AI Memory Shortage Deepens, Goldman Sees Boom Through 2028

Goldman Sachs analysts project the AI-driven memory chip shortage will intensify through 2027 and persist into 2028, potentially marking the deepest supply crunch the industry has ever seen. The bank forecasts DRAM prices to surge over 300% year-over-year in 2026, with NAND prices rising more than 250%, as demand for AI infrastructure massively outpaces supply.

Three key structural changes underpin this forecast. First, demand is now anchored by the massive and growing AI server market, estimated at $449 billion—over seven times larger than during the last major upcycle. Second, supply growth is constrained as advanced High-Bandwidth Memory (HBM) production consumes 3-4x more wafer capacity than traditional DRAM. Third, customers are locking in supply with long-term contracts, providing memory makers with unprecedented earnings visibility and stability.

This prolonged shortage fundamentally reshapes the competitive landscape. The clear winners are memory manufacturers with leading HBM technology and scale, like SK Hynix and Samsung, who can command premium pricing in a tight market. Goldman's massive price target increases for these firms reflect expectations for explosive profit growth.

The shift matters because it breaks the memory sector's notorious volatility. If profits remain elevated for years, the historical discount applied to these stocks could vanish, leading to significant valuation multiple expansion. Companies reliant on buying memory, such as certain PC makers or cloud providers without locked-in contracts, face rising costs and potential supply constraints, putting pressure on their margins. The era of memory as a cheap, plentiful commodity is over, replaced by a strategic, capacity-constrained component essential for AI.