According to recent reports, NVIDIA, a prominent AI chip manufacturer, has allocated a substantial $1.3 billion budget to ensure smooth shipments of its GH200 and H200 chips this year. The company has secured a portion of the new high-bandwidth memory (HBM3e) production capacity from Micron and SK hynix to support this effort.
Industry experts caution that while the credibility of this budget remains to be confirmed, NVIDIA's strategy aims to capitalize on early market opportunities in high-performance computing (HPC). By potentially gaining an early market share lead, NVIDIA aims to leverage the competitive advantage of launching its products ahead of competitors.
This year, major foundries including SK hynix, Micron, and Samsung are expected to scale their production capacity to accommodate approximately 750,000 chips, with HBM3e boasting an estimated yield rate of 90%. Global production of high-bandwidth memory (HBM) is projected to reach around 56 million chips, with significant capacity expansions anticipated in the latter half of the year.
Analyzing the advanced packaging capacity for GPUs using Chip-on-Wafer-on-Substrate (CoWoS) technology, global capabilities are forecasted to exceed 300,000 wafers by the fourth quarter. This capacity is crucial for meeting the demand projected for approximately 9 million GPU units utilizing CoWoS packaging this year.
Given the standard ratio of 1 GPU logic chip to 6 HBM storage chips, global demand for HBM is expected to exceed 54 million chips, primarily in 12-layer configurations.
Currently, the market price for 12-layer HBM chips stands at approximately $250 each. If NVIDIA's $1.3 billion budget is accurate, it would secure approximately 5.2 million chips, representing a significant investment in the high-performance memory segment.
The AI-HPC industry's growing demand for high-bandwidth memory underscores its increasing value and market potential, with prices significantly higher than traditional DRAM and DDR5 technologies.
While production capacity appears sufficient for this and the upcoming year, much of the production is scheduled to ramp up towards the end of the year. Therefore, early access to production capacity in the first half of the year is critical to ensuring timely product availability and market penetration.