With the rise in adoption of generative artificial intelligence (genAI), the infrastructure supporting this growth is facing a supply and demand bottleneck.
Research from IDC shows that 66% of enterprises globally are planning to invest in genAI in the next 18 months, with a significant portion of IT spending in 2024 allocated to infrastructure. However, a crucial hardware component required for AI infrastructure is currently facing shortages.
The rapid pace of AI adoption has strained the supply of high-performance chips necessary for genAI operations. While there has been much focus on the demand for Nvidia GPUs and alternatives, the surge in demand for high-bandwidth memory chips from SK Hynix has been overlooked.
SK Hynix recently announced that their high-bandwidth memory (HBM) products, essential for AI processing alongside high-performance GPUs, are fully booked through 2025 due to high demand. This has led to a price increase of 5% to 10% for HBMs, driven by premiums and capacity needs for AI chips.
SK Hynix’s HBM3 product offers the industry’s largest 24GB memory capacity, achieved through the stacking of 12 DRAM chips, providing high-capacity and high-performance capabilities.
SK Hynix
According to TrendForce, HBM chips are projected to contribute over 20% of the total DRAM market value by 2024, potentially exceeding 30% by 2025. This surge in demand has led buyers to accept higher prices to secure stable and quality supplies, as not all major suppliers have met customer qualifications for high-performance HBM.
2024-05-08 14:51:02
Original from www.computerworld.com