AMD launched its Instinct MI400 Series GPUs at its Advancing AI 2026 event on July 23, 2026, according to the company’s announcement. The family splits into two parts aimed at different buyers: the Instinct MI455X for “frontier AI and AI factory deployments,” and the Instinct MI430X for sovereign AI and high-performance computing (HPC). AMD says the line-up uses HBM4 memory.
On this page
The two MI400 Series GPUs
| GPU | Target workloads (per AMD) | Figures AMD published |
|---|---|---|
| Instinct MI455X | High-volume inference, frontier-model training and fine-tuning | HBM4; powers the AMD Helios rack |
| Instinct MI430X | Sovereign AI and HPC, converged AI and scientific workflows | HBM4; up to 288 TFLOPS of hardware-based FP64 |
AMD says the MI455X delivers “the compute, memory and networking performance required for high-volume inference, frontier-model training and fine-tuning.” The MI455X is the GPU inside the AMD Helios rackscale solution, which the company positions for frontier AI.
The MI430X is the double-precision part. AMD claims it delivers “uncompromised accuracy and throughput across converged AI and HPC workflows with up to 288 TFLOPS of hardware-based FP64 performance,” the kind of math scientific simulation codes depend on.
AMD also lists security features across the series, including “secure boot, encrypted GPU-to-GPU links and hardware-based protections,” and says the MI400 Series runs on its open ROCm software stack with “programming models, compilers, libraries, runtimes and deployment tools.”
What AMD said in its second-quarter results
AMD returned to the launch in its second-quarter 2026 results on August 4, 2026. The company reported revenue of $11.5 billion and Data Center segment revenue of $6.7 billion, up 107% year over year, which it attributed to demand for EPYC processors and Instinct GPUs. The release describes the MI400 Series as “bringing leadership compute, memory capacity and bandwidth for large-scale AI training and inference.”
In the same release AMD said Helios “is being deployed by leading AI labs and cloud providers including Anthropic, Cirrascale, HUMAIN, Meta, Microsoft, OpenAI, Oracle, Tensorwave, Vultr and others.” It also listed a strategic partnership with Anthropic “to deploy up to 2 gigawatts of MI450 Series GPUs in AMD Helios racks,” and the launch of the Instinct MI350P, a GPU “designed to bring seamless AI acceleration to existing infrastructure.”
What it changes
The MI400 Series gives AMD separate parts for rack-scale AI (MI455X in Helios) and FP64-heavy HPC (MI430X), rather than one GPU for both workloads. Helios, built on the MI455X, is the configuration AMD names alongside its lab and cloud deployments. Specs for every Instinct generation are in our AI accelerator catalog.
