AMD has unveiled the full product lineup for its MI400 chip series.

The company's CEO, Dr. Lisa Su, revealed details of the chips in her keynote speech at the Consumer Electronics Show (CES), currently taking place in Las Vegas.

Su said AMD's forthcoming double-wide Helios rack – slated for release in Q3 2026 – would deliver up to three AI exaflops in a single rack and would create a “blueprint for yotta-scale compute.”

In addition to Helios, which comprises 72 Instinct MI455X-series accelerators with 31TB of HBM4 (high-bandwidth memory), the chip portfolio also includes the AMD Instinct MI440X, a rack mounted server consisting of eight GPUs and one AMD Epyc Venice CPU that targets AI training and inference workloads, and the AMD Instinct MI430X, developed for sovereign AI, HPC, and hybrid computing.

Additionally, the MI430X, MI440X, and MI455X accelerators are expected to be the first GPUs to support UALink’s new interconnect for scale-up connectivity. For scale-out connectivity, the company’s Helios platform will feature Ultra Ethernet technology, which supports AMD’s Pensando Pollara 400G and forthcoming Pensando Vulcano 800G cards.

Su also teased additional details about its forthcoming MI500 series GPUs during her speech, saying that the chips, set to be launched in 2027, would deliver up to a 1,000x increase in AI performance compared to the AMD Instinct MI300X with AMD CDNA 6 architecture.

Both the MI400 and MI500 series will be manufactured using TSMC’s 2nm process technology, although the MI500 will be produced with technology that is reportedly more advanced and offers unspecified enhancements.

In October 2025, AMD announced it was partnering with OpenAI for the supply of 6GW-worth of GPUs, with the deal expected to kick off in the second half of 2026, with OpenAI saying it would build a 1GW data center using AMD MI450 chips, starting this year.

During her speech, Su was joined on stage by OpenAI’s president and co-founder Greg Brockman. However, while Su quipped that every time she sees Brockman, “you tell me you need more compute,” no additional information about the agreement was provided. No details regarding where the proposed data center will be built or who the developer will be have been disclosed by the two companies either.

“At CES, our partners joined us to show what’s possible when the industry comes together to bring AI everywhere, for everyone,” said Su. “As AI adoption accelerates, we are entering the era of yotta-scale computing, driven by unprecedented growth in both training and inference. AMD is building the compute foundation for this next phase of AI through end-to-end technology leadership, open platforms, and deep co-innovation with partners across the ecosystem.”