GitHub Code Leak Reveals: Anthropic Reportedly Testing AMD AI Chips to Diversify Compute Supply and Reduce Reliance on Nvidia

AI startup Anthropic is reportedly expanding its AI chip supply chain. Semiconductor research firm SemiAnalysis claims that by analyzing code publicly posted on GitHub by a senior director of AMD's AI business, it discovered references indicating that Anthropic could become a new customer of AMD. If confirmed, this would mark AMD's first entry into Anthropic's compute infrastructure, signaling that major AI model developers are accelerating efforts to build diversified chip supply strategies.

SemiAnalysis announced on Monday via the social platform X that its research team, after analyzing the GitHub code published by the AMD executive, found project-related content involving Anthropic. The firm inferred that Anthropic has begun testing or integrating AMD's AI GPUs and plans to release further technical details and code analysis. However, neither AMD nor Anthropic has officially confirmed the report as of now.

The disclosure of this information is highly unusual—not stemming from an official announcement or media report, but from data uncovered in a public code repository. Given that GitHub code is publicly searchable and the poster is a senior AMD AI executive, the revelation has drawn significant market attention, highlighting how public code can inadvertently leak commercial partnership information.

If Anthropic formally adopts AMD GPUs, it would represent a major breakthrough for AMD in the high-end AI training market. Currently, Anthropic's primary compute resources come from Google's parent company, Alphabet (GOOGL-US), via its TPU offerings, as well as Nvidia (NVDA-US) GPUs. Samsung Electronics also serves as a partial hardware supplier. With AMD's inclusion, Anthropic would further diversify its compute infrastructure, reducing reliance on any single supplier.

Anthropic has received substantial investment from Alphabet in recent years and has long relied on Google Cloud TPUs to train its Claude series of large language models, while also heavily utilizing Nvidia GPUs to build its AI infrastructure. Industry observers suggest that adopting AMD's MI-series AI accelerators would enable Anthropic to operate a hybrid architecture combining GPUs and TPUs, achieving better balance across supply stability, cost control, and technical flexibility.

For AMD, securing Anthropic as a customer would be highly symbolic. The company has been aggressively targeting the data center AI market, with its MI300 and next-generation MI350 series GPUs positioned as key challengers to Nvidia's Hopper and Blackwell platforms. AMD's recently published AI chip roadmap indicates continued launches of next-gen AI GPUs over the coming years, aiming to expand data center market share and intensify direct competition with Nvidia.

Market analysts note that nearly all major global AI model developers are currently re-evaluating their chip supply strategies. Beyond Anthropic, OpenAI is expanding procurement across multiple cloud providers, Alphabet is advancing its TPU platform, and Meta Platforms (META-US) and Microsoft (MSFT-US) are investing in proprietary AI chips and diversified supply chains to reduce dependency on single vendors.

Meanwhile, Intel (INTC-US) is actively promoting its Gaudi AI accelerators to capture more enterprise AI customers. With global demand for AI model training and inference growing rapidly, it is widely expected that AI companies will increasingly adopt a mix of Nvidia GPUs, in-house chips, and platforms from AMD and others to mitigate supply chain risks and strengthen bargaining power.

Analysts believe that reports of Anthropic testing AMD GPUs not only suggest AMD could gain a high-profile AI customer but also reflect a broader industry shift—from heavy reliance on a single chip supplier toward a new era of diversified compute architectures. If the collaboration is officially confirmed, it could help AMD scale its data center AI business and further intensify competition in the AI chip market.