Anthropic Plans to Scale Compute to 5 GW by Year-End and 10 GW Next Year
Anthropic has reportedly informed investors of plans to expand its available compute capacity to approximately 5 gigawatts (GW) by the end of this year and double it to 10 GW by next year. This aggressive scaling plan directly matches the projected compute expansion roadmap of rival OpenAI. The move signals that Anthropic is shifting away from its previously cautious spending stance to keep pace in the high-stakes frontier model race. It underscores how leading AI labs are competing aggressively not just on algorithm design, but on securing massive physical data center and energy capacity. As of last year, Anthropic trailed OpenAI in compute reserves with 1.5 GW compared to OpenAI's 2 GW. The planned scale-up to 5 GW and subsequently 10 GW highlights a strategic pivot by CEO Dario Amodei, who had previously criticized competitors for reckless spending on compute infrastructure.
## BACKGROUND
In frontier AI development, compute capacity is increasingly measured in gigawatts (GW), reflecting the total electrical power consumed by data centers housing hundreds of thousands of AI accelerators. A single gigawatt can power roughly 750,000 homes, making grid access and energy infrastructure primary bottlenecks for training next-generation large language models.