Life insurers are now the hidden funding engine of the AI boom. Bloomberg reports that insurance firms facing record annuity demand are buying long-term, high-grade corporate bonds from tech borrowers—a direct capital channel that sits outside traditional banking scrutiny. For UK insurers regulated under PRA SS1/23, this matters profoundly: you are now material lenders to an industry whose fundamental credit metrics remain opaque. The question is not whether this capital should flow. It is whether the due diligence frameworks exist to assess what you are actually funding. They do not.
This story exposes a pattern we have watched for two years. AI infrastructure spending has become decoupled from earned revenue. Companies borrow on investor faith that AI will eventually justify trillion-dollar hardware costs. Insurers, desperate for yield in a low-rate world, have become the lenders of last resort. What Bloomberg tactfully calls 'high-grade corporate bonds' are often credit instruments issued by entities with zero mature revenue streams. The FCA Consumer Duty (PS22/9) requires firms to act in customers' best interests. Pension holders and annuitants do not know their security now depends on whether Anthropic, xAI, or smaller GPU-cluster operators can monetise inference. That is a risk disclosure problem masquerading as a market opportunity.
The honest truth: AI vendor due diligence is broken. Firms using large language models like OpenAI's GPT-4, Claude, or Mistral to assess counterparty creditworthiness are teaching machines to pattern-match on historical data that has no predictive power in this context. Tools like Harvey or Luminance excel at document review in law and compliance, where the task is classification of known categories. Credit assessment of pre-revenue AI labs is not a classification problem. It is a judgment call under uncertainty—exactly what generative AI cannot do reliably. Worse, it creates the illusion of diligence. Trovix Aria was built on the opposite principle: human expertise preserved and accessible, AI used only to surface and synthesize information that humans then validate. Your chief investment officer should be calling the shots on AI infrastructure credit, not delegating to a black box. And Trovix Watch is designed to flag when regulatory guidance changes—because PRA and FCA expectations on AI counterparty risk assessment will harden within months, not years.
If you are an insurance CIO or CFO right now, audit your AI-backed due diligence workflows immediately. Ask: where in my credit assessment is AI making the final call, and where am I pretending it is? Insist on human sign-off for any bond purchase that funds AI infrastructure. Do not accept vendor assurance that 'the model was trained on similar deals.' This is not similar to anything. Second, check your regulatory reporting. PRA expects concentration risk disclosure; AI infrastructure is now a material counterparty class, and the FCA will want to know you have named it and quantified it. Third, engage your legal team on whether private bond purchases of AI assets trigger new disclosure obligations under the EU AI Act equivalence regime (now live for UK firms). Finally, demand that your data and analytics teams build transparency into AI risk assessment, not opacity. The insurers who will still be profitable in 2032 are those who lend to AI infrastructure with eyes open, not with models trained on a world that no longer exists.
Source: Bloomberg News