Goldman Sachs just raised its wafer fab equipment (WFE) spending forecast to $281 billion by 2028. The market cheered. I read the fine print and saw a different number: the probability that this forecast is built on a house of cards called 'AI demand persistence.'
This isn't a call to short ASML. It's a structural audit of the assumptions underpinning the most crowded trade in the semiconductor complex. And for anyone holding crypto assets tied to the AI narrative—which is to say, most of the market—the implications are direct.
The Context: A Supply Chain Under Tension
The WFE market is an oligopoly. ASML controls 100% of EUV lithography. Applied Materials, Lam Research, and Tokyo Electron dominate etch and deposition. KLA owns inspection. These five firms extract 10-15% of the industry's total profit pool, but their 'choke point' leverage is worth ten times that.
Goldman's revised forecast implies a 20%+ CAGR from 2024 to 2028. That's not a trend; that's a supercycle. The drivers are clear: 2nm GAA logic from TSMC and Samsung, HBM4 memory from SK Hynix and Micron, and the relentless expansion of CoWoS advanced packaging capacity. TSMC alone is spending $65 billion in Arizona. SK Hynix is committing $90 billion to a memory cluster in Yongin. These are real projects with real timelines.
But here's the structural tension the forecast glosses over: ASML's EUV production capacity is roughly 50-60 units per year. Delivery lead times are 12-18 months. High-NA EUV, required for 2nm and below, costs over $300 million per unit. The supply chain cannot simply scale to meet a $281 billion demand curve. The bottleneck isn't demand; it's the physical capacity to build the machines that build the chips.
The Core: Deconstructing the Forecast's Hidden Assumptions
I've audited enough tokenomics to know that every forecast is a narrative with a P&L attached. Goldman's WFE projection rests on three implicit assumptions that deserve scrutiny.
First, the yield curve. The forecast assumes 2nm and HBM4 yields will ramp quickly—by 2026-2027. This is optimistic. TSMC's 3nm took two years to reach 80% yields. Samsung's 3nm GAA is still struggling at 60-70%. HBM4 requires hybrid bonding, a fundamentally different packaging technology with its own learning curve. If yields lag, equipment purchases get delayed, not cancelled. But delayed spending in 2026 becomes cancelled spending in 2028.
Second, the demand persistence. The forecast requires AI capex from hyperscalers—Meta, Google, Microsoft—to remain at current levels through 2028. That's a bold assumption. We've seen this movie before. In 2021, every CSP was building out capacity for a metaverse that never materialized. The subsequent correction in semiconductor capital equipment was brutal. AI is different, the bulls argue. It has real revenue. True. But the market is pricing in perfection, and perfection is a fragile state.
Third, the structural shift in spending. Goldman's forecast implies memory will overtake logic as the primary WFE driver. This is a significant change. Memory fabs require a different equipment mix—more etch, more deposition, more advanced packaging tools. This benefits Lam Research and TEL, but it also means the industry is becoming more cyclical, not less. Memory is the most boom-and-bust segment of the semiconductor industry. Basing a multi-year growth forecast on the most volatile end-market is a choice.
The Contrarian Angle: The Crowd Sees a Supercycle; I See Optionable Variance
The consensus view is that this time is different. AI is a structural shift, not a cyclical uptick. The equipment makers are 'picks and shovels' plays with pricing power and secular tailwinds. The crowd sees a smooth upward trajectory to 2028.
I see a volatility surface that's mispriced. The market is paying for linear growth, but the underlying asset is inherently non-linear. The WFE cycle has historically overshot to the upside and the downside. The 2018-2019 downturn saw WFE spending drop over 10%. The 2022-2023 correction was similar. The industry has a 'herd mentality' problem: when times are good, everyone builds capacity simultaneously, creating a glut that destroys the next cycle's returns.
Goldman's forecast, by its very existence, accelerates this dynamic. It gives CFOs at TSMC, Samsung, and SK Hynix the cover to approve aggressive expansion plans. The result will be a capacity overshoot in 2029-2030. The smart money isn't buying the equipment makers at these valuations; it's positioning for the inevitable repricing when the first major AI capex cut is announced.
There's also a geopolitical overlay the forecast underweights. The US, Europe, Japan, and China are all subsidizing domestic fab construction. This is a classic collective action problem. Each region is building capacity for self-sufficiency, but the aggregate result is global overcapacity. The CHIPS Act, the European Chips Act, Japan's semiconductor revival plan—these are all demand-side stimuli for WFE, but they're also creating a fragmented, inefficient supply chain that will eventually rationalize through a brutal downcycle.
The Takeaway: Position for the Repricing, Not the Trend
I didn't flee the ICO crash; I shorted the panic. The same playbook applies here. The WFE supercycle narrative is real, but it's already priced in. The opportunity is in the variance, not the mean.
Watch the leading indicators: ASML's order backlog, memory contract prices, and hyperscaler capex guidance. When those start to roll over, the equipment trade will unwind fast. Volatility is the premium you pay for opportunity, and right now, the market is offering a discount on that premium.
Leverage amplifies truth, it doesn't create it. The truth is that semiconductor equipment is a cyclical business masquerading as a secular growth story. The cycle will turn. The only question is whether you're positioned to profit from the turn or be crushed by it.
The crowd sees a $281 billion opportunity. I see a $281 billion risk that's being systematically underpriced. The smart play isn't to fade the trend; it's to own the optionality that comes from understanding its fragility. Theta decay doesn't care about your feelings, and neither does the semiconductor cycle.