Anthropic Weighs In-House AI Chips to Secure Computing Independence
Anthropic has long relied on Nvidia GPUs and computing capacity and custom chips supplied by partners including Google and Broadcom to train and run its Claude models. As model sizes and enterprise demand grow, developing chips in-house could diversify supply risks, lower costs and strengthen Anthropic’s bargaining power with cloud and semiconductor providers.
As of July 19, 2026, Anthropic was evaluating whether to design its own AI chips and was reportedly in talks with Samsung about a custom-chip partnership. The company continues to diversify its supply chain even after recently signing long-term contracts with Google and Broadcom. Its annualized revenue has reportedly surpassed $30 billion, but it has yet to formally settle on a chip strategy.
All Coverage
5 original reportsThe Backstory
The history behind this eventAnthropic in Talks to Buy Fractile Inference Chips to Cut AI Operating Costs
Claude developer Anthropic currently uses chips from NVIDIA, Amazon and Google. AI inference costs are squeezing profits, making supplier diversification important to operational resilience. Fractile was founded in 2022 by Oxford PhD Walter Goodwin and places SRAM and compute units on the same die to reduce data-transfer latency and energy consumption.
Reports on May 2, 2026, said Anthropic was in preliminary talks to buy inference chips from London-based startup Fractile, with the value and volume undisclosed. Fractile claims its chips can run large language models 100 times faster than NVIDIA GPUs at one-tenth the cost and plans to bring them to market in 2027. The company is also seeking to raise $200 million at a valuation of more than $1 billion.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.