Chen Li-bai doesn’t mince words. The chairman of ADATA Technology sees no quick fix for the memory chip crunch gripping the industry. Shortages will drag on for another 10 years. Talk of an AI investment bubble? Save it for 2040 or beyond.
His comments landed this week amid fresh warnings from other executives. SK Hynix CEO Kwak Noh-jung told Reuters the global memory industry faces its worst-ever supply shortage in 2027. Demand will outstrip supply by a wide margin. Reuters captured the stark forecast.
But Chen goes further. He argues analysts underestimate the breadth of AI’s pull. Future applications will span B2B, B2G, B2C and B2B2C models at once. Judging the boom by short-term capital spending or single-company metrics? He calls that “a view of the sky through a pipe.” The TechRadar report quotes him directly.
Persistent Imbalance in Supply and Demand
Memory makers Samsung, SK Hynix and Micron push capacity higher. Yet Chen insists it won’t close the gap. Electricity, especially green sources, and memory itself rank as the two scarcest resources ahead. Manufacturers now favor measured expansion. No repeat of past overbuilding sprees that crashed prices.
Data backs the tension. IDC projects 2026 DRAM supply growth at just 16% year-over-year. Below historical norms. NAND fares little better at 17%. The IDC analysis highlights impacts on smartphones and PCs. Data centers could consume 70% of all memory chips produced in 2026. That leaves slimmer supplies for everything else.
Prices tell the story too. RAM jumped as much as 89% in 2026. DDR5 kits in Germany rose 7% in July alone. They now sit 448% above 2025 levels. Gaming rigs, GPUs and consumer devices feel the pinch. Shattered.io tracked the surge.
And the drivers run deeper than hyperscale data centers. Robots. Autonomous vehicles. Unmanned factories and stores. Smart homes. Low-orbit satellites plus supporting ground systems. All demand more memory. The three dominant producers can’t satisfy this across a decade. Even Chinese capacity adds limited relief. Chen’s outlook, echoed in Wccftech coverage, frames it as structural. Not cyclical.
High-bandwidth memory for AI accelerators accelerates the squeeze. HBM takes three to four times the wafer capacity of standard DRAM per bit. Its share of total DRAM production jumps from 18% in 2025 to 35% by 2028. Deutsche Bank models show demand exceeding supply every year from 2026 through 2030. Shortfalls start at 210,000 wafer starts per month in 2026. They peak near 795,000 in 2028. The bank’s analysis, discussed widely on X and in IEEE Spectrum, paints a multi-year imbalance.
HBM demand grows 56% to 67% annually in that window. Standard DRAM for servers, PCs and phones gets crowded out. New fabs require two to three years to ramp. Supply can’t pivot fast enough. Micron, one of the big three alongside Samsung and SK Hynix, already sells out its HBM output. Higher-margin mix expands almost automatically. Recent X posts from analysts like @SemiconductorsX and @MilkRoadAI highlight how steady AI demand alone sustains pricing power. No acceleration needed.
PC and smartphone markets shrink under the pressure. Up to 9% contraction possible in 2026. Contract prices for server DRAM soar. Some forecasts see 55-60% jumps in Q1 2026 alone. Insight Enterprises warns of “RAMageddon” driven by AI infrastructure. Their campaign brief urges planning for 10-20% monthly increases through year-end.
Chen Li-bai’s decade-long view stands out for its bluntness. He dismisses bubble fears outright. The AI boom, in his telling, has only begun. Electricity and memory constraints will define the next 10 years. Other voices align on timing if not exact duration. SK Hynix eyes 2027 as the nadir. Deutsche Bank sees tightness persisting to 2030. Tom’s Hardware covered Chen’s remarks in detail just days ago. Tom’s Hardware noted his call for rational industry behavior.
So what does this mean for the broader tech supply chain? Memory becomes the bottleneck. Not just for AI training clusters but across industrial automation, edge devices and consumer electronics. Chip designers scramble for efficiency gains. System architects rethink architectures to stretch every bit. Equipment makers and module producers like ADATA itself ride elevated pricing. Yet downstream buyers from automakers to PC OEMs absorb higher costs or delay projects.
Green electricity adds another layer. Data centers already strain power grids. Memory production itself is energy intensive. Chen’s pairing of the two scarcities underscores intertwined limits. Fab expansions face permitting, financing and energy availability hurdles. History shows the industry often overshoots on capacity. This time discipline may hold longer. But even disciplined growth lags explosive demand.
Recent market moves reflect the narrative. Memory-related stocks fluctuate on every new forecast. Micron benefits from sold-out HBM and shifting mix. Samsung and SK Hynix guide higher prices into 2027. X conversations buzz with charts showing persistent shortfalls. One post from @IntCyberDigest summarized Chen’s warnings on electricity and memory as the decade’s defining constraints.
Relief remains distant. Analysts see no balance before 2028 at the earliest. Many push into the 2030s. For industry insiders tracking capex, wafer starts and bit supply, the message is clear. Prepare for tight markets. Elevated prices. Strategic allocation of scarce DRAM and HBM. The AI wave doesn’t break soon. It builds. And memory sits at its foundation.
Chen’s perspective, grounded in decades of industry cycles, carries weight. He urges wider vision. Beyond quarterly capex reports. Beyond one firm’s utilization rates. The combined demands of intelligent systems, from data centers to satellites, reshape fundamentals. Supply responses take years. Demand compounds faster. That mismatch defines the outlook. For now and for years ahead.
WebProNews is an iEntry Publication