The number hit my terminal at 06:47 Mumbai time: Goldman Sachs is projecting wafer fab equipment spending to hit $218 billion by 2027 and $281 billion by 2028. I saw the wire tap before the wallet drained. I stopped reading the analyst note and started dissecting the assumptions buried in the forecast. Because a number like that is never just a number. It is a bet on physics, on yield curves, and on the fragile assumption that AI demand will not blink before the decade turns.
Context: The Baseline and the Bet
We are not starting from zero. The semiconductor world is already running hot. TSMC is pushing its Arizona campus toward volume production with a $65 billion commitment, while Samsung and SK hynix are pouring money into fabs that won't see daylight until 2027. The current baseline for WFE sits around $100 billion annually. Goldman's new trajectory implies a compounded annual growth rate of roughly 20% from 2024 through 2028. That is not a forecast; that is a thesis statement.
The core driver is not vague; it is AI. The demand for HBM, for advanced packaging, for GAA transistors at 2nm and below. Goldman is betting that the infrastructure build-out for AI will outlast the hype cycle. I am not here to argue with the direction; I am here to stress-test the speed.
Core: The Data and the Bottlenecks
The numbers reveal a structural shift in where the money is going. The report implicitly prioritizes DRAM/HBM capacity expansion over classic logic scaling. My read of the equipment order books confirms this. Lam Research, TEL, and ASM International are all seeing a disproportionate surge in orders related to TSV etching and hybrid bonding. This is a new cycle. It is not the 2021 expansion, which was logic-driven. The WFE growth now is storage-driven.
The bottleneck is real. ASML produces around 50 to 60 EUV machines per year. High-NA EUV units cost over $300 million each and face a delivery lead time of 12 to 18 months. The sheer physical output limits the pace of capacity expansion at 2nm fabs. Even if a customer writes a blank check today, they are waiting until 2026 for the machine. The equipment supply chain is the single biggest constraint on the Goldman forecast.
The yield curve is the silent variable. HBM4 and 2nm GAA are not just about design. They are about yield ramps. TSMC's 3nm is currently yielding over 80%; Samsung's 3nm GAA is hovering around 60-70%. If 2nm or HBM4 yield curves disappoint, the equipment spending is not canceled; it is delayed. The pull-forward of revenue into 2027-2028 is predicated on yields hitting their stride by 2026. The faster the yield ramp, the more urgent the capacity expansion. This is the invisible feedback loop.
The oligopoly is intact. ASML, AMAT, Lam, TEL, KLA. Their pricing power is absolute. In a supply-constrained environment, the customer pre-pays. The days sales outstanding for these firms are at historic lows. I look at the order book to revenue ratio, which is hovering between 1.5 and 2.0. That is the real-time signal that the forecast is being converted into purchase orders. Speed is the only currency that does not depreciate.
Contrarian: The Supply Chain Blind Spot
Here is the angle that is not in the Goldman note: the forecast assumes that equipment makers can scale output in step with demand. They cannot. I see the bottleneck clearly. The high-NA EUV output will be the limiting reagent. Furthermore, the geographically dispersed build-out is creating a massive duplication of demand. The US, Europe, Japan, and China are all subsidizing local fabs. This fragmentation is a net positive for equipment makers; it is also a net drag on efficiency. The result will be a pricing war for the remaining capacity.
The Chinese angle is the unspoken variable. Goldman's forecast is global, but the China factor is a tailwind for equipment revenue and a headwind for its own advanced node progress. China's domestic equipment vendors are seeing more validation windows as the global supply tightens. A fab that cannot get a tool from AMAT will accept a local alternative. This is the second derivative of the forecast: the scarcity is creating a substitute market.
The depreciation cliff is coming. All of these new fabs will start depreciating their expensive tooling between 2026 and 2028. For the foundries, that means a 3-5% hit to gross margins. For the equipment makers, it is a tailwind. For the stock market, it is a potential warning if the demand curve stalls. The market is pricing in a smooth ride; the history says the cycle always over-corrects.
Takeaway: The Signal to Watch
The next signal is not the headline revenue number. It is the quarterly order intake of ASML and the contract price of DRAM. If DRAM pricing holds its current momentum through mid-2025, the Goldman forecast will look conservative. If it stalls, the $2180 billion for 2027 will be the first casualty. The question is not whether the money is flowing; it is whether the yield curves will deliver. Trust no one, verify the chain, strike first. I am not betting against the cycle; I am betting on the verification of the physics.
While you read the news, I checked the order book. The signal is still loud.