The Ceiling Height

I. The announcement

In January 2026 the Technical University of Munich founded the Munich Advanced-Technology Center for High-Tech AI Chips, MACHT-AI for short. It is based at the Chair of AI Processor Design, its founding director is Hussam Amrouch, its key partner is TSMC, and it is funded by the Bavarian Ministries of Economic Affairs and of Science. Its stated aims are training in TSMCs FinFET technology and several successful tape-outs per year at the 16 and 7 nanometre nodes.

Bavarias Minister of Science, Markus Blume, comments that AI chips are the currency of the future and that, in order to remain competitive, independent and sovereign, the state intends to train its own talent and develop its own technologies.

On 4 February 2026 TUM announced the EUs first AI chip in 7-nanometre technology: a neuromorphic design based on the open RISC-V instruction set architecture, designed by Amrouch in accordance with TSMCs industry standards. University President Thomas Hofmann says that Amrouch impressively demonstrates how dependence on Asia and the USA can be reduced and how Europes technological self-determination can be strengthened.

The same release states that the group plans at least three new designs per year, to be manufactured from 2028 onward by the European Semiconductor Manufacturing Company in Dresden.

II. What is possible in Dresden

The last two sentences stand side by side and suggest a connection that does not exist.

The 7-nanometre chip was manufactured by TSMC in Taiwan. The ESMC fab in Dresden — a joint venture of TSMC, Bosch, Infineon and NXP, more than ten billion euros of investment, five billion of it federal money — will, by its own account, run two process technologies: 28/22 nanometre planar CMOS and 16/12 nanometre FinFET. A 7-nanometre design cannot be manufactured there. What can be built in Dresden from 2028 are designs at 16 or 12 nanometres.

According to a report by the Dresden trade publication Oiger of 16 February 2026, the reason lies not in the equipment but in the building: the ASML EUV scanners required for structures well below ten nanometres, and the associated systems, would not physically fit — above all with regard to room height. This statement rests on a single source. The fact it explains, by contrast, is stated in ESMCs own documents.

According to a further single report from May 2026, the DUV tools freed up by the modernisation of Fab 15 in Taiwan are being shipped to Dresden, while building 15B there, already equipped with EUV, is being converted to the 3-nanometre node.

One may regard this arrangement as sensible. Mature nodes are the backbone of automotive and industrial electronics, and that is where European demand lies; it was planned this way deliberately. But it is the opposite of what the word sovereignty asserts in the accompanying quotations.

III. Sovereignty as a specification

In these texts the term is used as a sentiment. One wants to be sovereign, therefore one is on the way there. Treated instead as a specification, it becomes testable.

A specification for sovereignty in semiconductor manufacturing would have to state, for every link in the chain, who controls it and what happens if it is interrupted. The links are: design tools, circuit design, process technology, lithography, process equipment, wafers, process chemicals, photoresists, masks, fabrication, packaging, test.

Working through that list produces a picture that matches neither the alarmism nor the announcements.

IV. What exists and what does not

Present and under European control. ASML in the Netherlands holds the monopoly on EUV lithography. Zeiss SMT supplies the optics required for it and is not replaceable. TRUMPF supplies the drive lasers. Siltronic and Soitec make wafers, ASM International and Aixtron deposition equipment, Merck and BASF process chemicals. Research at imec, Fraunhofer and CEA-Leti is world class. Infineon, STMicroelectronics and NXP design and manufacture in their own plants. For automotive electronics, industrial applications, power semiconductors and sensors, Europe is largely self-sufficient.

Not present. Leading-edge manufacturing: no European company can do seven nanometres or below. Design tools: Cadence and Synopsys are American; Siemens EDA is owned by a European group but developed in the USA and subject to American export control. Memory: nothing since the end of Qimonda in 2009. Advanced packaging: Taiwan and the USA. EUV photoresists: Japan. Test equipment: Japan and the USA.

So the picture is not dependent. It reads: strong at the front of the chain, self-sufficient at the mature nodes, absent at the leading edge.

V. The calculation already exists

In April 2025 the European Court of Auditors published Special Report 12/2025. Its subject was what EU industrial policy has achieved for the strategic autonomy of the microchip industry.

The result, in figures: to reach the twenty per cent share of global production set in 2022 by the year 2030, the EU would have to quadruple its production capacity. No such pace is currently discernible in any area. Of the 86 billion euros foreseen up to 2030, the Commission contributes five per cent, that is 4.5 billion; the remainder is to come from member states and industry. The worlds leading manufacturers invested 405 billion euros over the past three years.

The ratio of 4.5 to 405 is not a nuance. It is the matter itself.

The responsible auditor, Annemie Turtelboom, summarised it thus: the EU urgently needs a reality check on its strategy; the Chips Act is not an option, its revision is essential. In January, seventeen EU countries were affected by supply restrictions on chips.

The report has been public since April 2025. The founding texts of January 2026 read as though it did not exist.

VI. Two goals, one word

The reason is not ignorance. It is that two different goals are denoted by the same word in this debate.

Sovereignty means: we can make everything ourselves. Resilience means: an interruption does not kill us. These are different tasks with different price tags.

Resilience for Europes industrial requirement is achievable. By far the largest part of European value creation — vehicles, machinery, energy technology, medical technology, automation — runs on nodes between 28 and 90 nanometres, and there Europe is already self-sufficient. What is missing is packaging, test, second sources for chemicals and photoresists, and stockholding of critical components. That is a task for purchasers and engineers, in the single-digit billions, over three to five years.

Sovereignty at the leading edge is something else. A single fab today costs twenty to thirty billion euros; Intels Magdeburg project was budgeted at thirty billion and was cancelled. But the fab is the cheap part. The expensive part is process knowledge, and it cannot be bought: Intel had unlimited funds and still lost a decade at the ten and seven nanometre nodes.

The only reliable yardstick is China: roughly fifteen years, several hundred billion, state direction of the entire economy — and the result is seven nanometres by multiple patterning at moderate yield, plus, by now, a domestic DUV scanner. EUV not yet.

The order of magnitude is therefore: fifteen to twenty years, several hundred billion euros, and behind for most of that time.

That sentence cannot be announced. It contains no deadline within a parliamentary term, no ribbon to cut and no result the announcer will still be in office to see. A centre with a partner and a superlative, by contrast, is available immediately.

VII. No charge against the centre

It would be wrong to turn this into a charge against MACHT-AI.

A training centre for chip design at advanced nodes closes a real gap. European universities design predominantly in mature technologies because FinFET process design kits are expensive and access-restricted. Tape-out is moreover a hard criterion: the chip works or it does not, and that shows in silicon, not in a report. A centre that lets itself be measured by that is something other than one that publishes roadmaps.

Amrouch himself names the gap correctly. He says it will be crucial in future to cover all stages in Germany and Europe — from training chip specialists through technological development to chip manufacturing. That is a description of a goal. In the ministerial quotations beside it, it becomes a state of affairs in the present tense.

The difference between the two is the subject of this text.

VIII. What is missing is a calculation

No strategy is missing, and no further centre. What is missing is a balance sheet.

One line per link in the chain, and in the columns: who controls it, how long a substitution takes, what it costs, what happens after twelve months of interruption. That is not a strategic exercise but a bill of materials with one additional column. Every design engineer who releases an assembly does this for their suppliers.

Such a balance sheet could be produced in a few months, and it would be the basis for any sensible allocation of funds. It is not published. The reason is foreseeable: it would show that the money is not flowing to where the dependency sits, and that the critical links are other than the ones being talked about.

As long as it is missing, sovereignty remains a word without a unit of measurement.

In Dresden it now has one. It is measured in metres, it was fixed before construction began, and it cannot be changed afterwards.