JPMorgan's KLA call is really a bet on the WFE market rotating from memory to logic

On Friday, JPMorgan raised its wafer-fabrication-equipment (WFE) forecasts and named KLA its top pick among U.S. chip-equipment makers. Read the surface, and it looks like one more AI-equipment endorsement: analyst Mio Shikanai lifted the bank's WFE growth estimate to 31% in 2026, taking the market to $163 billion, then 38% in 2027 to $225 billion and 17% in 2028 to $263 billion, a 28% compound rate from 2025 through 2028. The driver JPMorgan names is demand — cloud providers investing at a 58% annual clip through 2028.

The more useful way to read the call is as a rotation. The WFE market has split into two sub-markets, and KLA is the leg that has not been credited yet.

The evidence of the split is in the year-to-date stock moves, which JPMorgan cites directly. KLA is up about 50% this year. Lam Research is up roughly 85% and Applied Materials up about 82%. The reason for the gap is not that KLA is a worse business. It is that 2026's equipment spending is memory-led — DRAM and HBM capacity — and Lam and Applied carry more direct exposure to that spending. KLA's revenue is process control: the inspection and metrology tools that watch every wafer, every layer, every node. Process-control content spreads across foundry, logic, and mature capacity, so KLA moves with a different sub-market than the memory lever did.

That is the two-market map in its cleanest form. 2026 favors the memory lever. The underperformer is the foundry-and-logic lever, and JPMorgan's top-pick call is a wager that 2027 flips the mix — that leading-edge logic becomes the constraint and the process-control vendor tied to it gets re-rated relative to the memory names that already ran.

The mechanism JPMorgan leans on for that flip is supply discipline and constraint location, not unit demand. Memory supply is expected to lag demand even into 2028, meaning suppliers are holding back capacity rather than flooding DRAM. And leading-edge capacity at TSMC is expected to stay above 100% utilization — the bottleneck sits at the most advanced logic node, the 2nm transition where AI accelerators are built. Where the constraint sits determines who has pricing power and where equipment dollars flow; JPMorgan is betting that constraint migrates from memory to leading-edge logic during 2027, with Intel's rising capex as a further upward-revision catalyst for KLA.

None of that makes the pick cheap. KLA still trades at roughly 49 times trailing earnings and near 37 times EV/EBITDA, with a market capitalization around $236 billion, even after lagging its peers. The "attractive risk/reward" JPMorgan describes is relative — a re-rate of the laggard within the group, not a value bargain. The quality underneath holds: gross margin above 60%, operating margin near 42%, free-cash-flow margin near 28%, and return on invested capital in the low 40s, making it the most profitable franchise in the sector.

The number that deserves the most scrutiny is the magnitude, because JPMorgan is far outside the industry's own forecast. The semiconductor trade body SEMI projects WFE of $143.9 billion in 2026, up 23.1%, reaching roughly $200 billion by 2028. JPMorgan is about $19 billion above SEMI for 2026 and years ahead of it through 2028.

WFE forecast ($B)	JPMorgan	SEMI
2026	$163	$143.9
2027	$225	~$175
2028	$263	~$200

Both series are forecasts, not reported numbers, and both point the same direction — record equipment spending for years. The gap is about pace, and it matters because the entire "top pick" thesis rests on JPMorgan being right about the magnitude and the timing of the rotation. A market that grows toward SEMI's path still supports KLA's revenue; it just removes the upward-revision fuel that an aggressive forecast implies.

The distinction between memory-led and logic-led growth is not academic for KLA's holders. It is the difference between a stock that re-rates because a second sub-market finally prices in, and a stock whose multiple was already rich. The live question is not whether total WFE grows — every forecast says it does. It is whether the mix rotates toward foundry and logic in 2027 the way JPMorgan expects, so that process-control revenue gets its turn.

The condition to watch is the mix, not the total. If leading-edge logic and Intel-driven capex keep TSMC above full utilization and pull equipment content toward the processes KLA measures, the laggard case holds and upward revisions follow. If the constraint instead stays in memory, the memory names keep taking the content and KLA's relative discount is a discount for a reason. The market has split; the pick is a bet on which side of the split earns the next cycle.