From the City to the high street, concern is growing about an AI Bubble and its effects on everything from the financial services to public services. AI financing is growing substantially, and investment in chip technology is squeezing availability for other purposes. While productivity appears to be increasing on a macro level, company-level experiences are mixed. The latest release of OpenAI’s Claude, an AI chatbot, may provide some insight on how AI will be successful and disrupt generic workflows, as did the calculator, for instance, but what could happen to more specialist financial services use cases were a bubble to burst. In this article, GreySpark looks at the implications of an AI bubble on AI software firms, Private Markets and Investment Banks.
By GreySpark’s Declan Sharp, Analyst Consultant and Rachel Lindstrom, Head of Capital Markets Intelligence practice
Firstly, AI may have real tangible productivity impacts in the market as proposed by macro-economists. The evidence of productivity gains in professional work and education are significant and supported by a significant number of studies since 2022. Looking at US labour productivity numbers (see Figure 1), it seems that there is an uptick in 10-year annualised productivity growth post-COVID, signifying a possibility of the start of a trend of productivity growth. Indeed, similar to the J-Curve experienced in the US before and during the adoption of computers. The uptick seen in 2023 onwards, may be the start of the curve in a new cycle related to AI productivity gains. Alternatively, the growth could be an amalgamation of the impact of AI and the return to work post-COVID. However, current macroeconomic studies predict that annual aggregate productivity growth due to AI could range between 0.4 to 1.3% in G7 countries.
Many senior industry leaders predict productivity gains and job losses in professional work due to the adoption of AI. This is because the productivity gains in administrative, financial and computer/software based occupations will be substantial. These sectors are so exposed to AI that significant job losses could occur. For those that remain, the effect of AI adoption appears to be an intensification of work, because employees are developing a broader set of skills, in theory boosting productivity.
Figure 1: US Labour Productivity (10-year Annualised Nonfarm Productivity Growth Rate)
Source: US Bureau of Labor Statistics, GreySpark analysis
(Click to enlarge)
Financial and fintech firms are early adopters of AI and are implementing AI at scale with mixed results. In an earnings call for Q3 2025 results, JPMorgan estimated that it had made USD 2 billion savings from AI efficiency gains following a USD 2 billion investment. As the firm extends AI into other areas, the firm expects ROI to become positive. On the other hand, Klarna implemented OpenAI powered agents for sales and customer service roles and although this was successful project, the CEO said the firm had ‘overpivoted’ and did require staff for some of the teams that were most affected by the cutback. MIT’s 2025 study on the state of the AI industry, highlighted that 95% of AI programmes had not seen any return and although individuals may be more productive, ‘these tools primarily enhance individual productivity, not P&L performance’. This calls into question the benefits firms are receiving from adopting AI, adding to a possibility that the growth in AI spending, and therefore growth in revenue for AI firms, may slow.
Sky-high capital expenditure (CAPEX) and borrowing from AI technology giants such as Google, are fuelling fears of an AI bubble. In November 2025, a US analyst and investor, Dr Michael Burry, of The Big Short fame, wrote that he believed the circular deals (wherein Company A invests in Company B, which in turn uses those funds to buy products or services from Company A) were fuelling an AI bubble that reminded him of the ENRON energy pricing and accounting scandal of the 2000s. NVIDIA’s Jensen Huang refuted the comparison, strongly stating that NVIDIA is ‘not ENRON’. A month later, in January 2026, NVIDIA walked backed from an agreement for USD 100 billion investment into OpenAI. However, OpenAI’s USD 300 billion spending commitment in Oracle’s cloud storage is expected to go ahead. OpenAIs spending commitments are mounting raising concern about the shockwaves across the whole market should they prove an inability to pay. In an investor call, Open AI’s CEO, Sam Altman disputed that it will be difficult for a company with USD 13 billion in revenue to pay its over USD 1 trillion in spending commitments, and later wrote on X that OpenAI is making USD 20 billion in revenue (still significantly short of the USD 1.4 trillion in spending commitments).
OpenAI’s CFO, Sarah Friar, in a townhall at the end of 2025 hinted that such large spending commitments and investments may require a government guarantee. OpenAI’s CEO, Altman, denied on X afterward that the company would be seeking a government bailout, and that the firm projects revenue to grow into the hundreds of billions before 2030.
MIT’s 2025 study on the state of the AI industry, highlighted that 95% of AI programmes had not seen any return and although individuals may be more productive, ‘these tools primarily enhance individual productivity, not P&L performance’.
The amount of debt pouring into technology is potentially a cause for concern for banks and private capital, as technology companies are moving to a new funding model. Hyperscalers, such as Google and Meta, are seeking bond and private debt funding for AI and data centres. Debt accumulation as a result of these endeavours reached USD 3 trillion at the end of 2025. In the case of Google, its release of the 100 year sterling bond in February 2025 and its announcement of further USD 185 billion investment into AI in 2026, suggests that it is diversifying the financing of its technology development. A recent bulletin from the Bank of International Settlements (BIS) highlights that even these hyperscalers are turning to debt issuance en masse to fund the CAPEX on AI and data after exhausting their cash flows. This is creating the new and worrying phenomenon of credit default swap issuances for hyperscaler traded debt as investors seek to hedge against the possibility of the giants being unable to meet their commitments. Massive investment from Private Credit (PC) on data centres, on the other hand, produce positive results if AI is to be successful. The outstanding amount of direct loans from PC to AI firms was now estimated to be roughly USD 200 billion at the end of 2025. The largest example of technology firms turning to private capital was in late 2025, when Meta reached an agreement with Private Capital giant, BlueOwl for a USD 27 billion joint venture (JV) for the creation of the Hyperion Data Centre in Louisiana.
BlueOwl will have 80% of the ownership, with Meta only putting up USD 5.4 billion as part of its 20% stake. Meta will lease the data centre from BlueOwl, thus most of the cost of the data centre is off balance sheet. BlueOwl is primarily funding its investment through debt, with lead investor PIMCO already selling some of its stake. This may indicate clever CAPEX accounting by Meta, or a safety net if the AI race cools down in the future. Figure 2 shows mounting commitments for hyperscalers.
Mentions across the industry of a looming ‘SaaSpocalypse’ (where AI companies take over some of the SaaS business, might deter consistent and large investment in software companies by Private Equity (PE) and PC. PE firms looking for exits in software companies may run into trouble if there is a compression in their valuations as business models are disrupted by AI, causing a possibility for PE investors to be at risk and PC lenders to be overly exposed to the industry. Banks could be impacted by their exposure to private market firms through the various types of lines of credit they have extended to private capital firms. Indeed, private capital firm exposure to SaaS may become a systemic risk according to Moody’s if there is spill-over from portfolio companies and private capital firms to bank loan delinquencies and ultimately defaults.
Figure 2: Hyperscaler Commitments to Leases that are not yet Commenced (USD Billions)
Source: Bloomberg
(Click to enlarge)
Infrastructure Investment for the Future
The elevated valuations, circular financing and rising debt are reminding many of the DotCom bubble in the early part of the millennium. Like that technology bubble, which burst in 2001, AI is here to stay. The spending on data centres, IT manufacturing, IT equipment and software is a rising and estimated by the BIS to be nearly 5% of GDP as of 2025. The massive upward trend on capital expenditure by AI and chip maker firms mirrors that of technology and ecommerce firms during the DotCom bubble.
AI and data infrastructure, like the server and cable infrastructure in the late 1990s, will most likely bear fruit eventually even if firms go bust, as they did when the DotCom bubble burst. Similarly, the infrastructure spend on data, AI, software and the energy and construction capabilities to fuel this build out could also be reminiscent of infrastructure investments for electricity, telephone or railways, and despite potential overexuberance, could be beneficial. However, despite this long term potential, significant market disruptions and systemic risks remain likely. Equity and debt investment in software companies with risky business models and the debt funding of hyperscalers could mean that there is potential systemic risk to banks and private capital firms due to their exposure to AI technology.
