• AMD CEO Lisa Su highlighted tightening supply for key commodities like memory at a UBS investor conference, confirming a severe industry-wide shortage.
  • The crunch, driven by a massive reallocation of production to high-bandwidth memory (HBM) for AI chips, has caused DRAM and NAND flash contract prices to double in recent months.
  • Analysts and industry executives see no quick relief, with shortages and elevated prices projected to persist into late 2027 or beyond, impacting PC makers, consumers, and enterprise systems.

At a UBS conference this week, Advanced Micro Devices Inc. CEO Lisa Su pointed to a growing supply constraint that is sending shockwaves through the global tech sector: memory is getting tight. Her comments, delivered to an audience of investors, align with an escalating crisis in the semiconductor supply chain where commodities like DRAM and NAND flash have become scarce and increasingly expensive.

The root of the problem is a fundamental structural shift. Major memory producers—Micron, Samsung, and SK Hynix—have aggressively pivoted production capacity away from standard commodity memory to manufacture high-bandwidth memory (HBM), the specialized, high-margin chips critical for powering AI accelerators from companies like AMD and Nvidia. This reallocation has created a severe deficit for the conventional memory used in everything from personal computers and servers to consumer electronics.

Market data reveals the stark reality. Contract prices for DRAM and NAND have surged, with some reports indicating month-on-month increases of 80% to 100% in late 2025. In just the past six months, NAND flash prices have more than doubled, while DDR4 memory is facing projected quarterly price hikes of up to 43%. The effects are already visible to consumers; the cost of some 64GB DDR5 RAM kits now rivals that of a premium gaming console.

"Availability is expected to worsen significantly in early 2026 once distribution stockpiles are exhausted," a senior executive at a major memory module maker recently warned, suggesting that securing allocation may soon become a challenge regardless of price. The safety stock buffers that normally cushion the supply chain have been depleted, according to industry analysts, due to a combination of tariff uncertainty, previous price declines that discouraged inventory building, and reduced secondary manufacturing.

The demand pull from artificial intelligence is staggering. Large-scale AI data center projects, including initiatives like OpenAI's reported Stargate, are consuming vast portions of global DRAM output. Cloud giants such as AWS, Google, and Microsoft are effectively locking up supply years in advance, creating a two-tier market where AI infrastructure takes absolute priority.

For the broader tech industry, the implications are immediate. Small to medium-sized PC manufacturers, which lack large inventory buffers, are under acute pressure. Analysts note that SSDs will likely follow DRAM with similar price increases, and even GPU supplies, particularly for RADEON cards, are experiencing constraints. While companies like Nvidia maintain larger memory inventories for their board partners, the overall cost structure for building any system with substantial memory is rising rapidly.

When asked for comment on the supply outlook following the CEO's remarks, an AMD spokesperson referred back to the public conference discussion. A memory industry source, speaking on condition of anonymity due to lack of authorization, stated bluntly that the next 6-9 months are "already screwed" in terms of pricing and availability recovery.

The timeline for any normalization appears distant. Building new semiconductor fabrication capacity is a multi-year endeavor. Industry projections suggest that even if new production decisions were made today, meaningful additional supply would not hit the market until late 2028 at the earliest, with full ramping in 2029. Most forecasts now point to sustained shortages and elevated costs through 2026 and 2027, a sobering outlook for businesses planning their technology budgets.

Correction: An earlier version of this article misstated the projected timeline for price normalization. Industry projections suggest late 2027 or later, not mid-2027.