Anthropic sees adoption, eyes chips
Reports say Anthropic’s tools have seen a jump in U.S. business adoption and the company is reportedly considering designing its own AI chips amid global compute shortages. (alltoc.com) The chip‑building idea is described as preliminary — signalling that access to dedicated silicon is moving from a cost issue to a strategic decision for model providers. (mezha.net)
Anthropic’s Claude is gaining ground with United States businesses as the company weighs a more expensive next step: designing its own artificial intelligence chips. (cnbc.com) Ramp said in its March 2026 Artificial Intelligence Index that nearly one in four businesses on its platform now pays for Claude, up from one in 25 a year earlier. The same update said OpenAI adoption on Ramp fell 1.5% over the period. (ramp.com) Anthropic said this week that its annualized revenue run rate has topped $30 billion, up from $9 billion at the end of 2025. Bloomberg reported that more than 1,000 business customers now spend over $1 million a year with Anthropic, more than double the February count. (bloomberg.com) Training and running large language models requires specialized processors that handle the matrix math behind text prediction, much like using industrial turbines instead of household fans. Reuters reported that Anthropic’s internal chip effort is still early and that the company could still decide to keep buying chips instead of designing them. (cnbc.com) Anthropic has already moved to lock in outside supply. The company said on April 6 that it expanded its partnership with Google and Broadcom, while keeping Amazon Web Services as its primary cloud provider and training partner through Project Rainier. (anthropic.com) Broadcom disclosed that expanded arrangement in a securities filing, and CNBC reported it gives Anthropic access to about 3.5 gigawatts of computing capacity based on Google’s tensor processing units starting in 2027. That is the kind of long-term power reservation usually associated with utilities, not software companies. (cnbc.com) The chip question is no longer only about price. Bloomberg reported this week that OpenAI told investors it has 1.9 gigawatts of computing capacity versus Anthropic’s 1.4 gigawatts, turning data center access into a competitive metric alongside model quality and sales. (bloomberg.com) Anthropic’s own research has shown where that demand is coming from. In a September 2025 report, the company said it was seeing enterprise use of Claude spread across business functions, offering an earlier signal that workplace adoption was shifting from experiments to regular software spending. (anthropic.com) If Anthropic keeps growing at this pace, it may end up doing both: reserving massive external capacity and building more of its own stack. For now, the clearest fact is simpler: the scramble for customers is starting to look like a scramble for electricity and silicon. (anthropic.com)
