Goldman Sachs has a message for anyone still treating memory chips as a boom-to-bust commodity. This time, the boom has years left to run.

In a Sunday note, analysts Giuni Lee and James Schneider indicated that the artificial-intelligence-driven squeeze in memory chips will run tighter in 2027 than in 2026 and stay tight into 2028. The firm described the stretch ahead as the deepest memory shortage on record.

The call underpins a bullish outlook for memory manufacturers including SanDisk Corp. (SNDK), Micron Technology Inc. (MU), SK Hynix and Samsung Electronics, which Goldman believes could enjoy years of unusually strong pricing power and profitability.

Why Goldman Thinks This Memory Cycle Is Different

Memory has historically been one of the semiconductor industry's most volatile businesses. Periods of soaring profits have often been followed by brutal downturns as manufacturers rushed to build capacity, flooding the market with excess supply.

Goldman indicates that three structural changes are making a repeat of that pattern less likely this time.

The first is demand. Unlike previous memory booms driven by smartphones or cloud computing, the current cycle is increasingly tied to artificial-intelligence infrastructure. The bank estimates the server memory market has grown to roughly $449 billion, more than seven times larger than during the 2017-2018 memory upcycle. Servers now account for roughly half of global DRAM demand and about 40% of NAND demand.

The second is supply. The rapid adoption of High-Bandwidth Memory, or HBM, is consuming wafer capacity that would otherwise be used for conventional memory production. Goldman estimates HBM requires three to four times more wafer capacity than traditional DRAM, limiting the industry's ability to rapidly increase supply. The bank expects conventional DRAM capacity growth through 2030 to remain materially below the pace seen during prior cycles.

That imbalance is producing extraordinary pricing forecasts. Goldman now expects average DRAM prices to rise more than 300% year-over-year in 2026, while NAND prices are projected to climb more than 250%. HBM pricing, meanwhile, is expected to increase another 44% in 2027 as AI demand continues to outstrip supply. The bank also lifted its forecast for the HBM market to $116 billion in 2027, up from a prior estimate of $75 billion, and now expects it to reach $168 billion by 2028.

The third change is contractual. Goldman says customers are increasingly signing long-term agreements that include prepayments, volume commitments and penalty clauses. Those arrangements improve earnings visibility and reduce the extreme swings that have historically defined the memory industry.

The Tightest Memory Market On Record?

At the heart of Goldman's thesis is a simple conclusion: demand growth continues to outpace supply. The bank now forecasts DRAM undersupply of 5.0% in 2026 and 5.9% in 2027. NAND shortages are expected at 4.4% and 4.6%, respectively, while HBM could remain even tighter. Goldman describes the coming years as potentially producing the most severe memory shortage the industry has ever experienced.

“We now expect 2027 S/D for conventional DRAM, NAND, and HBM to be all tighter than 2026 and the tightness to extend into 2028,” analysts wrote.

Why Memory Stocks Still Look Cheap

Despite the industry's improving fundamentals, Goldman argues that investors continue to value memory companies as if another bust is around the corner. The bank notes that many AI-related companies with durable growth profiles trade near 20 times earnings, while memory stocks still command valuations closer to single-digit or low-double-digit multiples because of their history of earnings volatility.

If the market becomes convinced that profits can remain elevated for years rather than quarters, Goldman sees substantial room for multiple expansion. The firm raised its 12-month price target on Samsung Electronics by 50%, lifted its target on SK Hynix by 94% and upgraded Kioxia to Buy.

The bullish view extends to U.S.-listed names. Goldman reiterated its Buy rating on SanDisk, setting a $1,200 target price, and continues to use Micron's historical upcycle valuation multiples as a benchmark for assessing peers.

Perhaps nowhere is Goldman's conviction clearer than in its Samsung forecasts. The bank now expects the company to generate operating profit of roughly W374 trillion in 2026, more than eight times last year's level, as it becomes one of the biggest beneficiaries of what Goldman increasingly sees as a multi-year AI memory supercycle.