JPMorgan Raises Wafer Fab Equipment Market Forecast to $263 Billion
nashnova research
JPMorgan raised its wafer fab equipment (WFE) forecast for the second time in weeks, projecting a $263 billion market by 2028 at a ~28% three-year CAGR — AI demand is pulling the entire chip-manufacturing chain into a rare, prolonged expansion cycle.
Why did JPMorgan raise its forecast twice in quick succession?
AI-related demand keeps overshooting expectations. The four largest U.S. cloud providers are projected to grow total capex at a 58% CAGR from 2025 to 2028, forcing chipmakers to accelerate investment in advanced nodes.
This means → the upgrade is not confined to one chip category. DRAM, TSMC, NAND, and logic assumptions were all raised — AI is lifting the ceiling for the entire supply chain.
The 2026 / 2027 / 2028 WFE market now maps to $163B / $225B / $263B, with year-on-year growth of 31% / 38% / 17%.
How tight are the DRAM and TSMC capacity gaps?
DRAM capacity is expected to rise from 1.9 million wafers per month at end-2025 to 2.85 million at end-2028. But HBM — high-bandwidth memory, the "memory partner" bolted onto AI chips — consumes 3–4× the wafer area of conventional DRAM, and its share keeps climbing.
In plain terms = capacity is growing, but HBM devours most of the new supply. JPMorgan estimates an additional 300,000 wafers/month would be needed by 2028 just to reach balance — and considers that target unlikely.
On the TSMC side, N2 / N3 / N5 utilization is forecast to stay above 100% through at least 2028. JPMorgan lifted its TSMC capex estimates to $62B / $81B / $90B for 2026–2028.
Will equipment price hikes become a new variable?
JPMorgan notes the industry's value-based pricing structure is unchanged — the more advanced and scarce the tool, the higher the price. But some companies have started passing cost increases through to customers; Tokyo Electron is the leading example.
This means → from 2027 onward, equipment revenue growth will no longer come solely from shipping more units. Price increases themselves become a meaningful driver.
JPMorgan suggests investors focus on three stock-picking criteria: ① revenue exposure to DRAM and TSMC; ② whether supply chains (materials, components) can keep up; ③ room to improve margins through pricing actions.
What role do Chinese fabs play in this expansion?
JPMorgan raised Chinese fab capex forecasts by 10% for 2026 and 9% for 2027, and introduced a first-ever 2028 estimate of $54.7 billion — a 7% three-year CAGR.
This reflects CXMT and YMTC expansions, the AI-driven memory supercycle, and the strategic priority China places on building a domestic AI chip supply chain.
The domestic equipment sourcing rate is expected to rise from roughly 25% in 2025 to 40% by 2028. AMEC has launched six new tools, but gaps in metrology/inspection and lithography remain wide — JPMorgan frames these as long-term opportunities, not near-term bottlenecks.
When does the memory supply–demand gap peak?
JPMorgan expects supply shortages to persist for two years, with 2027 the most severe — supply covering only 70–80% of demand, easing slightly in 2028.
HBM presents a seemingly contradictory signal: per-die bit demand forecasts were cut (longer 8-layer cycles, delayed 12-layer ramp), yet the 2027 HBM price forecast was raised 42%, pushing the total market estimate higher.
In plain terms = Nvidia's Rubin Ultra downgraded its HBM spec from 16-layer to 8/12-layer HBM4E — not because demand softened, but because supply simply cannot keep up. Higher GPU and AI CPU shipment volumes offset the lower per-unit value.
Is 2027 the make-or-break window for this forecast?
Memory makers' long-term agreements show strong bargaining power: prepayment ratios of roughly 20–25%, volume lock-ins covering 50–70%, and AI/server pricing running 30–40% above other applications.
This means → the supply–demand tension is already locked into contracts — it is not short-term speculation. But 2027 is the widest gap — whether equipment makers' order visibility and pricing power both deliver at that point will determine if this round of upgrades holds up.
Put simply = the money has been spent, capacity has not caught up, and 2027 is when the scorecard gets read.
市场有风险，内容仅供研究参考，不构成投资建议。