Most people see Goldman's revised wafer fab equipment forecast and think "bullish."
I see a lagging indicator priced for perfection. A $218 billion WFE market by 2027. $281 billion by 2028. That implies a 20%+ CAGR, sustained over four years. It's a beautiful projection built on a fragile stack of assumptions.
Let's quantify the fragility.
The first assumption is that AI-driven demand persists in a straight line to 2028. The second is that HBM4 and 2nm GAA yield curves ramp fast enough to justify the capex. The third, and most dangerous, is that the supply chain—ASML's EUV capacity, in particular—can actually deliver the tools.
Goldman's number is a demand-side forecast. The supply side has its own vote.
Most people anchor on the revenue opportunity. I anchor on the delivery bottleneck. Chaos is data waiting to be quantified. In this cycle, the data points to a capacity ceiling that will shave billions off that optimistic number.
Here's the core trade. Not just which equipment maker wins, but which structural bottleneck gets priced first.
The High-NA EUV Bottleneck
Let's start with the most obvious constraint: EUV lithography.
ASML is the sole supplier. No alternative. Their annual output of EUV machines is roughly 50-60 units. High-NA EUV, the tool required for 2nm and below, is even more scarce. Each unit costs over $300 million. Delivery lead times are 12-18 months, stretching to 24 months for the High-NA systems.
Now overlay Goldman's forecast.
To hit that $218 billion WFE number in 2027, you need TSMC, Samsung, and Intel all pulling forward their 2nm buildouts. You need SK Hynix, Samsung, and Micron to build out HBM4 capacity simultaneously. That means a simultaneous demand for the same critical tools: EUV, high-end etch, and advanced deposition. The math gets tight quickly.
Supply elasticity is near zero in the short term. You cannot just order more High-NA EUV machines. The optics supply chain—Carl Zeiss is the key supplier—is the true bottleneck. They cannot scale to meet a sudden surge in demand.
So Goldman's forecast is a technical possibility. But it is not a physical certainty.
The market is pricing in the demand curve. The physical supply curve is about to throw a wrench in that model.
## HBM's Dirty Secret: Yield The second driver is HBM. Specifically, HBM3E and HBM4.
The AI boom is memory hungry. High Bandwidth Memory is the new gold rush. But the market misunderstands the yield curve.
SK Hynix, the current leader, has an HBM3E yield of roughly 70-80%. That's good, but it's not a 95% logic yield. The complexity of TSV (Through-Silicon Via) etching, the stacking of 8-12 dies, and the thermal management create a challenging manufacturing process.
Now add HBM4. It requires a fundamental shift to hybrid bonding. This is a new tool set, a new process, and a new yield curve. The transition to hybrid bonding doesn't guarantee a fast ramp. It is a hard technological problem.
Goldman's forecast implies HBM4 yield ramps quickly enough to justify capacity expansion by 2026-2027. If the yield stays sticky at 60-70%, the capex may be delayed, not cancelled. The machines get ordered later. The WFE spend slips to 2029-2030.
The yield curve is the hidden variable. It's the number that can blow up the timeline.
Ego is the ultimate systemic risk. Goldman is confident, but the physics of memory manufacturing is not subject to their conviction.
The Chinese Market: The Unquantified Variable
Now, let's inject the variable that Wall Street consistently gets wrong: China.
Goldman's forecast is global. It assumes a rational, frictionless market. But China is a major consumer of semiconductor equipment, representing roughly 20-25% of global WFE. And they are locked out of the high-end tools.
The export controls have done something predictable: accelerated the Chinese domestic equipment push. It's a 10-15 year, but it's happening. For mature nodes, 28nm and above, Chinese tool makers like Naura and AMEC are already viable.
But the real story for the global WFE forecast is not Chinese. It's the duplication of capacity.
Globalization is dead. Regionalization is the new demand driver.
The US CHIPS Act, the European Chips Act, Japan's plans—each region is building its own fabs. This duplication is a boon for equipment makers. You need a TSMC fab in Arizona, a TSMC fab in Kumamoto, and an Intel fab in Germany. That's triple the equipment demand for the same logic.
But it's also a structural inefficiency. It will lead to overcapacity in the late 2020s. This is a classic supply chain tragedy: overbuilding in response to a perceived shortage, leading to a glut.
So, Goldman's forecast is correct on a trend, but I suspect they're underestimating the amplitude. The 2028 number may be too high. The 2029-2030 downturn will be brutal. The cycle will be amplified by the regionalization.
When the cycle turns, the capacity is already built. The WFE demand will fall off a cliff.
The Contrarian Play: Which Toolmaker Wins?
If the forecast is too high, which plays do you pick?
You don't buy the broad index. You buy the specific bottleneck.
The real value in this cycle is not in the EUV lithography (ASML) alone. The new bottleneck is in advanced packaging.
TSV etch tools, hybrid bonding tools, and the metrology that goes with it. Companies like Besi and ASM International are the new power players. Their tool sets are specifically designed for the HBM and CoWoS challenge.
TSMC's CoWoS capacity is the single most constrained node in the AI supply chain. They doubled capacity in 2024 and it's still not enough. That's not a demand story. It's a packaging yield and capacity story. The tools that enable that capacity are the ones to watch.
The WFE total number is not the story. The composition is.
Most people will buy the big names. The real edge is in the enabling technologies that allow the 2D/3D packaging to work. That's where the real scarcity is, and that's where the price points are.
Takeaway: The Signal and the Noise
Goldman's forecast is a beautiful top-down picture. It's the macro view. But the trade is in the micro.
Watch the monthly sales data from ASML. Watch the order book of Lam Research. If their backlog starts to slip, the forecast is dead.
Watch the yield reports from SK Hynix on HBM4. If they don't hit the 70% yield mark by Q2 2026, the 2027 capex number is at risk.
And watch the Chinese equipment procurement. A slowdown there is a canary in the coal mine.
Don't be an Ego-driven macro trader. Be a microstructure trader. The biggest trade in this cycle is not in the equipment's top line. It's in the yield curve and the supply chain choke points.
Liquidity vanishes. Conviction remains. The conviction is in the specific tools that solve the yield and packaging constraints, not the broad thesis.
Chaos is data waiting to be quantified. The 2028 forecast is a hypothesis. The order book is a fact. Watch the order books.