EU raises the SME threshold, lowers the barrier — the omnibus is industrial policy in compliance wrapping
The EU AI Omnibus agreement raises the SME-relief threshold from 250 employees / €50M revenue to 750 employees / €150M. That sweeps Mistral, the German-Swiss research labs, and Bull/Atos-derived AI ventures into the simplified-compliance bracket. It's not a regulatory rollback. It's industrial policy done through compliance language.
The EU AI Act has been characterized for two years as either too strict (industry view) or too lenient (civil-society view). Both characterizations missed the structural function of the omnibus negotiation that closed on May 7, 2026. The omnibus is industrial policy — designed specifically to keep European AI labs competitive with US-headquartered alternatives while maintaining the regulatory framework's external-facing strength.
What changed
The SME threshold for simplified compliance moved from 250 employees / €50M revenue to 750 employees / €150M. That single change brings most genuinely European Series C / D AI ventures into the simplified-compliance framework — Mistral well above the old threshold but comfortably below the new one, the major German-Swiss research labs (DFKI, ETH Zurich's industrial spin-offs), the French government-backed Bull/Atos-derived AI ventures, and the entire next-generation cohort of EU AI startups that would otherwise have been over the threshold.
The simplified compliance regime gets these companies four benefits: simplified guidance documents, reduced fines (capped lower than the standard 7% of global annual revenue), access to regulatory sandboxes where they can test compliance approaches with regulator feedback before formal commitment, and standardized documentation templates that reduce the per-product compliance overhead.
Why this is industrial policy
The threshold change isn't a regulatory rollback — the high-risk AI system rules still apply to everyone, the transparency obligations still come into effect August 2, 2026, the prohibition on certain AI practices (now extended to nudifier applications effective December 2026) still applies. What changed is the cohort of companies that get to interact with the regulatory regime in simplified-cooperative mode versus full-compliance-adversarial mode.
The cohort that gets simplified treatment is, almost without exception, European-headquartered companies competing with US-headquartered incumbents (Anthropic, OpenAI, Google) that are well above the threshold and get the full-compliance treatment. That's industrial policy: regulatory advantages for the domestic cohort, regulatory burdens for the foreign cohort, dressed in compliance language.
Whether it works depends on enforcement asymmetry
The Q1 2026 EU AI Act enforcement data showed €250M in fines across 50 cases, with Ireland handling 60% as the regulator-of-record for US-headquartered tech. If that enforcement cadence holds at 2026 full-year scale (~€1B in fines), the simplified-compliance framework for European SMEs becomes meaningfully valuable in cost-of-compliance terms. A US-headquartered company at the same size paying full-compliance overhead plus potential fines is at a procurement disadvantage versus a European company at the same size paying simplified-compliance overhead with reduced fine exposure.
The procurement consequence: European enterprise buyers that have a choice between an EU-domiciled SME-status vendor and a US-headquartered full-compliance vendor will see the European vendor's price advantage compound through 2026-2027 as compliance costs accrue. That's the policy goal — make European AI procurement cheaper for European customers than US-headquartered alternatives — and the omnibus is the mechanism.
The transatlantic asymmetry
The US response is not symmetric. The March 2026 US National AI Policy Framework is recommendations, not law. There's no US equivalent of the SME relief framework because there's no US equivalent of the AI Act enforcement regime to relieve against. US AI startups face a patchwork of state regulations (California SB-1047 framework, New York's transparency requirements, Texas's anti-discrimination rules) plus pending federal preemption discussions, but no unified compliance burden that the federal government could systematically lighten for domestic SMEs.
That asymmetry means the EU's industrial-policy-via-compliance-relief is unilateral. US AI ventures don't get equivalent treatment from their own government. The Brussels effect — EU regulations becoming the global standard because cross-jurisdictional companies architect to the strictest applicable rule — combined with the SME-relief asymmetry, creates a structural incentive for AI ventures to establish European entities to access the SME-relief framework. That's a real industrial-policy win for the EU's AI ecosystem, even if it doesn't make the news coverage of the omnibus deal.
Consilium — AI Council and Parliament agree to simplify and streamline rules → · Tech Policy Press — What the EU AI Omnibus Deal Changes for the AI Act → · Latham Watkins — AI Act Update EU Resolves to Change Rules Extend Deadlines →