SK hynix Powers Microsoft Maia 200 with HBM3E

According to Korean media reports, SK hynix is set to provide high-bandwidth memory exclusively for Microsoft's latest AI accelerator, the Maia 200. The chip will pair Microsoft's in-house AI silicon with SK hynix's newest HBM3E memory, marking a deep technical collaboration focused on next-generation AI workloads.

The Maia 200 is manufactured on TSMC's advanced 3nm process and is designed to significantly improve AI inference efficiency. Each accelerator integrates 216GB of HBM3E, built from six 12-high HBM3E stacks developed by SK hynix. Microsoft has already deployed Maia 200 in its Iowa data centers and is expanding rollout to facilities in Arizona, signaling broader production use rather than limited testing.

Industry insiders note that demand for HBM is no longer driven solely by NVIDIA. A growing wave of self-developed AI chips—aimed at reducing reliance on external GPU platforms—is creating fresh momentum for the HBM market. These efforts include Google's seventh-generation TPU, Ironwood, and Amazon's third-generation Trainium, alongside Microsoft's Maia series.

UBS analysts point out that SK hynix appears well positioned in serving custom ASIC-based AI designs, with strong ties to major technology players such as Google, Broadcom, and Amazon Web Services. Looking ahead, SK hynix has already established an HBM4 mass-production system starting in September 2025, while its next-generation HBM optimization project with NVIDIA has entered final certification, paving the way for full-scale production in the near future.