Market research firm TrendForce announced on September 7, 2026, that DRAM industry revenue reached $154.73 billion in the second quarter, up 59.5% from the previous quarter. Yet bit shipments—the measure of memory capacity actually shipped—grew only modestly. Supply allocation prioritizing AI servers, combined with rising contract prices, inflated revenue faster than volume. Conventional DRAM contract prices are expected to slow to a 13-18% increase in the third quarter, but reading this deceleration as a sign of supply recovery would misread what's actually happening in the market.
The Gap Between 59.5% Revenue Growth and Bit Shipments
TrendForce's estimated second-quarter industry revenue of $154.73 billion marked a sharp jump from the first quarter's $97 billion. The company attributes this primarily to rising contract prices for conventional DRAM, along with growth in shipments of HBM3E, LPDDR5X, and high-capacity server memory modules (RDIMMs) driven by AI server demand. Meanwhile, suppliers' inventories sit at historic lows, and most additional supply has gone toward server applications. Bit shipment growth was modest.
Revenue and bit shipments are not the same metric. Revenue is determined by a combination of shipped capacity, product mix, and average selling price (ASP). If the share of high-priced server DRAM and HBM rises, and conventional DRAM contract prices also increase, revenue can grow substantially even without a major increase in shipped capacity. TrendForce hasn't disclosed the contribution of each factor or revenue by product category, so it's impossible to determine how many percentage points of the 59.5% growth came from price increases. Still, it's clear that "the market grew by roughly 60%" and "production increased by roughly 60%" are two different things.
The same phenomenon occurred in the first quarter. Conventional DRAM contract prices spiked sharply, pushing industry revenue up 81% to $97 billion, while TrendForce attributed the bulk of supply growth to migration toward advanced process nodes, noting that building new cleanrooms takes time. The revenue surge in the second quarter is not evidence that supply capacity expanded at the same pace.
Top Three Players' Combined Share Slips Slightly; Race for Second Place Tightens
Calculating from the same TrendForce revenue data, the combined share of the top three companies fell 2.1 percentage points, from 89.7% in the first quarter to 87.6% in the second quarter, while the gap between SK hynix and Micron narrowed from 6.4 points to 1.6 points.
| Company | Q1 2026 Revenue Share | Q2 2026 Revenue Share | Q2 Revenue | Q2 QoQ Change |
|---|---|---|---|---|
| Samsung Electronics | 38.5% | 39.4% | $60.98 billion | +63.4% |
| SK hynix | 28.8% | 24.9% | $38.59 billion | +37.9% |
| Micron | 22.4% | 23.3% | $36.0 billion | +65.5% |
Treating the top three companies as a monolithic bloc obscures this shift. According to TrendForce, Samsung recorded the largest bit shipment growth among the three, with early HBM4 mass production and shipment, along with a significant rise in average selling price, boosting revenue. Micron, facing supply constraints, prioritized higher-priced server DRAM. SK hynix, by contrast, had the highest proportion of HBM in its bit shipment mix among the three companies, resulting in relatively limited ASP growth. This doesn't mean SK hynix's shipments or earnings declined—its DRAM revenue also grew 37.9% quarter-over-quarter.
The 2.1-point drop in combined share also reflects revenue growth among companies focused on mature process nodes. Nanya's second-quarter DRAM revenue rose 68.3% quarter-over-quarter to $2.612 billion, while Winbond's grew 75.8% to $998 million. PSMC's DRAM revenue, though small in absolute terms at $115 million, jumped 167.8%. As the major three companies shift production toward advanced nodes and server applications, demand for products with no migration path—such as DDR4 and DDR3—flows toward Taiwanese manufacturers. While 87.6% still represents strong market concentration, price increases have spread beyond the top players as well.
Prices Slow to 13-18%, But Supply Shortage Persists
TrendForce forecasts conventional DRAM contract prices will rise 13-18% quarter-over-quarter in the third quarter. The growth rate is slowing, but price levels themselves are still expected to climb further—and this remains a forecast, not confirmed transaction data. The deceleration factors TrendForce cites are PC and smartphone customers finding it harder to absorb additional price hikes, and some demand shifting from high-capacity modules to lower-capacity products. Recovery in supplier inventory is not among the reasons given.
For server DRAM, the forecast range is the same 13-18%, but the impact of price increases varies by customer. TrendForce expects that U.S. cloud providers with multi-year long-term agreements (LTAs) will benefit from price caps, while customers without such contracts—or those purchasing beyond their contracted volumes—will bear the brunt of price increases. The volumes, pricing formulas, and caps specified in individual contracts are not publicly disclosed. Average price increase figures alone cannot predict what any individual company will actually pay.
The shift toward lower-capacity modules also doesn't equate to a disappearance of demand. Some cloud providers and server manufacturers are shifting their configurations from 96GB/128GB modules to 32GB/64GB modules. This reflects both DRAM inventory buildup from delayed server assembly due to CPU shortages, and efforts to control procurement costs. However, TrendForce hasn't provided data on total memory capacity per server, completed unit counts, or performance impact. The fact that lower-capacity module shipments are increasing as a share of the total doesn't support a conclusion that overall server DRAM demand has declined.
In fact, TrendForce's preliminary estimates suggest that server memory bit supply growth in 2027 will be limited to 15-20% year-over-year, falling short of server CPU shipment growth. While the specific CPU growth rate isn't disclosed, if CPU supply improves in the second half of 2026, servers that had been waiting on CPU availability will require DRAM for assembly. The third-quarter slowdown looks less like a phase where supply shortages are resolving and more like a phase where buyers are adjusting capacity and contract terms to cope with high prices.
Migration to Advanced Nodes—Not New Fabs—Drives Supply Growth
As the primary means by which the top three companies will increase bit supply in 2026-2027, TrendForce points to migration toward advanced process nodes. While node shrinkage increases the memory capacity obtainable from a single wafer, wafer input volume is expected to see only modest increases—driven by manufacturing process improvements, cleanroom acquisitions, and conversions from other product lines. This approach can boost supply of high-value-added products, but it won't simultaneously resolve shortages across all product categories, including mature-node products.
Each company's disclosures reflect this time lag and prioritization of supply destinations. Samsung reported in its second-quarter earnings that its memory business achieved record revenue and operating profit, driven by expanded HBM4 sales. Even so, the company expects demand centered on server DRAM, enterprise SSDs, and HBM to keep growing in the second half of 2026, with supply constraints persisting. SK hynix has signed long-term supply agreements with roughly 10 customers and began mass shipment of HBM4 in the second quarter. The company is accelerating M15X mass production and plans to open the first phase of its Yongin cleanroom in early 2027, though investment will be executed in stages based on customer demand and efficiency.
Micron also disclosed multi-year strategic customer agreements and HBM4 mass shipments in its fiscal third-quarter 2026 earnings. However, the company's fiscal quarter ended May 28, differing in both timing and scope from TrendForce's calendar-quarter aggregation. Individual companies' consolidated earnings cannot simply be added back into market revenue estimates. SK hynix and Micron are both locking in demand through long-term contracts. What all three companies share is a strategy of prioritizing growing supply toward HBM and server applications.
Even if third-quarter contract prices land within the forecast 13-18% range, that alone cannot be taken as evidence that supply and demand have normalized. What needs verification is actual transacted contract prices, growth in wafer input volume versus bit shipments, and total procurement capacity that continues accumulating even after the shift to lower-capacity modules. Only when all these factors align can we determine whether the force driving revenue growth has genuinely shifted from price to volume.