Opening 0:00
Welcome to today's China report. Today it's trade, supply chains and industrial overcapacity, specifically, the machine that builds the capacity, set against a proposal to keep that machine's output out of America. Two reports, which is fewer than usual. It was a quiet month in the literature, and that's worth saying rather than padding around. One of the two isn't squarely on topic: Zongyuan Zoe Liu's essay for the Hoover Institution's China Leadership Monitor is about financial architecture, not trade. It earns its place because it describes the credit plumbing that funds the factories the second report wants to fence out. I'll be careful not to make it say things about overcapacity it never said.
Hoover on China's mission-driven finance 0:52
Liu's essay ran on the thirtieth of August in China Leadership Monitor, published by Hoover. She's at the Council on Foreign Relations and she holds a CFA charter, which shows, this is the most financially literate thing I've read on the subject in a while.
What it says. Since about 2020, Beijing has rebuilt its financial system around technological self-reliance, treating it as a defence against being cut off. Not by nationalising lending. By building what she calls a capital allocation circuit: centralised strategic direction, decentralised funding decisions, and selective socialisation of risk.
How it argues. The mechanism is the whole essay, so stay with me. The state doesn't pick the loans. It picks the categories, sets eligibility, then bends the economics so banks want to lend inside them. The central bank refinances qualifying technology loans at one point seven five percent while market funding sits around two and a half to two point eight five. Her careful phrase is that this "can be large enough to change the relative attractiveness of eligible technology loans at the margin." Then risk-sharing: state guarantees cover up to eighty percent of credit losses, but the lending bank must retain at least twenty. She treats that retention as load-bearing, it's what stops this being a blank cheque and keeps the bank screening. Add the STAR Market taking pre-profit firms, a bond technology board, technology insurance. For the conclusion to hold, banks have to keep behaving like banks inside the fence.
What it rests on. Almost entirely Chinese official sources, central bank announcements, State Council notices, regulator rules, Party press. Essentially no independent data, no outside scholarship, no interviews. And the outcome numbers are thin. One point five trillion yuan of lending under the unified relending facility by April, of which two hundred and eighteen point eight billion was classed as technology-innovation lending. Thirty-four thousand four hundred companies, a hundred and forty billion yuan, reported financing costs under five percent. No default rates. No productivity figures. No capacity built.
What it predicts. Nothing with a date on it. The essay treats twenty twenty-six as accomplished fact and closes on whether the design will produce genuine breakthroughs remaining uncertain. That's honest, and it's also a forecast no one can ever catch out.
Where it's weak. Grant the strong part: the twenty percent retention rule is a real observation, and she's right that cheap funding can't manufacture collateral or shorten a commercialisation cycle. She says so herself. But the entire claim is that this circuit moves money to firms banks wouldn't otherwise fund. Every outcome number she prints is a flow number, how much moved, and never a destination number: who got it. Except once. On the sci-tech innovation bond market she does measure destination, and ninety-four percent of issuance came from state-owned enterprises. To her credit she flags it. What she doesn't do is let it discipline anything else. The one place she checks where the money landed, it landed on the state firms the system was built to bypass, and the unmeasured channels get the benefit of the doubt anyway.
FDD on sectoral import controls 4:55
Craig Singleton's written testimony for the Foundation for Defense of Democracies, delivered to Congress on the sixteenth of September.
What it says. American technology-security policy has spent a decade denying China access to American capability, export controls, entity lists, while Chinese firms quietly took dominant positions inside supply chains pointed at the American market. Batteries, chips, robotics, power systems. His word is beachheads. The remedy is sectoral import controls: not banning companies, banning categories.
How it argues. Three steps. One, Beijing's model is innovation-to-scale, the fifteenth Five-Year Plan turns research into manufacturing clusters fast, and the stated goal is to tighten everyone else's dependence on China while reducing China's own. Two, it's working: Chinese firms took roughly seventy-eight percent of general-purpose embodied robot shipments last year, all ten of the largest battery-energy-storage cell suppliers in the first half of this year, about eighty percent of solar and battery inverter manufacturing capacity. Three, and this is the real argument: those positions are durable in ways tariffs don't touch, because the hardware keeps phoning home. Firmware updates, cloud services, remote control. So entity-by-entity enforcement always loses. Rename the subsidiary, start again.
What it rests on. Commercial market research, mostly, Omdia for robots and displays, Yole for LiDAR, Benchmark Mineral Intelligence and Wood Mackenzie for storage, the International Energy Agency for inverters, Gartner for PCs. Proprietary, so you can't check it, but it's what the industry itself buys and it's the right instrument for market share. The semiconductor claim is different. Thirty-three percent of mature-node capacity, up from nineteen percent in twenty fifteen, cited to the U.S.-China Economic and Security Review Commission. I went looking for where that number actually comes from. It's the Semiconductor Industry Association's, the American chip industry's own trade body, filed into a Section 301 trade case where its members stood to gain from restrictions. Then I went looking for a count that didn't start in Washington. Korea's KIEP, a government institute sitting inside this supply chain, publishes on Chinese semiconductors, and its work doesn't break capacity out by node. I couldn't find an independent measurement. That doesn't make the figure wrong. It means the most policy-consequential number in the testimony has one interested source.
What it predicts. Chinese firms taking nearly half of all new mature-node capacity over the next several years, no clock on it. The dated numbers are sharper, but they aren't his: humanoid deployment at ten-thousand-unit scale by the end of this year, pillar industries above ten trillion yuan of output by twenty thirty. Those are Beijing's plan targets, restated.
Where it's weak. The beachhead framing is good and the firmware point is the best thing in it, a genuine gap in any tariff-and-entity-list approach. But the testimony proposes excluding Chinese batteries, inverters, displays and mature-node chips from the American market and never once says what that costs or whether substitutes exist at scale. No price estimate, no transition timeline, no line on allied capacity. And his own evidence is why that's fatal. He reports China holding roughly eighty percent of inverter manufacturing, the IEA's number, from July, and all ten of the top ten storage cell suppliers. You can't cite eighty percent as the proof of danger and then make exclusion the baseline without saying where the other twenty percent scales from. The recommendation needs a number his own market data makes very hard to produce.
What to watch 9:34
The nearest thing that resolves is Beijing's humanoid robot target, ten-thousand-unit deployment scale by the end of this year, about three months out. Miss that badly and the innovation-to-scale story is slower than the testimony assumes. After that, watch what share of Chinese technology-innovation bond issuance goes to private firms instead of state-owned ones. Liu's ninety-four percent is the only destination number either report gives us, and movement there would tell you the credit machine is doing what it claims. Notice, finally, that these two never meet: one says the money is flowing and can't show where it lands, the other says the output is everywhere and can't say what replacing it costs.