If a new financial crisis develops in the coming years, it will probably not begin when everyone realises that something is wrong. Major crises tend to build much earlier, when assets considered sound are valued at prices incorporating increasingly optimistic expectations. The mechanism is fairly intuitive: when shares, property or other financial assets keep rising in value, their owners feel wealthier. Households, businesses and investors become more willing to spend, invest and borrow. Credit further fuels price growth and, for a while at least, everything seems to work. Problems emerge when expectations change. If asset values fall, perceived wealth declines, credit becomes harder to obtain, investment is postponed and consumption slows. If substantial debt has also been built on earlier valuations, a simple price correction can become something much more serious.
This happened in different ways during the dot-com bubble, the housing and financial crisis of 2007-2009 and numerous other episodes in economic history. Today's interesting question is therefore: where might the risk that markets are underestimating lie? The most immediate answer would be artificial intelligence. Many technology companies command high valuations, investment in data centres, semiconductors and AI infrastructure is reaching enormous proportions, and a growing share of the market incorporates very aggressive expectations about AI's future profitability. But looking only at technology shares may mean focusing on the most visible part of the problem. A much less familiar sector probably deserves just as much attention: private credit.
What private credit is
In simple terms, private credit is lending to businesses by entities other than traditional banks through privately negotiated loans. Instead of obtaining a conventional bank loan or issuing bonds in the market, a business can receive capital from specialised funds, Business Development Companies (BDCs, US vehicles that finance businesses and file reports with the SEC), asset managers, insurers or other institutional investors.
The sector has grown enormously since the 2008 financial crisis. Tighter banking regulation, combined with investors’ search for higher returns, has created more room for non-bank intermediaries. In its report of 6 May 2026, the Financial Stability Board estimates 1,500-2,000 billion dollars in assets at the end of 2024 within the scope considered. This is a dated snapshot dependent on the definition used, not a real-time measure of all non-bank credit. The Financial Stability Board also emphasises that the sector has become increasingly interconnected with banks, insurers, pension funds and private equity. This does not mean that 2,000 billion dollars are suddenly at risk. It does mean, however, that private credit is now large enough to warrant attention when considering financial system stability.
Why private credit has become so important
After 2008, banks became subject to stricter capital and liquidity requirements. At the same time, many companies, particularly medium-sized businesses or those with higher risk profiles, continued to need financing. Private credit funds progressively filled this space.
For companies, the advantage is clear: they can obtain more flexible and customised financing than public markets offer. For investors, the main benefit is yield, because loans to riskier companies normally need to pay higher interest. But this is precisely the other side of the coin: a higher return does not come free; it compensates for higher risk.
The connection with artificial intelligence
The relationship between AI and private credit is more complex than it may seem. The problem is not simply that funds have financed thousands of AI startups with no revenue. That would be an exaggeration. A substantial part of the exposure instead comes from the software sector, which in previous years became one of the main recipients of private credit, particularly through private equity transactions.
The Federal Reserve has noted that software had become the largest sector in several private credit funds’ portfolios. A study published by the Federal Reserve Bank of Boston on 5 August 2026 estimates that, at the beginning of 2026, internet and software accounted for around 20% of the number of loans at the median BDC, albeit with very wide differences between managers. The authors of the BIS Bulletin dated 14 July 2026 also estimated around 115 billion dollars of Business Development Company loans to software businesses, equivalent to roughly one fifth of their total lending.
This is where artificial intelligence enters the picture. For some software companies AI is an enormous opportunity, while for others it could become a threat. Software that only a few years ago required specialised services, large teams or expensive subscriptions could be replaced or put under pressure by AI-based tools. If these companies’ expected revenue falls while debt remains unchanged, their ability to repay financing could deteriorate rapidly. The risk, therefore, is not only that the AI equity bubble might burst. It could also be that AI makes some of the companies that accumulated substantial private debt in previous years less profitable. These are two completely different mechanisms, but they could interact.
The valuation problem
This brings us to what is probably the most interesting element. If you own a publicly listed share, you know almost at any moment what price the market is willing to pay. If it was worth 100 euros yesterday and 70 today, the loss is immediately visible. A private loan works differently. There is not necessarily a liquid market in which that same claim trades every second, so its value must be estimated.
Take an extremely simple example. We lend 100 euros to a friend. As long as the friend pays regularly and nothing unusual happens, we can continue to value the claim at close to 100 euros. But imagine the friend loses their job, accumulates other debts and starts experiencing financial difficulties. Formally, we still have a nominal claim of 100 euros, but economically those 100 euros may no longer be worth 100. Someone willing to buy that claim today might offer only 70 euros because there is a risk of not recovering the full amount.
The issue with private credit is that this reassessment of value can occur more slowly than in public markets. That does not mean managers can simply pretend an impaired loan is always worth 100: valuation procedures, controls and accounting obligations exist. But the absence of a continuous market price inevitably makes valuation less immediate. Indeed, the Financial Stability Board identifies valuation opacity and data scarcity as vulnerabilities to monitor in the sector.
The risk is not that nobody knows what anything is worth
We should avoid the opposite extreme, however. Saying that hundreds of billions in loans might be completely overvalued simply because they do not trade daily would be wrong. Funds recognise write-downs and periodically update portfolio values. For example, in its communication of 4 June 2026, using data at 30 April, Blackstone stated that its BCRED fund valued its overall portfolio at around 96.1 cents per dollar of par value, while the weakest 5% of its private debt portfolio was valued at an average of around 68.3. Potential losses are therefore not necessarily ignored until default. The real issue is subtler. When an asset lacks a continuous public price, the speed at which economic deterioration appears in valuations may differ from that of a listed asset. During a crisis, that difference can become very important.
What happens if investors all want to exit at once?
This is probably the most sensitive issue. Historically, many private credit funds were structured with long-term capital: investors committed money knowing that it would remain tied up for many years. In recent years, however, so-called semi-liquid funds have grown, allowing investors to request periodic redemption of their holdings.
The Federal Reserve's May 2026 Financial Stability Report states that the semi-liquid vehicles considered held 425 billion dollars in gross assets and 241 billion in net assets, representing around 20% of net assets managed by private credit vehicles within its scope. Gross assets and net assets are different measures (the latter accounts for liabilities) and should not be indiscriminately compared with global estimates using other definitions. This creates an obvious problem: investors can request their money back, but the loans held by the fund cannot necessarily be sold immediately. This is a form of liquidity mismatch, a difference between the liquidity promised to investors and that of the assets held. To avoid forced sales, many funds impose redemption limits, usually called gates. And that is exactly what we have seen in 2026.
Blackstone and Blue Owl: what actually happened
In the second quarter of 2026, investors in the Blackstone Private Credit Fund, BCRED, requested redemptions of around 10% of the fund's shares. In its communication of 4 June, Blackstone announced that it would repurchase up to 5% of outstanding shares under the approved programme. This is the proportion of the fund eligible for repurchase, not 5% of each request. It does not mean that BCRED was insolvent, nor does it, by itself, demonstrate insufficient liquidity at the manager: the fund and the management company are separate entities. The limit is part of the fund's structure and exists precisely to prevent large redemption requests from forcing the manager to sell illiquid loans rapidly.
According to investor communications reported by Reuters, second-quarter 2026 requests amounted to 18.8% for Blue Owl Credit Income Corp. (OCIC) and 38.1% for Blue Owl Technology Income Corp. (OTIC); repurchases remained capped at 5% of shares. In the 2 October update covering the third quarter, OCIC fell to 16.8%, while OTIC reached 39%. Many requests resubmitted previously unfulfilled applications: they therefore do not all represent new decisions to exit. The pressure is real but differs between funds, and requested percentages are not the same as actual outflows.
Why this can become a problem
Take a simplified example. A fund owns 100 billion dollars in private loans and investors ask to withdraw 20. The fund cannot simply call the financed companies and demand immediate repayment, because those loans have contractual maturities. It can use available cash, contractual loan repayments, new capital, credit facilities or asset sales, within applicable limits. If it applies a repurchase cap, moreover, it does not necessarily have to meet the entire 20 billion requested in the same period.
This is where the problem arises. The assets easiest to sell also tend to be the highest-quality or most liquid. If a fund is forced progressively to sell those very instruments, it may end up holding a growing share of loans that are harder to trade or riskier. Other investors could then become more worried and also ask to exit, increasing pressure on the fund's liquidity. This is the classic dynamic in which a confidence problem can become a financial problem, and it is precisely why gates exist.
Contagion can follow
At this point, some might think the problem affects only fund investors. In fact, private credit is closely connected to the rest of the financial system. Banks lend to funds, insurers invest in them, pension funds buy holdings and private equity firms use private credit to finance their companies. Some businesses also use bank facilities and private loans simultaneously. According to the Financial Stability Board, known direct bank exposures in the form of drawn or available credit facilities to private credit funds total around 220 billion dollars, while some commercial estimates range from 270 to 500 billion. Indirect exposures to common borrowers may add to these facilities: the full network of connections is not captured by a single balance sheet item.
Precision is needed, however: simply saying that if private credit collapses, banks will collapse too would be wrong. Bank exposures exist, but for now appear relatively limited compared with the size of the overall system. The Reserve Bank of Australia estimates that bank lending to private credit funds represents less than 0.5% of banking assets in the main jurisdictions considered, while the Federal Reserve Bank of Boston estimates that bank credit lines to BDCs represent less than 2% of large US banks’ Tier 1 capital.
Many private credit funds also have capital structures very different from banks. A 2026 NBER working paper estimates that equity generally represents 65-80% of private credit funds’ assets, compared with around 10% at the US banks analysed. In the funds examined, vehicle lifetimes are also generally very long relative to the loans held, reducing the classic maturity transformation problem characteristic of banking. Private credit therefore presents real risks, but it is not simply a new version of the banks of 2008.
How a crisis might develop
Now imagine a hypothetical scenario. Investment in artificial intelligence keeps growing, expectations of future profits rise ever higher, and many technology and software companies are financed on the basis of expected future growth. Then AI's economic results arrive more slowly than expected. This does not necessarily mean artificial intelligence has failed: it could simply mean that economic returns require ten years rather than the three investors had priced in.
Technology company valuations then begin to fall, some highly indebted software companies experience declining revenue and margins, defaults rise and private credit funds must write down some loans. Investors in semi-liquid vehicles begin requesting redemptions, funds apply gates, sentiment deteriorates further and new capital flowing into private credit declines. Businesses most dependent on this financing consequently find refinancing harder, defaults and restructurings increase further and, meanwhile, banks, insurers and other institutional investors begin reducing risk exposure. Credit conditions then tighten, companies invest less, employment growth slows and consumption begins to decline. This is how an initially financial problem can progressively reach the real economy. Not necessarily through a single spectacular event, but through a chain of mutually reinforcing effects.
Artificial intelligence might only be the trigger
This is where the discussion becomes more interesting. Talk of an “AI bubble” almost always focuses on the valuations of Nvidia, Microsoft, Meta or other large technology companies. But the industry's biggest companies also have enormous revenue, cash resources and access to financial markets. Systemic risk might instead lie in the connections built around AI investment: debt to build data centres, loans to software companies, infrastructure financing, private credit, private equity, bond markets, banks and insurers. In 2026, the bond market is also absorbing enormous amounts of new AI-related debt. The point is therefore not necessarily to identify a single “bubble”, but to understand where the same economic expectation has simultaneously been embedded in shares, bonds, private loans and infrastructure investment.
The private credit paradox
Finally, there is a particularly interesting aspect. One feature that makes private credit appear more stable in normal markets, namely the absence of continuously updated prices, can simultaneously become a source of uncertainty during a crisis. A share can lose 20% in three days, while a private loan may appear much more stable. But this does not automatically mean its underlying economic risk is lower. It may simply mean that risk is measured at a different frequency. Observed volatility and actual risk are therefore not necessarily the same thing. An asset whose price does not change every second is not automatically a less risky asset.
Are we therefore on the eve of another financial crisis?
No. The available data do not support that claim. Indeed, several factors point in the opposite direction. In its May 2026 Financial Stability Report, the Federal Reserve still considered pressures from redemptions at semi-liquid funds manageable. In its April 2026 briefing, the International Monetary Fund noted that redemption limits had contained pressures and that the aggregate exposures examined remained manageable, although monitoring was required. These assessments relate to those dates and conditions; they are not guarantees about the future. Blue Owl's redemption requests also declined in the third quarter relative to the second for its main fund, suggesting that conditions are not necessarily accelerating towards a crisis. Talk of another Lehman Brothers today would therefore be unfounded. But ignoring the signals would be equally wrong. The Financial Stability Board, Federal Reserve, BIS and other institutions are devoting growing attention to private credit. And when market size, sector concentration, redemption requests and links with the banking system increase simultaneously, it is worth understanding what is happening.
What I would watch over the coming months
To understand whether risk is really increasing, I would not look only at the Nasdaq. I would focus above all on private credit defaults, because a rapid rise would signal direct deterioration among borrowers; portfolio write-downs, which can show how much value loans are losing; redemption requests at semi-liquid funds, because a single quarter means little but persistent growth would be much more significant; new fundraising, because if investors stop supplying capital, the sector loses an essential source of financing; software exposure, one of the sectors where disruption from artificial intelligence could emerge fastest; and finally the relationship between banks and private credit, because the more interconnections grow, the larger the potential contagion channel becomes. I would also watch the AI-related debt market. If spreads, defaults and refinancing difficulties rose simultaneously, that would be a much more significant signal than a simple correction in technology shares.
In conclusion
The next financial crisis, if and when it comes, is unlikely to look exactly like the previous one. That is precisely the problem. The 2007-2009 crisis demonstrated how subprime mortgage risk could spread through securitisations, derivatives and bank balance sheets. Today everyone is talking about artificial intelligence, but perhaps the more interesting question is not simply what AI companies are really worth.
The more important question is how much debt has been built around expectations generated by artificial intelligence, and where that debt has ended up. Private credit is one area to watch, not because it is necessarily destined to trigger a crisis, nor because there is evidence today of an imminent systemic crisis, but because it combines three characteristics that always deserve attention in financial history: very rapid growth, less transparent valuations and increasing interconnections with the rest of the financial system. The real question is therefore not: “Will AI trigger the next crisis?” It is: “If expectations about AI prove too optimistic, who financed those expectations, and who would bear the losses?” Those connections are a useful starting point for analysis, without turning a possible scenario into a certain forecast.
Update note. Data and sources checked as of 3 October 2026. The scenarios described are hypothetical: this article does not predict a financial crisis or constitute an investment recommendation.