Introduction
There is no doubt that the biggest innovation race in the global economy is for AI. Some estimates attribute up to a third of US economic growth in 2026 to AI.
Law firms are no stranger to this, from Latham & Watkins' dedicated practice in AI, to Kirkland & Ellis' $500mn investment in an in-house LLM (Language Learning Model), to Freshfields launching an 'AI Law Degree' through its partnership with KCL. It sends the message that AI will shape how law is practised.
Yet while enthusiasm persists, several economists have started to ring alarm bells over the ballooning debt, and the prospect of the investments failing to pay off. With parallels drawn between AI and the Dot Com bubble, it is important to evaluate whether a new crisis looms, and what the consequences could be.
The case for an AI Bubble
Many AI startups cannot build entire data centres from the start. As a result, they resort to cloud computing providers such as CoreWeave, renting GPU capacity instead of owning it.
These data centres themselves are funded by syndicates that include the likes of Blackstone, JP Morgan and Goldman Sachs, with loans up to $8.5 billion in one recent case. The same GPUs used to power these data centres are taken as collateral.
The issue arises from the value of these GPUs, which tend to depreciate rapidly as tech titan NVIDIA continues to release newer models. Returning to CoreWeave, a company with a debt-to-equity ratio above 1,000%, insolvency would leave lenders holding collateral worth far less than the debt it secures.
Rather than seize hardware they have no capacity to operate, lenders are more likely to restructure. But a provider working to service that debt, in a sector where capital has become more expensive, will pass those costs on to the AI companies renting its capacity.
As a result, the cost of the services provided will rise. Clients of those companies, including firms using Harvey or Legora in commercial law, would see products repriced, features withdrawn, or providers consolidating into fewer hands.
What could this mean for Law Firms
This situation would be a blow to firms that have invested heavily. Even so, most could absorb rising licence costs, or in Kirkland's case the compute behind their own model. At worst, firms revert to working as they did three years ago, as AI has supported rather than changed working methods.
The most profound impact lies with the work, as restructuring and disputes teams will have to deal with the fallout from the distress of cloud computing companies, and the AI companies that face the second-order consequences. The lenders themselves will also be frequent clients in such cases, seeking to recover what they can. Inversely, transactional practices, particularly those specialising in tech and AI, will see a retraction in work after a loss in confidence in the industry.
What Next?
This remains a thought exercise, and lender optimism suggests it may stay one. Despite the real risks, lenders, firms and tech companies themselves expect efficiency gains to support AI's profitability. It is worth watching whether lenders flag GPU depreciation in future financing deals as an indicator of confidence in a heavily leveraged industry. The risk is unevenly spread. Firms that bought licences can walk away from them; those that committed capital cannot.