units, and a separate supply assurance agreement covers networking parts for Google’s next artificial intelligence racks through up to 2031. That means Google locked in both the engines and the cables for the same machine. (sec.gov) Anthropic is the customer on the other end of that machine. Broadcom said Anthropic will access about 3.5 gigawatts of Google Tensor Processing Unit compute beginning in 2027 as part of a larger multi-gigawatt commitment. (sec.gov) A Tensor Processing Unit is Google’s in-house artificial intelligence chip, built for training and running models instead of general-purpose computing. Google Cloud describes these chips as custom-designed accelerators optimized for artificial intelligence training and inference. (cloud.google.com) Google has been building these chips for years instead of renting all of its artificial intelligence horsepower from Nvidia. Google says work on the first Tensor Processing Unit started in 2013, and the first cloud version arrived in 2017. (cloud.google.com) The newest public generations show why Google wants control here. Google introduced its sixth-generation Trillium chip in May 2024 and its seventh-generation Ironwood chip in April 2025, with Ironwood described as its most performant and scalable custom artificial intelligence accelerator to date. (cloud.google.com) (blog.google) Anthropic is not a small side customer filling spare capacity. Anthropic said on April 6 that its run-rate revenue had passed $30 billion, up from about $9 billion at the end of 2025, and that more than 1,000 business customers now spend over $1 million a year with the company. (anthropic.com) That demand explains why the deal is measured in power-plant language instead of server counts. Anthropic said the new capacity will come online starting in 2027, and that most of the new compute will be located in the United States. (anthropic.com) Google also gets something immediate out of this beyond chip supply. Google Cloud said Anthropic is expanding its use of Google Cloud services including BigQuery, Cloud Run, and AlloyDB, and that thousands of customers already access Claude through Google Cloud. (googlecloudpresscorner.com) Anthropic is still spreading its bets across clouds, which makes Google’s win more notable. Anthropic said Amazon Web Services remains its primary cloud provider and training partner, while Claude is also available through Google Cloud and Microsoft Azure. (anthropic.com) So this is not just one chip contract and not just one cloud sale. It is Google using Broadcom to secure the hardware roadmap through 2031, then using that roadmap to sell reserved computing capacity to one of the fastest-growing model companies in artificial intelligence. (sec.gov) (anthropic.com)
