The number landed in my terminal like a dropped order book. Goldman Sachs is now calling for global wafer fab equipment spending to hit $281 billion by 2028, growing at a 36% CAGR from 2025 to 2027. That is not an incremental upgrade. That's a complete repricing of the semiconductor cycle.
I've been staring at equipment cycles since 2017, when I audited Golem's smart contracts in Lagos and realized that market narratives always lag the technical reality underneath. The same discipline applies here. Before we accept this forecast, we need to break down what it actually assumes. Because if this number hits, it changes everything. If it misses, the collateral damage reaches far beyond chipmakers.
Let me walk you through the seven dimensions of this equipment cycle, the hidden assumptions baked into the forecast, and where I think the market narrative gets it wrong.
The Context: Why WFE Is the Real Battlefield
Semiconductor equipment is not just another link in the supply chain. It's the toll booth. Equipment represents 70-80% of a fab's total capital expenditure. The companies that make these machines enjoy gross margins of 45-60%, backed by technical monopolies that make the barriers to entry nearly absolute.
ASML controls 100% of the EUV lithography market. Its single-unit price for High-NA EUV systems? Three to four hundred million euros. Applied Materials, Lam Research, and KLA are oligopolies in deposition, etch, and metrology respectively. These are the most profitable companies in the entire semiconductor value chain.
When Goldman says WFE spending will grow at 36% CAGR for three consecutive years, it's making a set of assumptions about every major process node, every advanced packaging line, and every storage fab across the globe. We need to dissect those assumptions.
The Core Analysis: Seven Layers of the WFE Cycle
1. Technical Process: The GAA Window and High-NA EUV
Goldman's forecast implicitly covers a critical period for the semiconductor industry's process evolution. From 2026 to 2028, we're looking at:
- TSMC's N2 (2nm GAA) volume production
- Intel's 18A and 14A roadmap progress
- Samsung's 2nm GAA mass production
These aren't just incremental upgrades. The transition from FinFET to GAA (Gate-All-Around) architecture is one of the most significant process changes in a decade. And yield ramping requires massive equipment input. Every percentage point of yield improvement in these advanced nodes demands additional process iteration, defect detection, and metrology tools.
The more subtle implication is timing. This forecast implicitly assumes ASML's High-NA EUV systems will achieve batch delivery in 2026-2027. Without High-NA EUV, the expansion of advanced process capacity at the scale needed to justify $281 billion in WFE spending is hard to imagine. High-NA EUV has a numerical aperture of at least 0.55, compared to the current 0.33 NA of existing EUV systems. It's the key tool for sub-2nm node mass production.
Confidence in this technical assumption? I'd rate it 6 out of 10. The complexity of High-NA EUV integration cannot be underestimated.
2. Memory's Dominance: The Storage Shift
Here's a significant detail that most coverage of this forecast misses. Goldman lists DRAM/HBM as the primary growth driver for WFE spending. This implies a major structural shift: storage-related equipment spending will exceed logic foundry spending over the next three years.
Why? HBM expansion requires both front-end (DRAM process) and back-end (advanced packaging) equipment investments. The equipment intensity per unit of capacity for HBM is significantly higher than for logic chips. HBM3E consumes 3-4 times the DRAM die area of regular DDR5. HBM4, which enters mass production in 2025-2026, will use hybrid bonding—a technology that requires unprecedented precision in packaging equipment.
The implication here is that memory makers (SK hynix, Samsung, Micron) are set to see their capex-to-revenue ratio reach historic highs. If storage accounts for 40% of the $218 billion in 2027 WFE spending—about $87 billion—memory industry revenues would need to be around $220 billion. That's a 40% capex-to-revenue ratio, far above the historical average of 25-30%.
This only works if DRAM supply remains tight through 2028. And that's a direct bet on HBM demand exceeding current expectations.
3. Capacity Expansion: The Inevitable Bottleneck
The capacity expansion plans are clear and already in motion. TSMC's Arizona Fab 21 Phase 2 is allocating $40 billion for 30,000 wafers per month at N3. Samsung's Taylor fab is investing $25 billion for 2nm capacity. SK hynix is building out its Yongin cluster for HBM. Micron is expanding its Idaho fab for DRAM. SMIC is pushing mature nodes in Beijing and Shanghai.
But here's the issue: equipment supply cannot keep up. ASML's EUV annual production capacity is only 50-60 units. Applied Materials and Lam Research lead times still run 12-18 months. The equipment industry's expansion cycle takes 2-3 years—and demand is growing at 36% annually.
This creates a structural bottleneck. Equipment delivery constraints might limit actual WFE spending by 10-15% below the forecast. The equipment makers hold the pricing power. They can raise prices 5-10% per year in a seller's market.
4. Demand Dynamics: AI's Insatiable Appetite
The end-market distribution tells us where the growth comes from:
• HPC/AI training: 25-30% of revenue, growing at 40-50% • AI inference: 10-15% of revenue, growing at 60%+ • Storage (DRAM/NAND): 20-25%, growing at 30-40%
AI chips are all advanced nodes—NVIDIA's B200, AMD's MI350/MI400—all using 5nm/4nm or below. The CoWoS capacity is expanding 3-4 times between 2024 and 2026, yet it remains the most bottlenecked point in the AI supply chain.
AI inference demand is growing even faster than training. Inference requires 2-3 times the HBM capacity of training chips. As large language models spread to edge devices and production deployment, this demand curve only steepens.
The Goldman forecast implicitly assumes AI infrastructure investment (data centers, GPU clusters) maintains 40%+ growth from 2025 to 2027. WFE spending is a lagging indicator of AI capex by 6-12 months. If AI capex stalls in 2026-2027—which I see as a 30-40% probability given the historically cyclical nature of capital spending—WFE estimates could be revised down 30-50%.
5. Geopolitical Overlays: The Decoupling Tax
The geopolitical dimension adds another layer of uncertainty. US export controls currently restrict advanced process equipment to China (16/14nm and below), EUV technology, specific EDA tools, and HBM-related tech. ASML cannot export EUV and advanced DUV to China since 2024. Japan has imposed controls on advanced process equipment and photoresists since 2023.
China's response includes:
• Export controls on gallium, germanium, and rare earths • The National Fund III (344 billion yuan) focused on equipment and materials localization
Here's the blind spot in Goldman's model. The forecast likely underestimates geopolitical impact on WFE spending. China accounts for 20-25% of global WFE spending. If the US tightens controls further—potentially restricting mature-node equipment—Chinese WFE could drop 50%, translating to a 10-12% reduction in global spending.
But there's a counter-argument: the "localized production" trend. The US CHIPS Act ($52.7 billion), Europe's Chip Act (€43 billion), Japan's semiconductor revival plan (¥2 trillion), and China's national fund are all driving regional fab construction. This geographic fragmentation is an efficiency tax—10-15% higher costs due to redundant construction and capacity overlap. But it also means more total WFE demand than a purely efficiency-driven market would allocate.
6. Competitive Landscape: The Toll Booth's Toll Booth
The competitive structure remains remarkably stable:
• ASML: 80%+ share in lithography (100% EUV) • LAM/AMAT/TEL: 30%/25%/20% in etch • AMAT/TEL/ASM: 35%/20%/15% in deposition • KLA: 50% in metrology
The key unknown is China's equipment makers. Companies like North Fortune, AMEC, and Piotech are advancing rapidly in mature nodes. The domestic substitution rate is currently 20-25% for mature processes and less than 5% for advanced nodes.
If the localization policy and National Fund III support accelerate, these companies could achieve 30-50% annual growth from 2026 to 2028. China's mature-node equipment market is worth about $20-30 billion annually. Moving localization from 20% to 30% would create $3-4 billion in incremental domestic market share—directly squeezing Applied Materials, TEL, and other international players.
The threat from new entrants in advanced nodes remains low. The combination of patent barriers, customer qualification cycles, and scale effects creates a triple defensive moat.
7. Financial Profile: Pricing for a Growth Story
The equipment industry's financials are pristine:
• ASML: 50-55% gross margin, 40-50% ROE • KLA: 60%+ gross margin, 50%+ ROE • Applied Materials: 45-48% gross margin, 30-35% ROE • LAM: 45-48% gross margin, 35-40% ROE
ROIC for these companies exceeds 20%, far above their WACC of 8-10%. They generate massive free cash flow—ASML around €4-5 billion annually, AMAT $2.5-3 billion.
Valuations are at historic mid-to-high levels: ASML at 30-35x PE, AMAT at 20-25x, LAM at 25-30x, KLA at 25-30x. But if the WFE forecast materializes, current valuations still have support.
Here's a contrarian point I want to flag. The market has historically valued these as cyclical stocks with low PE multiples (20-25x). If AI demand persists through 2028, the industry may transition from "strong cyclical" to "growth cycle," potentially shifting the PE range from 20-25x to 30-35x. That would be a systemic re-rating, not just earnings growth.
The Contrarian View: Three Blind Spots
Let me push back on the consensus optimism. There are three structural risks that the forecast doesn't sufficiently account for.
Blind Spot #1: AI Capital Spending Could Correct Earlier Than Expected
The core assumption is that AI demand remains strong through 2028. But AI capital spending has a history of being cyclical. If we see:
• Large model commercialization misses expectations • Cloud providers slowing capex growth • AI chip competition driving price wars
Then the AI capex bubble could burst in 2026-2027, dragging WFE expectations down 30-50%. The probability of this scenario is 30-40%. The market is pricing AI demand at near-certainty. We've seen this movie before in 2021, and then the equipment cycle went through a sharp correction.
Blind Second: The Supply Chain Bottleneck Could Actually Limit Spending
ASML's EUV capacity is 50-60 units per year. The lead time for advanced equipment is 12-18 months. The equipment supply chain is a bottleneck that could restrict actual spending. The WFE industry's expansion cycle is 2-3 years. If demand grows 36% annually, equipment supply cannot keep pace.
The result: actual WFE spending might be 10-15% below the forecast. But this also means equipment makers' pricing power is stronger than expected. They can raise prices 5-10% annually. The market might see higher revenue per unit, but lower total volume.
Blind Spot #3: Geopolitical Decoupling Could Accelerate
China's response to export controls is not just a passive acceptance of supply constraints. The National Fund III is targeting the most critical equipment categories. If China succeeds in maturing-node equipment localization by 2026-2027—which I think is plausible—the global equipment market structure shifts.
The regionalization of supply chains is already occurring. The US, Europe, Japan, Korea, and Taiwan are building separate, parallel ecosystems. This is a 10-15% efficiency tax on the entire semiconductor industry. But it also means higher total equipment spending in the short term, as each region builds out its own capacity.
The question isn't whether the WFE cycle will be strong. It's whether the strength will be sustained. And that depends entirely on AI capex sustainability.
The Takeaway: What This Means for Investors
Let me distill this into actionable signals.
The Goldman forecast is a high-conviction bet that AI demand fundamentally reshapes the semiconductor cycle. It's saying that the industry can sustain 36% annual WFE spending growth for three consecutive years. That's not a normal cyclical upswing. That's a structural shift.
Three things I'm watching closely:
First, the leading indicators. Cloud providers' capex guidance for 2026 is the early warning signal. If they're projecting higher spending, WFE stays on track. If they're hedging, watch out.
Second, the HBM adoption curve. HBM4 mass production in 2025-2026 and hybrid bonding technology adoption are the critical inflection points. If HBM demand exceeds expectations, memory capex could actually overshoot the forecast.
Third, the equipment supply chain. Track ASML order backlog and delivery lead times. If the bottleneck persists, actual WFE may be lower than forecast, but equipment pricing power will remain strong.
The key question for 2026-2028: Is the semiconductor equipment cycle being redefined as a growth industry, or are we just seeing a high-grade cyclical peak?
Based on my experience auditing tech infrastructure—whether it's smart contracts or market structures—the answer is always in the details. The technology is real. The demand is real. But the market's pricing can always get ahead of the fundamentals.
We're in a rare window where AI, memory, and geopolitical factors are simultaneously aligning to drive a semiconductor equipment boom. Trust the technology. Verify the demand. And pay attention to when the market starts to overprice the certainty.
