AIG's Eric Andersen is right. AI data centre underwriting is breaking traditional P&C capacity models — not because insurers lack appetite, but because they lack the tools to assess unfamiliar risk profiles fast enough. Data centres demand simultaneous coverage across project finance, build-out, operational resilience and cyber exposure. Most UK mid-market insurers are still managing this with spreadsheets, manual comparatives and claims teams that learn lessons too late. This matters because the FCA Consumer Duty (PS22/9) now requires firms to demonstrate they understand the products they sell and the risks they carry. When capacity is maxed out, corners get cut. When corners get cut, consumer harm follows.

This story is symptomatic of a broader truth: the insurance industry is experiencing a mismatch between the velocity of new risk and the velocity of human underwriting. AI infrastructure itself is creating entirely new risk classes — thermal runoff, supply chain concentration, regulatory exposure across jurisdictions that don't yet have AI governance frameworks. The EU AI Act, incoming UK AI regulation, and evolving PRA expectations around operational resilience (SS1/23) mean that underwriters need to assess not just the physical asset but the governance model of the entity operating it. This is not a scaling problem that hiring solves. It is an intelligence problem that only AI can solve.

Trovix's position is clear: the answer is not to deploy generic large language models like ChatGPT or Microsoft Copilot across your underwriting team and hope for coherence. Products like Harvey and Legora promise to 'democratise' legal and financial analysis, but they work best on highly structured, known-quantity problems — contract review, memo drafting, precedent identification. Underwriting new risk categories is the opposite. It requires enterprise knowledge systems that can hold your firm's own underwriting philosophy, learn from your claims data, integrate your regulatory obligations (SRA Code, FRC ISA UK), and produce defensible decisions with audit trails. Trovix Audit exists precisely because mid-market firms need to prove to their regulators that their AI is used consistently, fairly and within appetite. Trovix Aria builds RAG-based assistance that teaches your underwriters to reason from your own risk models, not from internet-scale training data. That is the difference between delegation and augmentation.

What should you do on Monday morning? Audit your underwriting stack. Find out where manual bottlenecks are costing you capacity and where human bias is creating inconsistency. Map the new risk categories you are being asked to underwrite and the regulatory obligations they trigger. Then build AI assist into those specific workflows — not as a replacement for judgment, but as a force multiplier for consistency and speed. If you are a law firm advising insurers, the same logic applies: your clients' capacity crisis is your opportunity to sell them better process design. If you are an insurer yourself, capacity constraints are forcing you to choose: hire more underwriters you cannot find, or invest in AI systems that make your current team superhuman. The Lloyd's Blueprint Two and PRA expectations mean regulators will ask how you chose. Get ahead of that question now.

Source: Bloomberg News