SEMI Forecast $116 Billion for Chip Equipment Spending in 2026. The Real Number Came In at $133 Billion.

Taiwan Semiconductor Manufacturing Company closed at a record $485.80 a share in New York trading on Monday. That put its market capitalization at roughly $2.4 trillion, a new high for the company. The same week, Morgan Stanley raised its price target on Lam Research, a company that makes none of the chips TSMC sells but builds the machines that print them, to $385 from $367. Those two moves share a cause, and the cause explains more about where AI spending is actually headed than either stock price does on its own.

Start with the forecast. In October 2025, SEMI, the trade association that tracks global chip-equipment spending, projected that chipmakers worldwide would spend $116 billion on 300mm wafer fab equipment in 2026. That number has since been revised to $133 billion, an increase of 18% on a forecast that was barely a year old. The group's 2027 estimate moved even further, from $120 billion to $151 billion, a jump of 26%. Trade groups that track an industry closely don't usually miss their own numbers by double digits, twice, inside the same forecasting cycle.

[[IMG1]]

TSMC's own guidance shows the same pattern playing out at the company level. In January, the company told investors to expect 2026 capital spending of $52 billion to $56 billion. By July, that range had moved to $60 billion to $64 billion, roughly 15% higher at the midpoint. Normally a company raising its own budget would be the headline. Here it's almost a footnote, because the equipment TSMC actually needed grew faster than the money it set aside to buy it.

Cliff Hou, TSMC's deputy co-chief operating officer, described the mismatch directly this summer. The equipment TSMC had planned to buy at the end of 2025 had already grown to 1.9 times that original estimate by July, a roughly 90% increase in just over half a year. In his three decades at the company, Hou said, he had never seen demand grow at such speed and frequency, and even that expanded buildout still wasn't enough to meet the full market. TSMC's own capex budget, meanwhile, rose only about 15% over that same stretch. Something in that gap has to give, and increasingly it looks like equipment makers are the ones capturing it.

Why It Matters

That's the backdrop for Morgan Stanley's upgrade of Lam Research. The firm expects Lam to guide to more than $9 billion in revenue for its December quarter, well above the roughly $8.1 billion analysts had modeled, and it now projects a wafer fab equipment spending run-rate above $200 billion, annualized, by that same quarter, 60% higher than a year earlier. That figure sits well above SEMI's full-year number for 2026, and the gap is worth sitting with rather than smoothing over. Morgan Stanley is describing a run-rate measured at the fastest point of the year; SEMI is averaging spending across all twelve months, including the slower start. Both can be true at once, and the spread between them is really a measure of how quickly the spending curve is bending upward within the year itself.

Lam's piece of that pie is growing too. Morgan Stanley puts the company's share of logic-chip equipment spending at 11.8% in 2026, an extension of roughly 300 basis points of share gained in 2025, with another 80 basis points expected next year. Overall, Lam is projected to hold 12.3% of total wafer fab equipment spending. None of that happens unless chipmakers are short on tools and willing to pay for whoever can deliver them, which is exactly what TSMC's own numbers describe.

The packaging side of TSMC's business tells a similar story from another angle. Goldman Sachs has raised its estimate for CoWoS, the advanced packaging process that binds memory to logic for AI accelerators, to 664,000 wafers in 2025 and 1.56 million by 2027, more than double over two years. TSMC CEO C.C. Wei said in July that the company's packaging capacity is "so tight that now it's limiting my customers' growth." The company's own memory partners are feeling a similar squeeze from the other direction. A record quarter at Micron still wasn't enough to lift Korea's chip stocks, a sign that even strong demand doesn't automatically translate into strong supplier margins the way it's doing for equipment makers right now.

What to Watch Next

TSMC reports third-quarter results on October 15, with guidance already pointing to $44.6 billion to $45.8 billion in revenue and a 65% to 67% gross margin. That call, and the capex number that comes with it, will show whether the budget gap Hou described in July has widened further or started to close. Equipment peers reporting in the following weeks, Applied Materials, KLA and ASML among them, are the real test of whether Lam's upgrade is company-specific or the first of several. And the breadth question matters here as much as anywhere else in this market: Marvell's own AI forecast has already drawn scrutiny over how few S&P 500 constituents are actually confirming this rally, and a capex boom concentrated in a handful of equipment names would deserve the same scrutiny.

The Pulse24 Take

The interesting thing about this story isn't that chip stocks are near records. Plenty of things are near records right now, and not all of them deserve the label. What's different about this one is that it's backed by a forecasting error that happened twice, in the same twelve months, from the industry's own trade association, plus a specific company admitting its tool orders outran its budget by a factor that embarrassed its own cost planning. That combination is harder to wave away as sentiment.

It's also a reminder of where the actual bottleneck in the AI buildout sits. Chip designs can be revised in software. Capital can be raised in a quarter. Lithography tools, etch chambers and packaging lines take months to build and longer to qualify, and that's a constraint nobody can spend their way around on short notice. That's the case for why equipment makers, not just the chipmakers buying from them, have room to keep repricing higher as long as the demand holds.

The honest caveat is that physical bottlenecks cut both ways. A company like Lam carries real operating leverage, which means a slowdown in AI capex would hit its margins harder than it hit TSMC's. The forecast revisions so far have only gone in one direction, but a trade group that missed its number by 18% on the upside this year has no particular reason to be more accurate next time, in either direction.