Anthropic IPO filing signals risk to Nvidia's pricing power

Anthropic’s IPO filing on Monday highlighted a computing strategy that may strengthen its negotiating position against Nvidia over time by spreading AI workloads across multiple chip platforms. In its S-1, the AI company said it manages compute as a single, interchangeable pool supplied by strategic partners, with software able to direct jobs to whichever processors are most productive. Anthropic said it uses hardware from four chip families: Amazon’s Trainium, Google’s TPUs, Nvidia’s GPUs and AMD’s GPUs. The disclosure underscores how leading AI developers are reducing their dependence on a single supplier. Analysts said multichip operations could erode Nvidia’s pricing power because customers that can shift workloads across platforms have greater leverage in negotiations. The approach may also cushion Anthropic against Nvidia supply shortages. The strategy has required years of engineering and large investment, limiting its practicality to major labs with deep technical and financial resources. The near-term threat to Nvidia remains constrained. Moving training workloads between processors is harder than moving inference tasks, and Anthropic continues to spend heavily on Nvidia compute. Smaller companies may lack the capacity to justify the effort, while Nvidia’s advantages in performance, networking and software remain substantial. A wider industry concern is that AI compute could become more commoditized. If supply outstrips demand, rental prices could decline sharply, as Citadel’s Ken Griffin has warned. Nvidia did not immediately respond to a request for comment.