Still Dark #1: AI factories go up on time...

I have spent two decades working on infrastructure projects. Oil, gas, chemicals, nuclear, data centres, utilities and construction. Stone and Webster, Technip, Bechtel, Bentley Systems. Four continents. Hundreds of facilities.

Everywhere I go, the same problem is waiting at the end of the project. The facility is finished. The engineering is "done", construction is "done done" and commissioning chaos ensues pre-turnover...

I invested in Vistergy to fix that. Along the way, I realised the pattern is bigger than any single project or sector. It is structural. And not many are writing about it honestly.

So this is Still Dark. A weekly newsletter about the gap between digital delivery and operations. The 12 to 18 months after the facility is "finished" where nothing works yet. Real data, real sites, real standards. No theory. No product pitches. Just the problem, why it persists, and what is changing.

This is a pilot. I am running eight editions to see if this resonates. If it does, it becomes permanent. If it does not, I will have learned something. Either way, I would genuinely value your feedback. Reply, comment, disagree. Tell me what you want to read about (use the comments feature). This is as much yours as it is mine.

Let's get into it.

What if the most expensive problem in infrastructure has nothing to do with construction?

£393bn ($500bn, €470bn) is committed to AI factory construction through 2028. Stargate. Google. Microsoft. Meta. Amazon. 50+ facilities across Texas, Tennessee, the UK, Singapore, Germany, the Nordics, and beyond.

The buildings are going up on time. First phases are completing ahead of schedule.

Then the facilities sit dark for 12 to 18 months.

Not because of engineering. Because of documents.

A single 100 megawatt (MW) data centre generates more than 30,000 documents during construction. PDF. Building Information Modelling (BIM). Computer-Aided Design (CAD). Excel. Revit.

Every discipline uses different naming conventions. Every subcontractor follows a different schema. Every geography enforces different standards.

Electrical documentation alone exceeds 5,000 files governed by 15 to 25 standards simultaneously:

None of them talk to each other.

The facility cannot operate until they do. So engineers sort files manually. Record by record. Standard by standard. Twelve to eighteen months of it.

Cost per facility: £4m to £12m ($5m to $15m, €5m to €14m).

Construction teams spent years compressing the build. Then handover takes the same duration they just saved.

£79m ($100m, €95m).

That is the daily cost of commissioning delay for a single one gigawatt (GW) AI factory. Lost compute revenue. Every day the facility sits dark.

Across the industry build rate, handover bottlenecks represent £43bn ($55bn, €52bn) in annual opportunity cost.

The hyperscaler that commissions first wins compute capacity allocation from OpenAI, Anthropic, and Google DeepMind. Speed to commission is not an operational metric. It is competitive advantage.

And the constraint is not engineering. It is information.

This is not theoretical. This week, AtkinsRealis announced a collaboration with NVIDIA to accelerate deployment of nuclear-powered AI factories using Omniverse and OpenUSD. A world-class nuclear engineering company and the world's leading compute platform company, designing facilities as 3D digital twins before breaking ground.

The construction and simulation capabilities are converging. The question is what happens after the digital twin is built and the physical facility is complete. Who ensures 30,000 documents from engineering flow into operations without 18 months of manual remediation? The digital twin models the facility. It does not model the handover.

That gap is where this newsletter lives.

This is the question I hear in every conversation, across every sector. The answer surprises people.

It is not a technology problem. The standards exist:

These have been developed over decades by some of the most experienced engineers in the world.

The problem is that a single facility must comply with 15 to 25 of them at once. Each was developed independently. Each assumes it is the primary schema. And each conflicts with the others in ways that only surface during commissioning.

Fire safety conflicts with cooling. Security conflicts with accessibility. Efficiency conflicts with reliability. The electrical classification differs from the mechanical classification, which differs from civil works.

No single engineer holds all of these in their head. No single system harmonises across all of them.

So it happens manually. And it takes 12 to 18 months. Every time.

If you work in healthcare, you already know. 50,000+ documents per renovation. Semiconductor fabs exceed the same number. Gigafactories push past 100,000. Nuclear plants generate documentation that must remain traceable for 60 to 150 years.

Every sector that builds complex infrastructure hits the same wall. Construction accelerates. Handover stalls. The gap between completion and commissioning is where value dies.

The common thread is not volume. It is the absence of an architecture that connects standards, data, and systems across the full lifecycle.

The data exists. The standards exist. The tools exist. What is missing is the intelligence layer between them.

That is what I work on. Across all of these sectors. Simultaneously.

Edition 2: 30,000 files. One site. Nobody can find the wiring diagram.

Edition 3: One switch fails. £7m ($9m, €8.5m) gone. Every hour.

Edition 4: 18 months of handover. One team did it in four weeks.

Edition 5: 25 standards walk into a facility. None of them agree.

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

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, covering satellite and geospatial monitoring, facility lifecycle intelligence, and standards architecture across LNG, nuclear, data centres, utilities and construction. If you want the full picture, subscribe to both.

David Ayeni / Founder, Vistergy Ltd