Samsung locks up AI chip supply for five years as memory shortage looms

Samsung has pre-allocated 60-70% of its future memory capacity into five-year supply agreements with major global data-center operators, part of roughly $950 billion in aggregate deals across Korean chipmakers. The unprecedented shift locks in predictable, long-horizon demand and is explicitly designed to weather surging AI-driven memory needs that Samsung and analysts expect to trigger shortages through 2027-2028. Samsung's foundry business projects its AI-related revenue share will exceed 30% this year.

The mechanics matter for the whole industry: by committing the majority of forward capacity to a handful of hyperscale buyers on multi-year terms, Samsung effectively removes that supply from the spot market, meaning smaller buyers and the broader electronics sector will face tighter availability and higher prices. It's a structural bet that AI compute demand is durable rather than a bubble — and a hedge against the boom-bust cycles that have historically whipsawed memory pricing.

The move lands amid a broader hardware-scarcity narrative: NVIDIA is reportedly set to raise GeForce RTX prices by up to 30%, and community threads lament GPU prices 'nearing $5,000' and 'new car prices' for next-gen cards. Memory pre-allocation is the upstream version of the same squeeze.

Skeptics will note that locking capacity at today's demand assumptions carries risk if AI capex cools — Samsung would be committed to volumes at prices that could look unfavorable in a downturn. But with hyperscaler backlogs like AWS's $496B suggesting sustained demand, the near-term read is a tightening memory market. Watch DRAM/HBM spot prices and whether competitors like SK Hynix and Micron follow with similar lock-ups.