AIG's warning that AI data center demand is 'maxing out' P&C insurers is not a headline about the future — it is a present crisis. The problem is neither the scale of the opportunity nor the speed of AI deployment. It is that standard insurance underwriting workflows, built for traditional risk assessment, cannot absorb the complexity and concentration that data center portfolios demand. UK insurers regulated under PRA SS1/23 and FCA Consumer Duty PS22/9 are exposed to exactly this tension: they must move fast on underwriting new infrastructure risk, but they cannot cut corners on documentation, risk assessment or audit trails. Mid-market firms in London, Manchester and Edinburgh are already feeling this pinch. The challenge is not whether AI can help — it is whether the AI tools chosen will actually solve the problem or just make it easier to move fast in the wrong direction.

This story reveals a pattern that defines insurance tech in 2026: generic large language models built for customer service or research are being pressed into roles they were never designed for. Harvey, Luminance and similar legal-first AI tools have their place in due diligence and contract review. Microsoft Copilot in Microsoft 365 is useful for productivity. But neither of these approaches was built to extract, validate and cross-reference the highly structured, domain-specific data that insurance underwriting actually requires. When an underwriter needs to validate whether a data center's fire suppression systems meet ISO 27001 standards, or trace inconsistencies in electrical load declarations across three engineering reports, a general-purpose AI reading paragraphs is not the right tool. What is emerging instead is that insurers are doubling down on document intelligence platforms and RAG systems that sit on top of the firm's actual underwriting knowledge base — systems that can answer questions anchored to verified risk data, not to probability and guesswork.

Trovix's view is direct: the firms winning this moment are not the ones racing to automate everything with the latest LLM. They are the ones automating document intake and data extraction with precision, then routing that validated data through human underwriters who have better information faster. Trovix Sift handles exactly this — it extracts structured data from the chaotic documents (engineering assessments, fire safety reports, electrical designs, environmental impact studies) that come with AI infrastructure projects. That data then flows into the underwriter's workflow cleanly, with confidence scores and source traceability. The difference between this and a chatbot answering questions about the same documents is the difference between a validated signal and a sophisticated guess. On top of that, Trovix Brief automates intake and triage so underwriters are not manually keyboarding data from intake forms into case management systems. Trovix Aria then lets underwriters query their own firm's historical decisions and risk appetite rules in natural language, grounded in real precedent, not hallucination.

If you are a mid-market insurance firm or a financial services outfit with underwriting exposure, the move is immediate and specific. First: audit your document intake process for data center and AI infrastructure projects. You are almost certainly losing data in the handoff between business development, underwriting support and the underwriter's decision. Second: test a document intelligence tool on a sample of recent submissions — not to replace underwriters, but to give them structured data they can actually use. Third: map your risk appetite rules and underwriting precedents into a knowledge system that underwriters can query. The firms that will write AI infrastructure risk profitably over the next two years are the ones that accept that speed and safety are not opposites — they are dependent on each other. Generic AI tools deliver speed. They deliver neither the safety nor the structural integrity that a regulated firm actually needs.

Source: Bloomberg News