Sovereign Tech
Sovereign AI, and the Europe that's building it
Sovereign AI is the sharp end of European strategic autonomy: the compute, models, cloud, and governance the continent is racing to own rather than rent. This is the definition, the policy that turned it into a budget line, the companies delivering it, and the live capital behind them.
The definition
What is sovereign AI?
Sovereign AI is a country or bloc's capacity to build, run, and govern artificial intelligence on compute, data, and talent it controls, without depending on foreign hardware, cloud platforms, or models that answer to another government's law. It is the difference between using AI and owning the conditions under which AI runs.
The phrase started as a slide in ministerial decks and has hardened into a procurement requirement. When a hospital network, a finance ministry, or a court system asks whether a model can be subpoenaed by a foreign authority, hosted on infrastructure outside its jurisdiction, or switched off by a vendor it does not control, it is asking about sovereign AI. The answer increasingly decides who wins the contract.
It is not autarky. No serious European policymaker thinks the continent will print its own GPUs or refuse American research. Sovereignty here means optionality: the ability to keep critical workloads running, and critical data governed by European law, even when the geopolitical weather turns. That is a narrower and more achievable goal than independence, and it is the one the money is now chasing.
The four layers of a sovereign stack
Compute
The chips, datacentres, and interconnect that training and inference run on. Europe imports most of its advanced silicon and rents most of its large-scale compute from US hyperscalers, which is the dependency every other layer inherits.
Models
Foundation and frontier models whose weights, training data, and alignment choices are controlled inside the bloc. Open weights matter here: a model you can self-host is a model no foreign vendor can revoke.
Cloud & data
Where workloads execute and where sensitive data sits, and crucially whose law reaches it. The US Cloud Act lets American authorities compel American providers to hand over data wherever it physically lives, which is the exact exposure sovereign-cloud tiers are written to close.
Governance
The ability to audit, constrain, and shut down AI systems already deployed. As enterprises wire agents into real operations, the off switch becomes infrastructure, not an afterthought.
Europe's case
Why the continent stopped asking nicely.
Europe runs on AI it does not own. The continent imports more than 80% of its digital products and services, the most advanced model APIs are American or Chinese, and the cloud regions those models run in are governed, ultimately, by foreign law. For a decade the response was a speech about strategic autonomy, which funds nothing and ages badly.
That changed on 3 June 2026, when the European Commission put technological sovereignty into legislation. The package bundles a Cloud and AI Development Act that restricts US hyperscalers from handling sensitive public data, a Chips Act 2.0 aimed at advanced semiconductor capacity, an open-source strategy, and an energy roadmap. “We want to be sure nobody has a kill switch,” Executive Vice-President Henna Virkkunen told reporters.
Read past the acronyms and the move is simpler than it looks. A continent that imports its digital stack just wrote itself a purchase order. CADA does not merely encourage European alternatives; it defines tiers of cloud sovereignty and requires the highest ones to be structurally independent of third-country law, which is something AWS and Azure cannot offer as currently built. That is not a ban. That is demand, with a legal deadline attached, and demand is the one input European deep tech has always been short of.
Either this is the moment European sovereign AI stops being a speech and starts being a budget line, or it is the same dependency, relabeled and re-funded at home. Possibly both. Either way, the companies that have to supply it are already raising, and we are counting them.
What the sovereignty package actually does
The procurement spec
CADA's top sovereignty tiers require EU ownership, EU-national personnel, and immunity from the US Cloud Act for sensitive public-sector workloads in healthcare, finance, and the judiciary. The private sector is untouched, so nobody is forced off Azure. But every public buyer on the continent now has a compliance reason, not merely a patriotic one, to find a supplier the incumbents cannot structurally match.
Chips Act 2.0
The successor act shifts from building fabs to stimulating demand for European-made chips and securing design capability the continent mostly lacks. Reports put an advanced 3nm facility near €30B; the headline target is €120B of investment by 2035. This is industrial policy measured in elections rather than quarters, but it is the demand signal compute-layer founders have been waiting on.
Compute, tripled
CADA sets out to triple EU datacentre capacity within five to seven years, roughly €200B in mostly private investment, so European AI can run on European infrastructure rather than route through US hyperscalers governed by US law. “Mostly private” is the load-bearing phrase: the Commission is naming a number it hopes the market will fund.
The energy footnote
Tripling compute means tripling demand for power, so the package pairs cloud and chips with an energy roadmap. It reads like a footnote and is not one. Sovereign infrastructure that cannot be powered cheaply and on the continent is a press release. The countries that solve clean energy will host the sovereign datacentres; the rest will keep renting.
Who's building it
The companies on the receiving end of the purchase order.
A policy can only ever write the cheque. What decides whether sovereign AI becomes real is whether the companies cashing it are any good. Grouped by the layer of the stack they defend, every one below is a European round we have tracked, with the reason it counts as strategic autonomy.
Foundation models & agents
1 companyAMI Labs
$1.03B · Seed · France
Yann LeCun's bid for a European alternative to US large language models.
Semiconductors, quantum & photonics
3 companieseleQtron
€57M · Series A · Germany
Trapped-ion quantum computing built in Germany, backed by the EIC Fund and NRW.BANK.
Eyeo
€40M · Series A · Netherlands
Next-generation photonic image sensors, a European semiconductor play, with imec and Invest-NL.
Wayve
€1B · Series D · United Kingdom
A homegrown autonomous-driving stack, backed in part by the British Business Bank.
Defense & dual-use
2 companiesComand AI
€32M · Series A · France
Sovereign command-and-control software for European militaries, with Saab backing.
Harmattan AI
$200M · Series B · France
France's first defense unicorn, autonomous drones, with Dassault Aviation on the cap table.
AI governance & security
2 companiesNeuralTrust
€17.2M · Seed · Spain
Governance and an off switch for the AI agents Europe is deploying.
Geordie AI
$30M · Series A · United Kingdom
Securing and policing autonomous enterprise AI agents, from London.
Energy & critical minerals
4 companiesMantle8
€31M · Series A · France
Drilling for natural hydrogen to loosen Europe's energy dependence, backed by Bpifrance.
Lithosquare
$25M · Seed · France
Hunting the critical minerals the energy transition and European supply chains need.
terralayr
€112M · Series A · Switzerland
Grid-scale storage hardening Europe's energy supply.
Cloover
$1.22B · Series A + debt · Germany
An operating system for European energy independence, $22M equity and $1.2B in debt.
Robotics & industrial autonomy
1 companyTHEKER
$85M · Series A · Spain
Industrial robotics built for European production lines, the continent's largest-ever robotics Series A.
The full live tracker, filterable by country, sector, and stage, sits below, alongside the investors writing the cheques.