We didn't just hunt alpha; we rewired the game. That's the phrase I keep muttering to myself when I look at Goldman Sachs' latest wafer fab equipment (WFE) forecast. On August 25, 2025, the bank quietly extended its semiconductor equipment cycle projection all the way to 2028, predicting WFE spending will balloon from $150 billion in 2026 to $281 billion by 2028. Growth rates of 36% in 2026, 45% in 2027, then a sharp deceleration to 29% in 2028. The market read this as bullish โ I read it as a supercycle that smells eerily familiar. Having spent years auditing smart contracts in the Ethereum trenches and then watching the Terra collapse unfold in real-time, I've learned that every prolonged euphoria carries a hidden assumption that can't survive contact with reality. And this forecast is built on a mountain of them.
Let's set the stage. The WFE market is the "pick-and-shovel" business of the chip world. ASML, AMAT, Lam Research, TEL โ these firms don't make chips; they make the tools that make the chips. Their revenue is a leading indicator of the entire semiconductor industry's health. The current expansion is driven by three engines: DRAM, HBM, and advanced logic manufacturing. The DRAM memory market is in the middle of a genuine supply squeeze โ inventory sits at 4-6 weeks versus a normal 8-10, and contract prices are up 15-25% quarter-over-quarter. HBM (High Bandwidth Memory) is the new gold rush โ SK Hynix, Samsung, and Micron are the only three players, and demand is so insatiable that SK Hynix alone holds 50%+ of the market. Behind all this sits the AI chip boom: NVIDIA's H100/H200/B100/B200 are sold out, and training chip revenue is expected to exceed $150 billion in 2025. AI training silicon runs on 5nm and below, which pulls advanced foundry capacity โ TSMC is the sole leader at 2nm GAA, which enters production in 2025. The conclusion Goldman draws: AI demand persists through 2028, and therefore WFE spending rises persistently.
But here's where I need to put on my skeptical mentor hat. This forecast is technically coherent, but it hides three structural assumptions that deserve scrutiny. First, the HBM story is a capacity cannibalization narrative. HBM3E, with its 8-layer to 12-layer stacking, consumes 3-4 times the wafer capacity of standard DDR5. So when you see DRAM supply tightening and prices surging, it's not because demand is growing organically โ it's because the memory makers are literally shoveling wafers into HBM production to feed the AI beast. This is a beautiful narrative until you realize it's a zero-sum game: HBM production physically takes wafers away from other memory products, creating an artificial scarcity that can correct brutally once the AI bubble shows signs of fatigue. The second hidden assumption is about equipment economics. The forecast implies that the capex per wafer continues to rise โ and it does, because high-NA EUV lithography machines cost over $300 million each. ASML has 100% market share in EUV, and its annual production capacity is only about 50-60 units. The forecast assumes ASML can scale production fast enough to meet demand โ that's not a given; it's a hope. Delivery lead times are already 12-18 months for EUV, and 24+ months for high-NA.
The contrarian angle โ and I want to be very clear here โ is that this supercycle forecast is dangerously optimistic on the demand side. I've been in the trenches since 2017, auditing smart contracts and later building educational platforms for Indonesian crypto enthusiasts. The psychology is identical: every bull market tells you it's different this time. The AI capex boom is real, but the ROI question is unresolved. Cloud providers are spending billions on AI infrastructure, yet the actual revenue from AI applications is still unclear. If 2026-2027 brings a first-wave AI correction โ and the 29% growth rate forecast for 2028 already hints at a plateau โ the WFE spending will be cut dramatically. I've watched this exact pattern in crypto: the 2021-2022 cycle had a similar "trustless" narrative, and when the music stopped, capital expenditures were slashed 30-50%. The second unspoken risk is geopolitics. The forecast assumes that the U.S.-China tech decoupling stays controlled. But the U.S. has been tightening export controls on advanced semiconductor equipment, and if this escalates to include mature processes, China's expansion โ which is a massive component of global WFE โ will shift to domestic equipment makers. Chinese companies like Naura and AMEC are already moving into the top global ranks, and they're not buying from ASML or AMAT. If export controls tighten further, the forecast is structurally broken.
The deeper lesson here is that the semiconductor industry is now the new mining rig for the mind. Just as miners in the early days of Bitcoin bought GPUs to mine blocks, the modern gold rush is buying lithography tools to mint AI capabilities. But from my years in the crypto trenches, I know that mining rigs are the first to get sold when the bubble bursts. The equipment makers will be the biggest winners in this cycle โ their gross margins are 45-55%, they have deep pricing power, and their earnings visibility is 12-18 months. But the storage companies โ SK Hynix, Samsung, Micron โ they're the ones with the biggest earnings elasticity. They will swing more than anyone. And they're the ones who will break first if the AI demand curve falters. I've seen this pattern in the Terra collapse, in the DeFi summer. The underlying technology is real โ AI, HBM, advanced packaging โ but the price that the market is paying is pricing in perfection. The forecast is a mirror of the market's mood, not a prediction of the future.
When the market sleeps, the architects wake up. That's the lesson I take from Goldman's number. The real signal isn't the $281 billion peak; it's the fact that the growth rate decelerates from 45% to 29% in 2028. That's the tell โ it says the AI supercycle's first wave will saturate by 2028, and the next wave must come from something entirely different, like embodied intelligence or mass-scale AI agents. The smart money is not on the equipment makers โ it's on the leading indicators: NVIDIA's earnings calls, the cloud provider capex guidance, and the contract price of DRAM. Watch those, not the forecast. We didn't just hunt alpha; we rewired the game. The question is whether the industry can actually rewire itself before the market comes to its senses.