NVIDIA Boosts Memory Commitments to $279B, Signaling Pricing Power for HBM Makers

NVIDIA's latest earnings report has signaled a significant shift in the AI hardware supply chain, revealing that access to High-Bandwidth Memory (HBM) has become a primary constraint on growth. The chip designer disclosed a massive increase in its long-term memory purchase commitments and warned customers of upcoming price hikes for AI servers, underscoring the growing pricing power of key memory manufacturers.

The $279 Billion Signal

In its Q2 FY2027 earnings report on August 26, NVIDIA announced its multi-year memory purchase commitments had surged by approximately 135% quarter-over-quarter to $279 billion. The company explicitly stated this increase was driven "primarily for memory," a move that highlights an aggressive strategy to secure the critical components needed for its AI accelerators.

At the same time, NVIDIA has reportedly warned customers of AI server price increases exceeding 15%, directly attributing the rise to escalating memory costs. This indicates the company is passing the higher input costs downstream to end users, effectively validating the price inflation from HBM suppliers.

Market Impact and Margin Shift

The financial implications for NVIDIA and its suppliers are becoming clear. NVIDIA's own gross margin guidance suggests a near-term compression, moving from a peak of 75% in the reported quarter to a projected 71–72% by the fourth quarter. This anticipated margin decline reflects a transfer of profitability to HBM producers, who are capitalizing on the supply-demand imbalance.

Ad

Investors responded positively to the news for memory makers. Following the report, on August 27:

• SK Hynix (000660.KS), the leading HBM supplier, saw its shares rise +3.55%.
• Samsung Electronics (005930.KS), which is ramping up its HBM3E production, gained +2.49%.

The HBM Bottleneck

The core issue is a bottleneck in HBM supply, which is now seen as a more significant limiting factor than compute capacity for AI infrastructure expansion. At the Jefferies Semiconductor Conference on August 27, sources indicated NVIDIA is evaluating different HBM configurations for future chips to maximize the number of accelerators it can ship using the limited pool of available memory dies.

However, the tailwind is not lifting the entire semiconductor sector evenly. Test equipment manufacturers like Advantest saw their stock decline, suggesting the constraint lies with memory die availability rather than a need for new manufacturing capacity. This indicates that while memory makers' margins are improving, the benefits may not fully trickle down to the broader equipment supply chain.