NVIDIA says it anticipated memory surge and pre-ordered supplies

Lead

NVIDIA (英伟达) says it foresaw a sharp jump in memory prices and moved early to secure supply, a move that it credits with insulating the company from the current shortages roiling the market. It has been reported that CFO Colette Kress told interviewer Tae Kim that NVIDIA anticipated rising memory costs and placed advance orders rather than buying spot inventory.

Why memory is tight

AI accelerators demand both high-bandwidth memory (HBM) and large volumes of DDR/LPDDR — and both types compete for similar production equipment. Reportedly, NVIDIA’s Rubin AI platform alone will require more LPDDR by 2027 than Apple and Samsung combined, a figure cited in the report that underlines how quickly demand patterns have shifted. The squeeze has produced sharp price moves, big bonuses for SK Hynix (SK海力士) staff and worker protests at Samsung (三星).

NVIDIA’s approach

Kress said NVIDIA did not simply buy existing memory on the spot market. Instead, it contracted with multiple suppliers to produce chips to order, coordinating custom development and committed volumes across three major memory vendors, it has been reported. “Suppliers worked with us on co‑development and then calculated supply volumes,” she said, according to the interview — a procurement playbook many rivals did not apparently adopt.

Broader implications

These supply shifts matter far beyond corporate P&Ls. They reshape memory makers’ capacity plans, pushing fabs to favor HBM and shrinking DDR allocations — which in turn feeds price volatility. And the changes come amid wider geopolitical pressures on the semiconductor supply chain, including U.S.-China export controls and trade frictions that complicate sourcing and investment choices. Who plans early, and who reacts late? The answer looks increasingly like a competitive advantage in the AI race.