Goldman Sachs just extended the semiconductor equipment cycle forecast to 2028. The market will read this as a bull signal for every chip stock between here and Taipei. I read it as something else entirely: a carefully worded admission that the traditional cyclicality of this industry is dead, replaced by a structural demand curve that most portfolios are not positioned for.
Let me be precise about what changed. The revised forecast puts wafer fab equipment (WFE) spending at $150 billion in 2026, climbing to $281 billion by 2028. That is not a gentle upward slope. That is a 45% annual growth rate at its peak in 2027, before decelerating to 29% in 2028. The numbers are large enough to make any macro analyst pause and check their liquidity assumptions.
Here is the context most coverage is missing. This is not a traditional up-cycle driven by consumer electronics replacement or enterprise IT refreshes. This is an AI-driven infrastructure buildout, and it is concentrated in three specific areas: DRAM, HBM, and leading-edge foundry. The old cycle logic dictated that memory capex spikes were followed by oversupply and price crashes. The new logic, embedded in the Goldman forecast, suggests the AI demand curve is structural enough to extend the investment horizon to 2028 and beyond.
I have been tracking WFE spending for fifteen years. I have never seen a forecast this bold. It is not the size of the numbers that stands out; it is the underlying assumption that the DRAM supercycle will last twice as long as the 2017-2018 peak. For that to hold, the AI demand must not be a bubble, but a new economic baseline. And here is where I start to smell the contradiction.
Let me get to the core technical analysis.
The forecast implicitly assumes HBM4 enters production smoothly in 2026, with 16-layer stacks becoming the standard. HBM3E currently consumes three to four times the wafer capacity of a standard DDR5 die due to the stacking complexity. HBM4 will push that further. This means the capex is not just about adding new fabs; it is about deploying a more expensive, more complex equipment mix.
The equipment density per wafer is rising. The transition from EUV to High-NA EUV is not a simple upgrade. A single High-NA EUV unit costs over $300 million and has a delivery lead time of 24 months or more. The 2026-2028 spending trajectory reflects this, but it also reflects a hidden bet: that ASML can actually produce enough units to meet demand. Their current annual EUV capacity is roughly 50 to 60 units. The forecast demand does not leave room for a production slip.
Now let me apply the anti-yield rationality framework I use in crypto markets. In crypto, we chase APY, and that is where the system breaks. Here, the system is chasing the AI narrative. The analysts are extrapolating current demand curves as if they are permanent. I have seen this pattern before. The signal is not in the headline number; it is in the deceleration. The 2028 growth rate drops to 29% from 45% the year prior. That is not a plateau; that is the first sign of a cyclical peak.
The market will focus on the 2027 peak. The smart money should be looking at 2028 and asking what happens after the infrastructure buildout is complete. There is an assumption that demand will continue to absorb the supply. I am not convinced.
The contrarian angle here is more uncomfortable than the usual bearish take. The geopolitical risk is not the primary risk to this forecast. The real risk is the AI bubble hypothesis. We are seeing capital expenditure at levels that have historically preceded a painful correction. In crypto, we call it yield farming. In semiconductors, we call it infrastructure investment. The dynamics are the same: money chases a narrative, capacity expands, and the demand curve eventually flattens.
I am also looking at the supply chain concentration. ASML has a 100% monopoly on EUV lithography. The top five equipment vendors control over 80% of the advanced etching and deposition market. This is a fragile ecosystem, but the fragility is not where the market expects it. The market worries about the US export controls on China. That is a real risk, but the forecast already absorbs that risk, as China continues to expand using domestic equipment for mature nodes. The actual blind spot is the concentration of the entire industry on a single demand driver: AI data centers.
If the AI infrastructure investment wave matures by 2028, as the Goldman forecast implicitly suggests, the next cycle will be brutal. The depreciation costs from 2025-2027 capex will hit the P&L in 2027-2029. Storage companies will see gross margin pressure of 5 to 10 percentage points. The equipment makers will see a demand cliff. The stock market will have to price that in.
Institutions smell blood when retail smells profit. The retail investor sees a three-year growth cycle. I see a three-year growth cycle followed by a structural downturn. The market is always lying at the top.
I will not argue against the short-term thesis. The data supports strong WFE spending through 2026. But the positioning must be different. Instead of chasing the equipment makers at these valuations, consider the memory makers with the highest HBM exposure, such as SK Hynix, which still trade at a reasonable PE and have the earnings elasticity. Or consider the companies that have pricing power in the packaging space, as CoWoS is a true bottleneck.
The signal is not in the forecast itself. The signal is in the shape of the forecast. A straight line to 2028 is the market's way of saying the cycle is dead. But the cycle is not dead. It is just hiding, waiting for the right moment to reassert itself.
The next 24 months will be the most bullish period for the industry. The smart move is not to be the last one out of the door. The smart move is to be the one who identifies the 2028 turning point and positions before the market sees it.
Here is the cold truth from my own audit of the data: the infrastructure buildout is real, but the narrative around the buildout is not. The numbers are the result of a specific set of assumptions about AI demand, about geopolitical stability, about equipment supply, about process yield improvements. Any one of those assumptions fails, and the forecast changes.
The market wants to believe in the forecast. The market wants to believe the AI demand is structural. I have seen enough cycles to know that the market always believes in the forecast at the top. The question is whether the top is now or in 2028. My bet is on a more nuanced timeline. The equipment spending will continue, but the curve will not be smooth. It will be a series of fits and starts, corrections, and recoveries.
Watch the liquidity, ignore the narrative. The narrative is the forecast. The liquidity is the actual capital flow. When the liquidity starts to run out, the forecast will be revised again. And the market will be surprised, because it always is.
I am not saying the semiconductor cycle is over. I am saying the cycle is being repriced. The 2028 target is a marker, not a guarantee. The systemic risk hides where the charts are too clean. And the charts here are too clean.
The signal is not weak. The signal is just early. The noise is the 2028 forecast. The signal is the deceleration at the end.
Position accordingly. The yield is in the price of the equipment makers now, not in the delivery of the cycle. I will be watching the quarterly earnings, the delivery times, and the yield reports. The noise is deafening, but the signal is getting clearer.
One last thought: I was here in 2017, watching the ICO mania and the empty promises. I am here in 2025, watching the AI buildout and the equally confident promises. The names are different, the numbers are larger, but the structure is the same. The structure precedes price. The price will follow the structure. And the structure has a ceiling.
