China builds machines, France sells shares

Three days separate two news items from late July 2026. On the 28th, Reuters reported that China has begun volume production of its own immersion DUV lithography machines. On the 31st, Bpifrance announced the sale of 2.5% of Orange’s capital for €1.1 billion. Nobody connected the two, and that is precisely the problem.

A word about the object itself, because everything that follows turns on it. Lithography is the step where a circuit pattern is printed onto a silicon wafer by projecting light through a mask. The shorter the wavelength, the finer the features. DUV — deep ultraviolet — is a 193-nanometer laser. Immersion means slipping a film of ultrapure water between the final lens and the wafer: water has a refractive index of 1.44, which brings the effective wavelength down to roughly 134 nanometers. A water film less than a millimeter thick, held stable beneath optics sweeping across the wafer at several hundred millimeters per second, and there you have a decade of shrink gained without changing the light source.

Every chip made today goes through lithography, often thirty or forty times over, and everything below 40 nanometers goes through immersion. An immersion scanner weighs tens of tons, contains tens of thousands of parts, costs between fifty and eighty million euros, and exactly one company in the world truly knows how to build one: ASML, in Veldhoven, in the Netherlands. Nikon and Canon survive at the margins. This is the single chokepoint Washington closed to China in 2023. It is the chokepoint Shanghai has just reopened on its own terms.

What happened in Shanghai

The story has been circulating under two different names, which was enough to muddle its reception in the French press. The Information first pointed to Shanghai Yuliangsheng Technology, the startup affiliated with SiCarrier, the equipment maker backed by Huawei, whose prototype SMIC has been testing since September 2025. Reuters then named the actual prime contractor: Shanghai Aishengna Electronic Technology Group, a state-owned company founded in August 2023 with registered capital of 7 billion yuan, roughly a billion dollars, held by two public shareholders, Shanghai Electric Holding and a subsidiary of Shanghai International Trust.

Aishengna has no website. It has never disclosed anything about its operations. It shares an address with Yuliangsheng and has absorbed that company’s teams along with those of SMEE, the country’s veteran lithography outfit. Put plainly: Beijing gathered every engineer in the country capable of focusing a 193-nanometer beam through a film of water, placed them inside a structure with no shop window, and put a billion dollars on the table.

The announced volumes are modest. Five machines delivered in 2026, roughly twenty in 2027, going to the foundries SMIC and Hua Hong and to memory maker CXMT. ASML, by comparison, ships about 130 immersion systems a year. The ratio is one to twenty-five. On paper, there is no story here.

Except that Asian markets fell more than 10% in Seoul and 4% in Tokyo in the days that followed. The markets, at least, understood.

Why five machines matter more than a hundred and thirty

A technology embargo has no proportional value. Its value is binary. As long as no substitute exists at any price, the blockade is total and the approved supplier dictates terms. The moment a substitute exists, however mediocre, however slow, however limited to five units a year, the embargo stops being a wall and becomes a delay. And a delay can be measured, amortized, planned around.

The Chinese machine targets 28 nanometers in a single exposure, and 7 nanometers through multipatterning at the cost of degraded yield. This is not EUV, it is not TSMC, it is not the silicon inside the latest AI accelerators. But it is exactly where the overwhelming majority of real-world chips are made: microcontrollers, power management, sensors, memory, networking, automotive. DUV lithography is not the prestige technology. It is the volume technology.

I am doing nothing more here than replaying a theorem I have already laid out on different ground: good enough, fifty times cheaper. American export controls were meant to strangle Chinese AI. Instead they taught it frugality, and frugality became its pricing weapon. What played out with DeepSeek and Qwen is playing out again today in a Shanghai cleanroom, with the considerable difference that we are no longer talking about weight files but about steel, optics, and precision mechanics. Inverse proportionality has crossed the border out of software.

The threshold that has just been crossed, then, is not one of performance. It is one of existence. And it does not get crossed back.

The embargo worked exactly as intended, for Beijing

This is the passage our ministers should read twice.

An entrant into the lithography market normally faces three insurmountable obstacles: a technically superior incumbent, customers with no reason whatsoever to gamble on an immature tool, and a market where a mistake is paid for in billions of dollars of lost yield. Export controls eliminated all three at once. They created guaranteed captive demand, they made the risk of adopting the local tool smaller than the risk of doing nothing, and they handed the Chinese state a political justification for pouring unlimited money into an industry that economic calculation alone would never have brought into being.

The most telling detail lies elsewhere. Barred from selling its best models in China, ASML is reported to have raised prices on its older generations, the only ones still cleared for export. Translation: the Western supplier personally widened the price gap that its local competitor is now walking through. As for the MATCH Act, introduced in April 2026, it extends restrictions to the servicing and technical support of machines already installed. In other words, it tells Chinese fabs that their existing fleet is a hostage. You could not design a better sales pitch on Aishengna’s behalf.

It is the same mechanism I described when I argued that the European auto industry is not dying of China but of its own protection. A barrier does not protect the industry it claims to defend. It sedates the one standing behind it and builds muscle on the one standing in front. And let no one put words in my mouth: I have granted Trump a measure of credit on tariffs, on the condition that they be understood as an openly acknowledged instrument of leverage with a defined expiration date, and not as industrial policy by proxy. An embargo on a technological chokepoint is not a tariff. It is a work order issued to the other side’s engineers.

Let me be clear: this is not a plea for central planning. The Chinese lithography program is twenty years of SMEE and unkept promises, financed at a loss. What changed the game was not the subsidy. It was the removal of the alternative. Prohibition accomplished what billions of yuan had failed to do, which was to make the domestic product the preferable one. This is an Austrian lesson in its purest form, and it is a cruel one. You cannot plan other people’s ignorance. Every intervention produces its unintended consequences, and those consequences are sized precisely to match the arrogance that preceded them.

Anyone who assumes the climb to the high end will take twenty years should revisit what I wrote about a robot vacuum manufacturer capable of unveiling a 1,876-horsepower hypercar. Chinese vertical integration does not respect our calendars.

What we do not yet know

Honesty requires stating the limits. The reporting rests on an anonymous source and a paywalled outlet. No throughput figures, no overlay data, no pilot-line yield has been published. Actual production at SMIC is announced for 2027 at the earliest, and “at the earliest” in this industry generally means later. Domestic EUV is at the prototype stage, which at that level of complexity means a decade.

Above all, there remains the real battlefield, the one nobody discusses because it has no photogenic cleanroom: the supply chain. A lithography machine is not a single object. It is an assembly of monopolies. The precision optics, which must hold tolerances measured in a handful of atoms across lenses weighing hundreds of pounds, come from Zeiss, and it is no accident that ASML eventually took a 24.9% stake in Zeiss SMT after buying Cymer, its light source supplier. ASML did not merely build the machine. It bought its own chokepoints, one after another, over twenty years. That is what an industrial strategy looks like.

The second lock is chemical. Photoresists, and particularly those formulated for ArF immersion, are a near-monopoly held in Japan, where a handful of names, JSR, Tokyo Ohka Kogyo, Shin-Etsu, and Fujifilm, account for most of the global market. In November 2025, Japan’s METI added a dozen critical materials, high-end resists among them, to its export control list for specifically named Chinese companies. Reuters notes, for that matter, that critical components of the Aishengna machine are still imported from Japan.

The lesson cuts sharply, and it cuts both ways. The scanner is only the visible part. Tokyo can remind Beijing, with a single administrative signature, that building the machine is not enough if you have neither the glass to see with nor the chemistry to print with. Except that this is exactly the reasoning ASML was relying on five years ago.

None of which cancels the central fact. A machine that exists badly is worth infinitely more than a machine that does not exist at all.

Meanwhile, in Paris

On July 31, Bpifrance sold 66.5 million Orange shares at €16.57 apiece, 2.5% of the company, for roughly €1.1 billion. The press release speaks of “active portfolio management” and a “gradual asset rotation” strategy. The French state and Bpifrance together retain 20.4% of the capital and 27% of the voting rights, and their board representation is unchanged.

Note the elegance of the arrangement. You sell the economic stake and keep the political one. Double voting rights mean that 20% of the capital carries 27% of the governance, and the board seats do not shift by an inch. The state steps back from the risk and stays at the wheel. For minority shareholders this is the worst of both worlds: they fund a company whose direction continues to be arbitrated by an actor who is reducing his own exposure. Anyone who has forgotten what political stewardship of a national telecom operator produces can revisit Thierry Breton’s trajectory from France Télécom to Atos.

I am not asking the state to hold on to Orange. I have always maintained that it has no business owning a telecom operator, and that a full exit, openly decided and openly announced, would be good news. What sets my teeth on edge is the salami slicing. Selling in 2.5% increments and calling it a strategy, banking a billion to close a budget nobody dares to reform, and continuing to exercise control that is no longer being paid for. We have seen this film before: in 2005 the state unloaded 90% of the national motorway network for €14.8 billion, and the French are still paying that bill at every toll booth.

Above all, look at what becomes of the money. A billion euros vanishes into the bottomless barrel of a state whose debt trajectory is what it is, and nothing will remain of it. On the other side, a billion dollars went into a company with no website in order to learn how to build a machine. I do not endorse the Chinese method and I do not believe it could be replicated here. But I note that for an identical sum, one party bought productive capital and the other bought three months of cash flow.

The objection will come, and on the facts the objector will be right, that Bpifrance is not throwing this money out the window: the Deeptech plan and French Tech Souveraineté exist, hundreds of young companies live off them, and a share of these divestments does flow back into domestic venture capital. I grant that readily. But it is the nature of that reinvestment that is the problem. Taking minority positions in three hundred companies is planting saplings in a field you have forgotten to irrigate. A few will survive, most will be acquired before they bear anything, and none will ever cast shade. Shanghai did the opposite. One object, one industrial chain, every team in the country inside a single structure, a billion dollars, and no communications strategy. This is not a question of resources, or even of doctrine. It is a question of concentration, and our chronic inability to concentrate is legible both in the deeptech scattershot and in the determination to keep a lost cause on life support.

Sovereignty is not a line in a portfolio

The word “sovereignty” has become a verbal tic. It gets attached to a cloud running on an American hyperscaler, to a public procurement order for Microsoft licenses, to a minority stake in a telecom operator.

Sovereignty is not holding securities. It is holding a chokepoint. The Netherlands does not carry weight in the negotiation between Washington and Beijing because it has a handsome public portfolio. It carries weight because one company on its soil manufactures an object nobody else knows how to build. A single mandatory point of passage is worth a thousand shareholdings, and China already demonstrated as much with rare earths: the day the holder of the chokepoint decides to squeeze, there is nothing left to negotiate.

Europe, for its part, still believes a chokepoint can be purchased with a budget line and a ribbon cutting. NanoIC and its €700 million stand as the monument to that belief, set against the $165 billion TSMC has committed to Arizona alone. France, meanwhile, holds real assets it has never consolidated: Soitec and its SOI substrates, Air Liquide and the ultrapure gases without which no fab runs, the Grenoble ecosystem built around CEA-Leti. Minority pieces, fragile ones, for which I have never once heard a plan worthy of the name. Whereas on the Orange line of the state’s portfolio, there is a press release and a team congratulating itself. It is the same old preference for regulating our decline rather than building our future.

That is what this week in July is telling us. One country spent the decade learning to build the tool because it had been refused the tool. Another spent the same decade selling off, in small pieces, what it still owned, and calling it active management.

One builds. The other sells. Ten years from now, nobody will remember where Orange closed on July 31, 2026, and everybody will know who had the machine.