The rapid growth of AI creates ever-expanding investment opportunities with potentially lucrative returns. While the ‘newness’ of the artificial intelligence world naturally generates eagerness to invest, there is a growing question about whether the financial risks involved are fully understood.
Insuring the future value of AI chips is not yet a tried and tested practice, meaning concerns focus on whether the safeguards being put in place are truly providing the expected (and very much required) protection.
The cases that invited closer scrutiny
When it came to light that Nvidia had been exploring insurance arrangements which would protect lenders who accepted AI chips as collateral, this demonstrated why thorough scrutiny must become standard practice. Proposed arrangements, which the FT reports ‘may not result in any deals’, would see insurers compensating lenders if borrowers defaulted and the sale of chips did not cover the outstanding debt.
Whilst these arrangements could heighten protection for lenders, Nvidia itself potentially stands to gain if the plans go ahead. This underscores why independent oversight is a crucial part of governance arrangements, especially where the party profiting from the lending has also organised the insurance cover.
Another case which attracted attention was that of Greensill Capital, the now defunct institution which found itself the subject of fresh headlines numerous times in 2026, including when an A$2.8b claim was settled, five and a half years after Credit Suisse froze their Greensill-linked funds. Action was taken following the expiration and subsequent non-renewal of insurance cover worth $4.6b used to support investments in supply chain finance funds.
The parallel is structural, not conduct based, but Greensill’s case further highlights that financial safeguards organised by organisations with a commercial interest in the transaction require independent scrutiny. Simply having safeguards in place does not guarantee that the underlying risk has been adequately addressed.
If the chips are down
Additional insurance protection may be sought due to the difficulty of predicting the future value of AI chips. These assets are attractive as collateral against loans to fund further investment, however in a rapidly-evolving technology market, new developments can change demand for existing products.
The consequence is that there is no definitive depreciation benchmark, and their useful economic life must be assumed based on predictions about the pace of technology advancement going forward, and the resulting impact on existing assets and their value.
It is an equation which even major corporations are yet to solve: in 2025, Amazon shortened the estimated lifespan of some of its servers at the same time Meta extended the assumed useful lifespan of its own. The widespread uncertainty was even referenced by The Bank of England in its Financial Stability report which noted the ‘mixed’ evidence about the speed of depreciation. Of course, this economic depreciation should not be confused with operational lifespan, as a chip may lose its commercial value before ceasing to function effectively.
With fluctuations such as rental prices for Nvidia’s H100 falling by about three quarters from their peak, then rising again in a shortage, there is a real uncertainty for lenders looking to value collateral which may not be sold for several years. This doesn’t preclude chips being used as collateral, but it does reinforce the importance of thorough understanding and independent testing of the assumptions underpinning any valuation.
The underlying risks
Whilst insurance is intended to protect against loss, there are risks and dependencies which must be thoroughly understood, especially in such a new market.
Take for instance a scenario where an insurance policy expires before a loan has been repaid: this leaves the lender relying on the collateral after all. At the same time, a loan secured against AI infrastructure may depend on generated revenue from a contract, which again could end before full repayment – meaning the lender must then sell equipment which may have already depreciated significantly in value.
This is precisely why individual safeguards and revenue commitments should not be considered separately; rather, the complete financial picture must be assessed prior to a deal being agreed. In more complex arrangements where part of the risk is transferred to a hedge fund, there are further questions to be asked around who is ultimately responsible for providing protection, the circumstances in which they will do so, and the speed of accessibility.
The five key questions
As the AI market matures and governance infrastructures around technology investments develop, boards, lenders, insurers and audit committees must ensure the right questions are being asked at the right time, of the right people.
Does the organisation understand its total exposure (especially where commitments are spread across multiple loans, contracts, guarantees and parties)?
Are valuations based on realistic assumptions, and whose price is being used as the basis for continual assessment of the relationship between the outstanding loan and the value of its collateral?
Could income such as customer contracts or alternative revenue streams end before financial obligations have been met?
Will the financial safeguards in place provide protection when needed, and how quickly would a claim actually pay out (especially given the five and a half year wait for the Greensill settlement)?
And have risk assessments been independently challenged to identify overly optimistic assumptions, potential conflicts of interest, and vulnerabilities which could otherwise remain overlooked?
These are all fundamental principles of effective governance which are not unique to AI investment, but they become increasingly important in a market which is developing rapidly – with methods of assessing risk still evolving.
Reassuringly, the Bank of England’s Financial Policy Committee judged AI companies’ debt to be ‘relatively modest’ at the beginning of 2026, but this came with a warning that increasing complexity and a lack of transparency in debt structures could create wider risks. Risks that could be overlooked if effective governance has not been established and embedded into organisations pursuing AI investment opportunities.
Whether or not current high expectations come to fruition, there is a fundamental need for robust financial arrangements to recognise potential challenges and build appropriate protections. Enthusiasm for new technology and its related opportunities must not lead us to overlook the risks today’s decisions may pose in tomorrow’s world.
James Walker is an independent board chair, non-executive director and Chartered Internal Auditor who has led internal audit teams at Credit Suisse and UBS. Currently he provides governance, risk and internal audit services through Lightning Oak, chairs the board of Nature Impact, is a Non-Executive Director and Chair of the Finance, Planning and Performance Committee at Avon and Wiltshire Mental Health Partnership NHS Trust, and advises the boards of Footstep AI and Fallon AI. James is a Fellow of the Association of Chartered Certified Accountants, and a Chartered Member of the Chartered Institute for Securities and Investment.
