The 5M EU and 0.85M UK Firms Legally Barred From Public-Cloud AI: Sizing the Real Wedge
A grounded count of the regulated businesses that cannot lawfully run hosted AI, and why owned, on-prem AI is the only compliant path left open to them.
The question nobody sizes properly
Most AI market maps start from the top: total enterprises, multiplied by an adoption curve, multiplied by an average contract value. That is the wrong way to find a durable wedge. A durable wedge is defined by constraint, not appetite. The right question is narrower and far more useful. How many businesses are legally prohibited from sending their data to someone else's cloud to be processed by AI they do not control?
When you count from the constraint instead of the hype, the number is large, specific, and stable. Roughly 0.85 million UK businesses, about 15 percent of the active population, sit inside regimes that make public-cloud AI a compliance failure rather than a procurement choice. Across the European Union the comparable figure is around 5 million firms. These are not laggards waiting to be convinced. They are organisations for whom hosted, multi-tenant AI is off the table by law.
I am Micky Irons, founder and CEO of Mickai. We built Mickai for exactly this population: the sovereign AI operating system that regulated businesses own and run inside their own walls, on-prem and air-gapped, with every action written to a tamper-evident, post-quantum-signed audit record. Mickai is built and live. This article sets out who the barred population actually is, why the constraint is permanent rather than transitional, and why owned AI is the only path that clears it.
What actually does the barring
The 0.85 million and 5 million figures are not a single rule. They are the overlap of several hard regimes, and most barred firms sit inside more than one at once.
Financial services resilience. The PRA's SS2/21 framework on outsourcing and third-party risk requires regulated firms to retain control, exit rights, and full auditability over critical functions. Pushing model inference and sensitive data into an opaque hosted endpoint does not satisfy that standard.
Special-category personal data. UK GDPR and its EU equivalent treat health, biometric, and other special-category data as a high bar for processing and transfer. Any architecture where the data leaves the controller's boundary to be read by a third-party model carries a burden most firms cannot discharge.
Healthcare assurance. The NHS Data Security and Protection Toolkit sets expectations that a multi-tenant public endpoint struggles to meet for patient-identifiable workloads.
High-risk AI. The EU AI Act places documentation, traceability, and human-oversight duties on high-risk systems. Those duties are far easier to evidence when the system runs inside your own estate with a complete record of every decision.
Export-controlled and defence data. ITAR and EAR restrict where controlled technical data may be processed and by whom. Air-gapped, owned compute is frequently the only lawful option.
Critical infrastructure. The NIS Regulations impose security and resilience duties on operators of essential services, which sits uneasily with dependence on an external inference cloud.
Jurisdiction. The US CLOUD Act means data held by a US-headquartered provider can be reached by US legal process regardless of where the server physically sits. For a European bank, hospital, or defence supplier, that single fact can disqualify an entire category of hosted AI.
Stack these regimes and you do not get a niche. You get a structural floor of demand that grows as regulation tightens, not one that erodes as models improve.
Why this constraint does not melt away
The common counter-argument is that hyperscalers will close the gap with sovereign regions, confidential computing, and customer-managed keys. Those are real and useful, but they soften the data-residency and encryption problems without removing the core issue: the firm still does not own the system, cannot fully audit it, and remains exposed to a foreign jurisdiction and to a control plane it does not run.
The regulated buyer's test is not whether the data is encrypted. It is whether the firm can prove, to a regulator, that it retained control of the function and holds an immutable record of every action that function took. That is an ownership question, not a hosting question. It is why the sovereign AI market is projected to grow from around USD 40 billion in 2025 to roughly USD 148 billion by 2032. The constraint is not a temporary friction on the way to full cloud adoption. It is the permanent shape of how regulated industry will run AI.
Why owned, on-prem AI is the only path that clears it
If the barring is structural, the answer has to be structural too. Mickai is built so the entire intelligence layer lives inside the customer's walls. The models run locally. Nothing has to leave the boundary to be useful. Every action, every retrieval, every decision is written to the Owned Audit Record, a tamper-evident log signed with post-quantum cryptography so it stands up years from now, not just today.
On top of that substrate we run a set of Greek-named Studios mapped to the work regulated firms actually do: Nemesis for fraud and AML, Plutus for finance, Tyche for underwriting, Prometheus for forecasting, Iris for customer service, Nomos for compliance, Astraea for legal, Panacea for clinical, Pythia for business intelligence, and Aletheia for audit. Around them sit Trust Agent, the AMT agent layer, Vinis voice, OAR-as-a-Service, and HELIOS hardware. This is built and live, manufactured in Birmingham, and being scaled now.
The defensibility underneath is an estate of 104 filed UK patent applications, roughly 2,340 claims, held by Mickai LTD with me as inventor. Filed, not granted, which gives priority and a prior-art moat rather than a marketing claim. Our own analysis maps 196 companies across 311 patent-company pairs as potential licensees, including names such as Microsoft, AWS, NVIDIA, Google, Adobe, and IBM. That is potential-licensee sizing, not booked revenue, but it tells you where the architecture sits relative to the field.
The dual-buyer thesis, and why a hyperscaler would care
Mickai is an ally, not an attempt to displace any frontier lab. The thesis is dual-buyer. The regulated enterprise buys sovereignty because it has no lawful alternative. And the hyperscaler, whose own platform is the very thing the CLOUD Act and SS2/21 make hard to sell into this population, gains a compliant on-prem path to a market it cannot otherwise reach. The same IP estate underwrites both. That is what makes this a category a large platform would rather own than compete with.
The momentum signal is independent and third-party. As of June 2026, Crunchbase ranked me at number four among founders, with the Mickai company profile sitting in the top one to two percent globally. That is a read on attention, not a valuation, and I treat it as such. The substance is the moat, the live product, and the regulated population that has nowhere else to go.
Where this is heading
The product is live. The IP is filed. The manufacturing is secured in Birmingham. The regulated demand is sized and structural, and it grows as regulation tightens. The path from here is scale: more Studios in production, more of the barred population served inside their own walls, and a patent estate that compounds in value as the sovereign category moves from edge case to default for regulated industry.
If you operate in or invest behind regulated industry and want the detail, the financial model and the diligence pack are available on request.
Contact: micky@mickai.co.uk
FAQ
How many businesses are legally barred from public-cloud AI? About 0.85 million UK businesses, roughly 15 percent of the active population, and around 5 million firms across the European Union sit inside regimes that make hosted, multi-tenant AI a compliance failure rather than a procurement choice.
Which regulations create the barrier? The main ones are PRA SS2/21 on financial outsourcing, UK and EU GDPR special-category data rules, the NHS Data Security and Protection Toolkit, the EU AI Act high-risk obligations, ITAR and EAR export controls, the NIS Regulations for essential services, and the US CLOUD Act on cross-border data access.
Why can on-prem ownership solve what sovereign cloud regions cannot? Sovereign regions and confidential computing improve residency and encryption, but the firm still does not own or fully control the system and remains exposed to foreign jurisdiction. Regulators test for retained control and an immutable audit trail, which is an ownership question. Mickai runs the models inside the customer's walls and writes every action to a post-quantum-signed Owned Audit Record.
What is Mickai? Mickai is a sovereign AI operating system: AI that regulated businesses own and run inside their own walls, on-prem and air-gapped, with every action recorded to a tamper-evident, post-quantum-signed audit record. It is built and live, with Greek-named Studios covering fraud and AML, finance, underwriting, forecasting, compliance, legal, clinical, business intelligence, and audit.
Frequently asked questions
How many businesses are legally barred from public-cloud AI?
About 0.85 million UK businesses, roughly 15 percent of the active population, and around 5 million firms across the European Union sit inside regimes that make hosted, multi-tenant AI a compliance failure rather than a procurement choice.
Which regulations create the barrier?
The main ones are PRA SS2/21 on financial outsourcing, UK and EU GDPR special-category data rules, the NHS Data Security and Protection Toolkit, the EU AI Act high-risk obligations, ITAR and EAR export controls, the NIS Regulations for essential services, and the US CLOUD Act on cross-border data access.
Why can on-prem ownership solve what sovereign cloud regions cannot?
Sovereign regions and confidential computing improve residency and encryption, but the firm still does not own or fully control the system and remains exposed to foreign jurisdiction. Regulators test for retained control and an immutable audit trail, which is an ownership question. Mickai runs the models inside the customer's walls and writes every action to a post-quantum-signed Owned Audit Record.
What is Mickai?
Mickai is a sovereign AI operating system: AI that regulated businesses own and run inside their own walls, on-prem and air-gapped, with every action recorded to a tamper-evident, post-quantum-signed audit record. It is built and live, with Greek-named Studios covering fraud and AML, finance, underwriting, forecasting, compliance, legal, clinical, business intelligence, and audit.