The ongoing shortage in AI computing supply is rapidly reshaping the chip procurement strategies of major tech firms, with Advanced Micro Devices emerging as a key beneficiary. According to media reports citing market sources, Anthropic is planning to purchase AMD's next-generation Instinct MI450 accelerator chips to bolster its server infrastructure. Analyst Ben Bajarin noted on social media, "It makes perfect sense for AMD to secure an order from Anthropic, given the current compute shortage."
This development would position Anthropic as another leading AI company—following OpenAI and Meta—to enter into a significant partnership with AMD, further strengthening AMD's competitive standing in the AI data center market. The rumored order centers on AMD’s upcoming MI450 chip, which is built on the CDNA 5 architecture. Official data indicates the chip delivers 40 PFLOPs of FP8 computing power, doubling the performance of the previous MI350 series. Memory capacity has also increased by approximately 50%, from 288GB of HBM3e to 432GB of HBM4, while memory bandwidth sees a substantial jump from 8 TB/s to 19.6 TB/s. These upgrades offer meaningful improvements for AI companies scaling large-scale inference and training workloads.
Persistent supply constraints in the global AI chip market are driving customers to diversify their sourcing strategies. Bajarin emphasized, "All available and viable compute capacity is being purchased," a view that aligns with current market dynamics—where hyperscale cloud providers and startups alike are actively securing next-generation computing resources. AMD has already positioned itself to capitalize on this trend, having previously disclosed that multiple "OpenAI-scale" customers are queuing up for its current and future Instinct accelerators. Among them, Meta has committed to purchasing up to 6 gigawatts of Instinct chips across multiple generations. Should the Anthropic order materialize, it would further expand AMD’s roster of major AI clients and provide positive momentum for its revenue outlook.
AMD has positioned the MI450 series to compete directly with Nvidia’s Vera Rubin platform. Internal comparisons show that the MI450 offers 1.5x greater memory capacity and scale-out bandwidth versus its competitor, while matching it in FP4/FP8 compute performance, memory bandwidth, and scale-out interconnect bandwidth. Specific parameters include 40 PFLOPs of FP8 compute, 432GB of HBM4 memory, 3.6 TB/s of scale-out bandwidth, and 300 GB/s per card for node-to-node communication. AMD markets this combination as the core of a rack-scale AI ecosystem under the "Helios" brand.
Anthropic’s compute strategy reflects a multi-supplier approach to mitigate risk. The company already uses Nvidia GPUs and Amazon’s Trainium chips in its infrastructure, and recently signed strategic agreements with Google and Broadcom to access next-generation TPU compute at multi-gigawatt scale starting in 2027. In an official statement, Anthropic indicated these TPU resources will support its Claude frontier models and global customer demand. Some analysts suggest that Anthropic’s collaboration with Broadcom—which has extensive experience in custom chip design—could hint at future plans to develop proprietary chips for Claude. While still speculative, this direction underscores Anthropic’s intent to secure high-quality suppliers, strengthening AMD’s negotiating position.
Beyond commercial clients, AMD is also expanding its presence in the government sector. The company recently announced a deepened partnership with the French government to support France’s national AI strategy. The collaboration includes efforts to spur local AI innovation, expand access to open computing resources, and contribute to the development of France’s first exascale supercomputer, "Alice Recoque." AMD will open its high-performance computing platforms and open software ecosystem to the French AI community.
In summary, if the Anthropic deal is finalized, it would further validate AMD’s commercial capabilities in the high-end AI chip market and bolster market expectations ahead of the official launch of its MI400 series.
Comments