Big Tech’s AI Chip Push Doesn’t Faze NVIDIA, But Opens Door for Samsung & Micron
SAN FRANCISCO – Despite a surge in self-reliant AI chip progress from tech giants like Meta, Microsoft, Google, and Amazon, NVIDIA remains confident in its market position, while the escalating demand for High Bandwidth Memory (HBM) is poised to benefit both SK Hynix – currently the market leader – and perhaps Samsung Electronics and micron. A new report from S&P Global Ratings forecasts continued HBM growth fueled by increasing demand for AI GPUs and the adoption of Application-specific Integrated Circuits (ASICs) by major technology companies.
The move toward in-house AI accelerators isn’t intended to supplant NVIDIA, but rather to complement its offerings, particularly for AI inference tasks. While NVIDIA CEO Hwang has asserted that asics won’t replace his company’s accelerators,S&P analysts believe a coexistence is more likely. This dynamic is creating a broader, more robust market for HBM, the high-performance memory crucial for both GPUs and ASICs. The semiconductor industry anticipates sustained HBM market expansion through at least 2027.
Meta plans to begin mass production of its 4th generation ‘MTIA’ AI accelerator next year, alongside Microsoft’s anticipated rollout of ‘Maia 200.’ google’s 7th generation ‘TPU’ will feature six layers of 12-layer HBM4, and Amazon’s ‘Trenium’ 3rd generation will incorporate four layers of 12-speed HBM3Es.
“Major AI companies are expected to actively utilize asics in fields that do not necessarily require advanced GPUs, such as learning or inference of lightweight models,” predicts Kim Woong, a senior researcher at Nice Credit Rating. “With the increase in users and traffic, the ASIC market will also gradually demand higher versions of HBM.”
Currently, SK Hynix dominates the HBM market. However, the growing prominence of ASICs could create new opportunities for Samsung Electronics and Micron to gain market share in the specialized memory sector.
