AMD’s fastest FP64 GPU won’t ship until 2027

AMD’s Instinct MI430X delivers 288 teraflops of hardware FP64 performance, according to the company’s MI400 press release. AMD calls it the fastest FP64 GPU ever built, but researchers and sovereign-AI customers will have to wait: first shipments are targeted for early 2027.

The MI430X was announced alongside the MI455X, AMD’s flagship accelerator for AI data centers. Both chips are built on AMD’s CDNA 5 architecture and include 432GB of HBM4 memory, but they target different workloads.

MI430X targets HPC and sovereign AI

The MI430X is tuned for high-performance computing, where FP64 precision is critical for scientific research and other workloads that prioritize numerical accuracy. AMD says its FP64 performance exceeds that of both the MI455X and Nvidia’s Rubin-based offerings.

The MI455X instead emphasizes lower-precision FP4 and FP8 performance, which is widely used for AI inference and other data-center workloads. AMD is prioritizing that market, targeting the MI455X for earlier shipments even as sovereign-AI customers line up for both products. Potential buyers named in the announcement include the U.S. Department of Energy, Oracle, and the UK government.

AMD claims the MI455X can deliver up to 34 times the token throughput of the MI355X at 18 times lower cost. The accelerator also powers Helios, AMD’s first rack-scale AI system, which contains 72 MI455X GPUs and 18 sixth-generation EPYC server CPUs in a 1:6 configuration per compute node.

Recommended reading

Kimi’s launch queue exposes AI’s next bottleneck

“The next generation of AI will span frontier AI, sovereign AI and scientific computing, and each requires infrastructure optimized for its unique demands. The AMD Instinct MI400 Series extends our AI portfolio with purpose-built solutions optimized for the full spectrum of AI and HPC deployments, built on an open software foundation that gives customers the flexibility to innovate at every scale.”

AMD’s expanding accelerator commitments

AMD is also securing demand for its Helios platform through a major agreement with Anthropic. AMD has agreed to invest $5 billion in the AI company, which will deploy 2 gigawatts of AMD’s MI450 accelerators. The deal gives AMD another high-profile customer for its data-center hardware, even as Anthropic’s AI chief recently expressed concern about a perceived performance drop in Claude Code in early 2026.

The MI430X’s delayed availability leaves AMD’s near-term AI push centered on the MI455X, while the FP64-focused accelerator is aimed at customers willing to wait for specialized HPC capacity.