In partnership with
In Brussels, a €200 billion programme aims to make Europe an AI continent. In practice it is a plan to site and power five giant compute plants. In London, the first five AI Growth Zones come with reserved grid capacity, not just tax relief. In Abu Dhabi and Seoul, the AI budget and the power budget are the same document. Sovereign AI is written as a technology policy. It is delivered as an energy one.
The European Commission's InvestAI initiative mobilises €200 billion, about £171bn ($227bn, €200bn). At its core is a £17bn ($23bn, €20bn) fund for up to five AI gigafactories. Britain's first AI Growth Zone cohort is set to deliver £28.2bn ($38bn, €33bn) in investment and over 15,000 jobs. Both programmes stand or fall on megawatts.
The money is committed. The question is who can build fast.
The opening is a budget line. InvestAI's £171bn buys five sites of 100,000 chips each
Follow the money to the substation. The EIB and the Commission agreed in December 2025 to finance up to five gigafactories. Each will run about 100,000 of the most advanced AI chips, four times today's AI factories. The call drew 76 submissions covering 60 sites across 16 member states. A site with 100,000 accelerators is a power station's customer before it is a computer. Europe's AI sovereignty is therefore decided at the grid connection, not the model.
Britain reserves the grid. £28.2bn across five zones, with grid capacity held for them
London made the energy link explicit. The five AI Growth Zones are backed by up to £5m ($6.7m, €5.9m) each in targeted funding. The real instrument is a package of reforms. It reserves grid capacity, streamlines planning and cuts electricity costs for the zones. North Lanarkshire, the fifth zone, pairs 500 MW of data centre capacity with 1 GW of private-wire renewables, for £8.2bn ($11bn, €9.6bn). The state is not funding the chips. It is buying the queue position.
The Gulf and Korea write one document for compute and power
Two regions skip the separation entirely. Abu Dhabi's Stargate campus builds nuclear, solar and gas alongside its data halls. The energy policy and the AI policy are the same site plan. South Korea's AIDC Alliance targets 18.4 GW of AI capacity by 2035 with state financing. Its own grid operator is named as the binding constraint. The United States runs the private version. Hyperscalers have signed up to 13 GW of nuclear agreements, with the largest terms undisclosed. In each case the programme's success metric is delivered gigawatts, not trained models.
Your national AI programme has funding and a site. What decides whether it is built on time?
Money and megawatts are necessary. The standards decide the pace
Capital secures the site and power secures the load. Neither sets the build speed. That is decided by whether every party can exchange information without translation. The public standards spine does that work. ISO 19650 governs information management across a built asset's life. ISO 55000 governs how the asset is managed once it runs. ISO 23247 frames digital twins for manufacturing. ISO 18136 frames the nuclear digital ecosystem. CFIHOS specifies what information is handed over, and CII's Advanced Work Packaging defines the data that construction needs. Fluency in that spine is a build-speed advantage, not paperwork.
ISO 18136 published in October 2025. Nuclear now has a digital spine of its own
The newest strand matters most for firm power. ISO 18136-1:2025 was published in October 2025 by the industrial data committee ISO/TC 184/SC 4. It sets a shared framework for the nuclear digital ecosystem across every reactor-related facility. It spans every discipline, from civil to instrumentation. Beside it, ISO 23247 frames digital twins of manufacturing elements. Together they let digital twins of nuclear plants be built to a common structure. A regulator can inspect one twin's evidence the same way as another's. That is what shortens licensing, and licensing is the long pole in nuclear.
Handover is where sovereign programmes lose years. CFIHOS and AWP close the gap
Most schedule is lost between parties, not inside them. CFIHOS is the Capital Facilities Information Handover Specification. It exists so operators, contractors and suppliers hand over the same information the same way. CII's Working Group 19-01 built the data requirements for Advanced Work Packaging (AWP) because member firms found no common definitions. A gigafactory in Europe or a growth zone in Wales inherits this gap. Only the contract closes it, by naming the standards. The fast builders write them in from the first drawing.
Governed intelligence: intelligence informs, expertise decides
Here is where AI enters its own supply chain. The standards spine makes an asset's information visible. Visibility is not the same as decision-worthy. A model can read ISO 55000's requirements against a plant's records and show where they diverge. It cannot decide whether the divergence is acceptable. That judgement belongs to the operator, the regulator and the engineer. Our analysis calls the layer between the two governed intelligence. It sits between the standard and the decision, and it makes the evidence inspectable. Intelligence informs. Expertise decides.
Validation is coming: a joint USPI report for newsletter members first
Claims about standards need proof. Paul van Exel convenes the ISO working group behind ISO 18136. He and Youngsoo Jung are preparing a joint validation report with USPI. It examines how the standard is applied in practice. It will reach Facility Intelligence Newsletter members before it is published more widely. It covers the taxonomy, not any one vendor's implementation. Readers who want to test the build-speed thesis against a worked case will see it there first.
The investment implication: underwrite standards fluency as schedule risk
The capital case follows. Two programmes with the same money and the same power can finish years apart. The difference is whether information moves between parties without rework. Lenders price schedule risk; standards fluency is schedule risk in disguise. A programme showing ISO 19650 management, CFIHOS handover and AWP data from day one has removed a class of delay. One that cannot is carrying it unpriced. Korea's buildout needs over £514bn ($683bn, €601bn) by 2035. It and Europe's £171bn ($227bn, €200bn) will be judged on delivered gigawatts. Standards are how the years are saved.
The bottom line
Sovereign AI is an energy policy in a technology policy's clothing. Brussels, London, Abu Dhabi and Seoul have each committed the money and, increasingly, the megawatts. What separates the programmes that finish from those that slip is fluency in the public standards spine. That spine lets every party build to the same structure. Intelligence can read that spine. Expertise still decides. The standards decide who can build fast.
Next week: The handover gap had a name in January. By September it has a budget line.
Analytics on Live Data Without Leaving Postgres
When analytics on Postgres slows down, most teams add a second database. Then come the pipelines, the sync jobs, and a copy of your data that's always a little behind.
TimescaleDB takes a different approach: extend Postgres instead of splitting away from it. Hypertables partition your data automatically as volume grows. Hypercore compression cuts storage up to 95%. Continuous aggregates keep dashboards live without re-querying everything.
CERN runs Postgres this way for sensor data from the Large Hadron Collider.
No split architecture, no pipeline lag, no new query language to learn. Same SQL, same drivers, same tools.
Start on Tiger Cloud and get $1000 in credits.