Memory prices have already climbed sharply. They will keep rising. Industry forecasts point to another 40-50% jump in the third quarter of 2026 alone, followed by 30-40% more in the fourth. Digital Trends laid out the grim numbers from investment bank Jefferies in late June. Relief stays distant. New wafer capacity arrives only in 2028. Even then demand from data centers may swallow most of it.
But the story runs deeper than one bank’s projection. AI accelerators devour specialized high-bandwidth memory. That shift starves the rest of the electronics world of standard DRAM and NAND. Suppliers chase higher margins. Consumers, PC makers, automakers and smartphone brands absorb the pain. And the imbalance looks structural, not cyclical.
Supply Reallocation That Leaves Little for Everyone Else
Hyperscalers and GPU makers secured long-term contracts early. They now control half of total memory output. That share could climb to 70%. Digital Trends reported the lock-in leaves downstream buyers fighting over scraps. Manufacturers pivot production lines toward HBM stacks that power Nvidia and AMD accelerators. Each wafer sent to those stacks means one fewer for laptop SODIMMs or phone LPDDR chips.
Francisco Jeronimo, vice president at IDC, captured the trade-off neatly. “Every wafer allocated to an HBM stack for an Nvidia GPU is a wafer denied to the LPDDR5X module of a mid-range smartphone.” The research firm expects DRAM supply to grow just 16% year-over-year in 2026 and NAND 17%. Both figures sit well below historical averages. IDC published the revised outlook in December 2025. It has only grown more pessimistic since.
Prices reflect the squeeze. DRAM contract quotes rose 80-90% in one recent quarter, according to Counterpoint Research data cited across multiple outlets. Conventional DRAM now sometimes delivers supplier margins that rival or exceed those of HBM. So capacity keeps tilting. IEEE Spectrum examined the dynamic in detail this year. The piece notes HBM can consume three to four times more wafer area per bit than standard DRAM once advanced packaging enters the equation.
Intel CEO Lip-Bu Tan delivered a blunt assessment at a Cisco event. “There’s no relief until 2028.” Micron CEO Sanjay Mehrotra offered a similar timeline. He expects the HBM market to expand from $35 billion in 2025 to $100 billion by 2028, larger than the entire DRAM market was in 2024. Demand will outstrip supply “substantially for the foreseeable future,” he said. IEEE Spectrum quoted both executives directly.
New fabs will not solve the problem quickly. Samsung plans volume at its Pyeongtaek facility in 2028. SK Hynix targets output from its Cheongju site in 2027 and a U.S. plant in West Lafayette, Indiana, by the end of 2028. Micron schedules HBM production in Singapore and Taiwan for 2027 but pushes its New York DRAM investment to 2030. Even optimistic yield improvements and incremental expansions cannot close the gap before late in the decade. IEEE Spectrum mapped every announced timeline.
TrendForce tracked the revenue consequences. The firm projected the overall memory market would hit $551.6 billion in 2026 and surge again to $842.7 billion in 2027. DRAM revenue alone could reach $404 billion that year. Suppliers hold pricing power while tightness persists. TrendForce released the updated forecast in January 2026. Later updates pushed first-quarter 2026 contract price expectations even higher, to 90-95% sequential gains for conventional DRAM.
So PC and smartphone makers face higher bills. Gartner warned that combined DRAM and SSD costs could climb 130% by the end of 2026. Average PC prices would rise 17%. Smartphones would cost 13% more. On a $1,000 handset that adds $130. The firm also flagged the possible disappearance of entry-level PCs priced under $500 by 2028. Margins simply would not work. The same Digital Trends article carried those Gartner details.
IDC modeled comparable pressure. In a moderate scenario the global smartphone market contracts 2.9% in 2026 with average selling prices up 3-5%. A pessimistic case shows a 5.2% drop in volume and 6-8% ASP increase. PCs fare worse: 4.9% contraction or 8.9% in the downside case. Major vendors including Lenovo, Dell and HP have already signaled 15-20% price hikes to customers. The research house issued its analysis in December 2025. Subsequent reports from Bloomberg and Reuters in February and March 2026 revised shipment forecasts downward again, calling the expected smartphone decline the largest in more than a decade.
Automakers feel it too. Tesla and others have warned of production constraints in 2026. Gaming consoles, routers, even industrial sensors compete for the same constrained bits. Z2Data highlighted the risk of line shutdowns as early as March. Bloomberg noted Nintendo Switch 2 storage woes in a March 2026 article. The memory shortage does not stop at data centers.
Yet some suppliers see opportunity. China’s CXMT forecast first-half 2026 revenue between 110 billion and 120 billion yuan as prices lift its output. The company expanded capacity and improved its mix while global DRAM stayed tight. Reuters reported the prospectus update in May. SK Group chairman Chey Tae-won told reporters in March that wafer shortages could linger until 2030. He hinted SK Hynix would pursue strategies to moderate DRAM pricing swings. Reuters covered the comments.
SemiAnalysis called the current environment a “once-in-four-decades shortage.” Dylan Patel and colleagues argued in February that the supply-demand gap keeps worsening rather than normalizing. Their memory model shows incremental HBM growth crowds out commodity DRAM at a disproportionate rate. SemiAnalysis published the detailed forecast. Ray Wang, one of the co-authors, later discussed the dynamics on Bloomberg’s Odd Lots podcast.
TSMC moved to secure its own supply chain. The foundry partnered with Winbond to develop wafer-on-wafer 3D stacking using domestic DRAM wafers. Previously it relied on SK Hynix, Samsung and Micron. Recent social media discussion and Taiwanese media reports suggest the move aims to reduce risk amid the broader tightness. The partnership targets AI chips where memory stacked directly on logic improves bandwidth, latency and power. That development, while promising for high-end silicon, does little for the broader consumer market.
Prices will not fall quickly even after new capacity arrives. Mina Kim of Mkecon Insights told IEEE Spectrum that memory prices historically come down much more slowly than they rise. Thomas Coughlin of Coughlin Associates and Shawn DuBravac of the Global Electronics Association echoed concerns about prolonged tightness. A sudden slowdown in AI spending could trigger a boom-bust cycle. Most analysts assign low probability to that outcome before 2028.
Buyers face hard choices. Upgrade existing systems where possible. Lock in contracts early. Accept higher prices or reduced specifications. Entry-level devices may vanish or migrate to slower memory configurations. PC and smartphone shipments will likely miss earlier targets. The memory wall has become the new limiter on electronics growth. Suppliers enjoy rich margins for now. Everyone downstream pays the bill. And the invoices keep getting larger.
WebProNews is an iEntry Publication