The small reactor still turns a big turbine.

On 28 September I stood in a meeting room in Vienna. I talked to reactor people about AI factories and data centres. The International Atomic Energy Agency (IAEA) had called the meeting on powering and cooling data centres with Small Modular Reactors. The reactors were on the agenda. The point that drew the most interest was supply. Even the turbine makers came up as a bottleneck. People and safety came up too.

The Vistergy Brief has spent two weeks on the rest of the picture. Governments from Brussels to Seoul have committed the money. Every route to firm power is booked. What is left is the part nobody photographs: the machines, and the people who build and fit them.

Look at two small reactor plants that already feed a grid. At Pevek, in the Russian Arctic, a floating plant carries two small reactors on a barge. At Shidao Bay, on China's east coast, a pebble-bed plant runs two small reactors on land. World Nuclear News describes what those two Chinese reactors drive: a single 210-megawatt steam turbine. The planned next version runs six reactors into one turbine of 650 megawatts.

The reactor went modular. The turbine did not.

Every reactor in commercial service works the same way at the far end. It makes heat, the heat makes steam, and the steam turns a turbine. A small reactor changes the front of the plant. It does not remove the heavy machinery at the back. The turbine, the generator and the grid transformer are all still there.

Gas plants end in a turbine too, and their queue is public. GE Vernova, the American maker, said in July that its agreements already run into 2031. That is the fastest route to firm power, and it is almost sold out.

Whether the turbines behind small reactors face the same wait is not something I can put a number on. That uncertainty is the point. In the room in Vienna, the turbine makers were named beside everything else a plant needs. A reactor can be licensed and a site can be ready. If the machine at the back is not, the plant waits.

This is the Still Dark problem arriving from a new direction. I wrote in July about capacity that is built and not yet switched on. Late equipment is one way that happens.

A late machine is not only a late machine. Its delay lands in commissioning, the least forgiving weeks of any programme. Every other trade is waiting to test at the same time. The delay squeezes the testing that proves the plant works. Then it squeezes the handover record that proves the testing happened.

An announcement never shows this. The reactor is licensed, the money is committed, and the site is cleared. The gap opens later, in a turbine hall, in a week nobody planned for.

My argument in Vienna was simple. A power purchase agreement only holds if the plant arrives when the contract says. That makes supply chain management part of execution, not something procurement handles afterwards.

In practice it comes down to three habits. Name the long-lead machines on the first day, not the day the design freezes. Reserve factory slots against a programme that may still move. And write the standards into the contract. Then the records that travel with each machine arrive in a form the next party can use.

None of this is new to anyone who has built a power station. What is new is the pressure. AI factories want firm power on a timetable few power stations have had to meet.

I also asked the meeting to recommend IAEA guidance on where a reactor programme meets an AI programme. The only worry was time. Guidance like that can take years. The room wanted it faster, without giving up any of the safety or licensing rigour. That speed can only come from evidence that is ready when a reviewer asks for it, never from lighter checks.

Two more constraints came up in the room, and no factory slot solves either. The first is people. The worry was a shortage of skilled people. The second is safety. The tech community wants to go fast. A reactor programme earns its licence by going carefully, one proven step at a time. A buyer in a hurry does not change that, and it should not.

So the title is a reminder, not a paradox. The reactor is going small, modular and repeatable. That is real progress. But the plant is still a supply chain that ends in one large machine, delivered once and tested under pressure.

Good intelligence can read the order books, the queues and the schedule. It can show where they collide before a contract is signed. It cannot forge a rotor, train a commissioning engineer or sign a safety case. People do that, in a factory and on a site, one delivery at a time.

The reactor shrank. The schedule did not.

Still Dark returns when an argument demands the page. The podcast carries the rest. Permit to Operate is where the people who live in this gap speak in their own voices.

If you build, operate, invest in, or regulate infrastructure anywhere in the world, this is written for you. Subscribe to Still Dark.

This newsletter lives in the gap between digital delivery complete and permit to operate. That gap is where value dies, and where it can be recovered.

I also co-author The Vistergy Brief at vistergy.com/archive. Satellite and geospatial monitoring, facility lifecycle intelligence, and standards architecture across LNG, nuclear, data centres, utilities and construction. Subscribe to both for the full picture.

The podcast is live. Permit to Operate is available at vistergy.com/pod, Apple Podcasts, and Spotify.

Permit to Operate.

David Ayeni

P.S. Members go deeper.

What you get, £12 a month or £120 a year (17% annual discount):

Be among the first members: Upgrade