Product1 distinct publisher3 min readPublished
The bank's advanced-node shortfall falls from 92% to 34% over ten years, but the count of wafers China still cannot make roughly doubles. Hardware buyers should plan for a rationed second pool.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The sourcing lead deciding this quarter whether a Chinese-fabbed 7nm part belongs on a 2028 board is really weighing a single question: whether wafers get allocated against her order when the line starts. Goldman's own model answers that question underneath its headline figure.
Back the model out. Goldman puts 2035 domestic supply at 410,000 wafers a month against demand of 619,000 [2], with supply compounding at 46% a year and demand rising 17% [3]. Run 410,000 back through ten years of 46% growth and 2025 supply is about 9,300 wafers a month [1]. If that is 8% of demand, as a 92% shortfall implies, current demand is around 116,000 [2], so the gap today is roughly 107,000 wafers a month [3]. In 2035 the gap is 209,000 [4]. The share improves by 58 points [6] while the absolute hole nearly doubles [5]. Working from the demand side instead gives a 1.76x increase, so the direction holds either way [7].
That is the practical difference between a substitute and a second pool. A fab base that covers two thirds of its own domestic demand in 2035 has no exportable surplus, and its output is claimed before an outside buyer picks up the phone.
Yields carry much of the model's weight, and the 2035 assumption still lands below what TSMC posts today, seventeen years into a node it took to volume in 2018 [8]. At those two yield levels you start about 20% more wafers to get the same number of good die [9]. That yield gap raises the cost of every good die coming off a Chinese line, so the capacity reads as something buyers take because they have no alternative, not something they take because it is cheap.
The tool market shows where the ceiling comes from. Goldman sees China's wafer fab equipment market at $53 billion in 2027, with domestic vendors taking 38% by value in 2028 against 26% last year [8]. Chinese suppliers have moved out of etch and deposition into ion implantation, inspection and metrology [12]. Lithography is absent from that list of gains, state-owned SMEE remains behind its competitors according to the report [10], and the 62% of the tool market foreign vendors still hold in 2028 [10] is where the single dependency sits.
Treat the whole projection as carrying wide error bars, because Goldman does. The 2030 capital spending figure works back to roughly $46 billion in last year's version of the same model, a revision of about $36 billion in twelve months [11]. That kind of swing in a single year makes the 2035 figure a moving estimate, not a fixed planning input.
So the forcing function for a bill of materials review is two columns rather than one forecast. First column: does this part require 7nm or below, or is it a mature-node device where Chinese capacity is already plentiful. Second column: does the finished product ship inside China or outside it. Mature node, either market, and domestic sourcing is a straightforward call. Advanced node with China-only shipment puts you in the same queue as every domestic buyer, which argues for a qualified second fab and for reading allocation letters instead of price sheets. Advanced node plus export volume is the box where the 209,000 wafer gap becomes your problem rather than Beijing's, and the honest version of that plan books Chinese advanced-node supply as surge capacity that may not arrive.
Goldman's analysis reached the public through a Chinese media outlet [1], which is worth knowing when the numbers are this favourable to the domestic story and this open about a lithography gap: China still cannot buy the one machine class that sets the ceiling on all of this.
Ranked by verification strength, evidence, and original report placement.
Goldman Sachs projects China's domestic supply shortfall for wafers made on advanced processes of 7nm and below falling from 92% in 2025 to 34% by 2035, according to a Goldman Sachs analysis reported by a Chinese media outlet.
Goldman expects China's advanced-node wafer supply to reach 410,000 wafers per month by 2035, against monthly demand of 619,000 wafers.
Domestic advanced-wafer supply is projected to grow at a compound annual rate of 46% between 2025 and 2035, while demand is expected to increase by 17 percent.
Goldman's model assumes SMIC adds between 30,000 and 50,000 advanced-node wafers of monthly capacity each year from 2026 through 2031, then a further 20,000 wafers per month annually through 2035.
Production yields are projected to rise from 23% in 2026 to 50% in 2030 and 75% in 2035.
TSMC, the world's largest foundry, began mass-producing 7nm chips in 2018 and can achieve yields exceeding 90%, depending on factors including chip design and die size.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 29, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
build
SMIC's first $3 billion quarter comes with a wafer price increase attached1 distinct publisher
product
The export control that failed was Super Micro's own audit, and nine people are charged for it3 distinct publishers
product
Xiaomi qualifies CXMT's LPDDR6 for its September foldable1 distinct publisher
build
China's accelerator swap makes Cambricon supply, not export policy, your ship-date risk1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One relay, no primary note
The chain is three links long and thins at every one: Goldman's note, an unnamed Chinese outlet, then Interesting Engineering. Nobody in that chain shows the model, and the 2025 baseline needed to interpret the headline percentage is absent — two defensible ways of inferring it from the same reporting land more than a tenth apart. What can be checked independently is the historical furniture: TSMC's 2018 ramp, ASML's DUV monopoly position, SMIC's 2023 Huawei part. The decade of forward numbers rests on nothing a reader can inspect.
Real datapoints, very small base
Today's evidence of actual advanced-node output is a single 7nm part shipped to Huawei in 2023 and a 92% shortfall — implying on the order of ten thousand wafers a month against demand more than ten times that. The headline 70% self-sufficiency figure is measured by volume, and the reporting concedes the value-based picture is far worse, which is precisely where advanced nodes sit. Domestic tooling has genuinely spread into implant, inspection and metrology, but at 26% of equipment spending it is not yet displacing the categories that gate 7nm.
The percentage flatters the wafer count
'Slash the gap to 34%' is true and misleading at once. Demand in this model grows off a base more than ten times larger than supply, so the shortfall shrinks in ratio while the wafers China still cannot make rise to 209,000 a month — roughly double the implied 2025 shortfall. Add a 2035 yield assumption below TSMC's present-day capability, which quietly costs another fifth in wafer starts, and a lithography dependency the same reporting calls the major obstacle, and the optimistic framing is carrying weight the arithmetic does not support.
A bullish desk, quoted approvingly
This is research with a position: the bank states outright that it remains positive on China semis capex, and the piece carries that line as a conclusion rather than a disclosure. The revision tells the same story numerically — a 2030 capex view lifted 79%, about $36 billion, in a single year, in the direction the desk is constructive on. The relay adds its own pull: an unnamed Chinese outlet with reason to amplify a self-sufficiency narrative, and an aggregator publishing a paywalled bank call as news.
Arithmetic checks out, sourcing does not
We are confident about the internal shape of the story — the 2035 balance is stated plainly, and the absolute gap widens under every baseline the reporting allows. We are much less confident about the inputs, since the 2025 volumes must be inferred, the lithography assumption is a single unquantified sentence, and no second outlet has touched any of it. Treat the direction as sound and each specific figure as provisional until the underlying note surfaces.