Anthropic to Design Custom Hardware for Claude

Anthropic and OpenAI, two of the leading developers of generative‑AI models, are accelerating efforts to expand their compute capacity while lessening reliance on Nvidia’s graphics processing units (GPUs). Both firms have announced initiatives to diversify their hardware stacks, citing concerns over supply bottlenecks, pricing volatility, and the strategic risk of depending on a single supplier for the massive workloads required by large language models.

OpenAI has begun testing custom accelerator chips designed in partnership with Microsoft’s Azure cloud platform, aiming to run future model training and inference workloads on silicon optimized for its architecture. Anthropic, meanwhile, is evaluating alternative processors from AMD and exploring its own ASIC designs, with a pilot program that integrates these chips into its existing cloud infrastructure. The moves come as Nvidia reports sustained demand but also faces capacity constraints that have driven up prices for high‑end GPUs, prompting AI firms to seek cost‑effective, scalable alternatives. Industry analysts note that broader hardware competition could accelerate performance improvements and reduce overall operating expenses for AI providers.

If the diversification strategies succeed, both companies could achieve greater control over compute costs and mitigate supply‑chain risks, potentially reshaping the AI hardware market. Continued investment in non‑Nvidia solutions may also pressure Nvidia to adjust pricing and expand its product offerings, influencing the competitive dynamics of the rapidly growing generative‑AI sector.