The Contest Is for Second
The foundry table split into two races, and the movement is at second place
The quarterly foundry market-share numbers came out, and the headline everyone will quote is that TSMC edged up to 72.5 percent from 72.3 percent. That's not a story. A fifth of a point of movement at the top of a market this concentrated is noise. The story is what happened underneath the headline, because the foundry table has quietly split into two entirely separate races that happen to share a single ranking, and the interesting motion isn't at first place. It's at second.
Read the leader's number correctly
Start with TSMC, but read its share gain for what it actually is. Its 3-nanometer and 5-nanometer lines are at full utilization, sold out, and its revenue grew 12 percent sequentially, outpacing the roughly 11.5 percent average of the top ten, on a combination of higher average selling prices and more wafers.
The temptation is to call that a share win, as though TSMC out-competed someone for business. That's the wrong frame. When your most advanced lines are at full utilization, you aren't winning share by selling harder. You're capturing demand that has nowhere else to go, and pricing into it because you can. The leading-edge AI accelerators the whole industry needs can only be made in a handful of places, and increasingly in one, so the demand flows to TSMC by default and TSMC raises its prices into the queue. That's capacity-constrained pricing power, the same dynamic showing up across the AI supply chain: price and volume both rising because the buyer has no alternative. TSMC's 72.5 percent isn't the result of a good quarter of selling. It's the result of being the only fully-qualified door at the node that matters, and charging accordingly.
Samsung is the tell
The number that deserves the attention the headline won't give it is Samsung's. It remained second, but its share fell from 6.5 percent to 5.9 percent, and here is the part that inverts the obvious read: its revenue rose 1.8 percent, on higher 5-nanometer pricing and an HBM ramp. Revenue up, share down.
That combination is a specific signal, and it's the same one that showed up in the memory market this year with SK Hynix. A company loses share while growing revenue when the total market is growing faster than it is. Samsung isn't contracting. It's being outrun. It is participating in the same AI-driven boom, raising prices and ramping advanced product, and still slipping in relative position because the leader is capturing the incremental leading-edge demand and Samsung can't win the same allocation. That's the signature of a number-two that can't match the leader's pull, not a business in trouble, but a business losing a race it's running as hard as it can. For a foundry that has struggled with leading-edge yield and customer wins for years, revenue-up-share-down is the quantified version of that struggle.
The other race entirely
Now look at third place, because it's running on a completely different engine. SMIC jumped 20 percent in revenue to a 5.4 percent share, up from 5.1 percent, and the composition of that growth is the whole point. It came from seasonal inventory builds for PCs, notebooks, and consumer electronics, from AI-peripheral and server-networking chips, and from memory-related price increases, plus mature-node demand generally. Nowhere in that list is the leading edge, because the leading edge is gated away from SMIC by export controls.
So SMIC is climbing the same table as TSMC and Samsung while playing a different game. It's winning on the trailing edge, on mature nodes, on the AI-adjacent chips that surround the accelerators rather than the accelerators themselves, and on Chinese domestic demand and substitution that the leaders have effectively ceded. It cannot compete for the sold-out 3-nanometer AI business, and it isn't trying to. It's growing on the part of the market the leaders aren't fighting over. Same table, opposite engine.
The convergence that's the actual story
Put the second and third numbers together and you get the finding the headline buries. The gap between Samsung and SMIC collapsed from 1.4 percentage points to half a point in a single quarter, 6.5 to 5.9 for Samsung, 5.1 to 5.4 for SMIC, and the two are moving in opposite directions. One is falling, the other rising, and they are now within half a point of each other.
Extend those trajectories even one or two more quarters and a Chinese trailing-edge foundry takes the number-two spot in the global foundry rankings from Samsung. Sit with how strange that sentence is. It would happen not because SMIC beat Samsung at anything Samsung cares about, not at the leading edge, not on the AI accelerators, but because SMIC is winning a different race, on mature nodes and domestic demand, while Samsung loses ground in the race it's actually in. The number-two seat in global foundry could change hands between two companies that aren't competing with each other, decided by opposite forces acting on the same ranking.
That's why reading the table as one ordered list is misleading. TSMC on top, Samsung second, SMIC third looks like a single competitive hierarchy. It isn't. It's two races stacked into one column: a leading-edge race TSMC dominates and Samsung is losing, and a trailing-edge race SMIC is winning against competitors who mostly aren't the names above it. The ranking makes them look like they're all running the same course. They aren't.
The honest bounds
A few cautions keep this from being over-read. Quarterly foundry share is noisy, and one quarter of Samsung down and SMIC up is a data point, not a confirmed trend; a strong Samsung quarter or a seasonal SMIC dip could widen the gap back out. The convergence is a trajectory worth watching, not a foregone conclusion. Samsung's foundry business is also a small piece of Samsung overall, dwarfed by its memory operation, so losing the foundry number-two seat would be a symbolic and strategic blow more than a financial one. And SMIC's rise, real as it is, is capped where it matters most: it is structurally locked out of the leading edge by export controls, so it can climb the revenue table on mature nodes and still pose no threat whatsoever to TSMC's actual franchise. The two-races framing cuts both ways. SMIC winning its race doesn't touch the race that determines who makes the world's most advanced chips.
What to actually watch
Read the foundry table, then, as two questions rather than one. The first, who leads the leading edge, is settled and getting more settled: TSMC, sold out, pricing at will, with Samsung unable to keep pace. The second, and the live one, is who holds second place, and it's being decided by a company being outrun from above meeting a company climbing from below, on a trajectory that has them crossing. The headline number, 72.5 percent, tells you about the race that's already over. The contest that's actually in motion is the one for second, and it's the one the headline never mentions.
You do not remove a bottleneck. You relocate it. TSMC's sold-out leading-edge capacity is the bottleneck everyone can see, and it's what makes the top of this table boring and settled. The interesting consequence is the one nobody's tracking: by owning the leading edge so completely, TSMC pushed the entire rest of the industry into a different race, and that race, the one for everything below the frontier, is where the standings are actually moving.