The ECB is not saying the AI boom is 2008 all over again. It is saying you should stop pretending the debt behind it is someone else's problem.
Nvidia went back to the bond market on June 15, and that should make the AI buildout feel less abstract. MarketWatch reported that the chipmaker planned to raise $20 billion across seven tranches, with maturities running from 2028 to 2056, in its first corporate bond sale since 2021. Goldman Sachs, JPMorgan and Morgan Stanley were among the banks leading the deal. For a company throwing off huge cash, that is a striking move. Even Nvidia is borrowing.
This is where Christine Lagarde's European Central Bank has put its finger on the right problem. The May 2026 Financial Stability Review did not call the top of the AI trade, and it did not say private credit is about to break the financial system. It said the part most investors cannot see clearly is now large enough to matter.
According to the Financial Times, the ECB warned that an AI investment boom increasingly financed through private credit could expose euro area insurers and pension funds to losses if expected AI-related cash flows disappoint. In the ECB's stress exercise, direct bank losses were limited, but insurers faced losses of about 4% of assets and pension funds were hit in the 5% to 6% range. That is not an accounting footnote if you are the person whose retirement money sits behind those funds.
The comparison everyone noticed was the ECB's sizing of the US private credit market. The central bank put it at roughly $1.4 trillion at the end of 2024, close in dollar terms to the $1.5 trillion US subprime mortgage market in 2006. The Wall Street Journal noted the caveat that matters: private credit was about 4.7% of US GDP, while subprime mortgages were nearly 11% of GDP before the financial crisis. So no, this is not a copy of 2008. Frankly, that is the wrong test.
The better question is simpler. Where does the loss go if the data center math gets worse?
The Financial Stability Board gave one answer in May. As The Guardian reported, AI accounted for more than a third of private credit deals in 2025, up from 17% over the previous five years. The watchdog warned that a sharp correction in AI valuations, delays caused by electricity shortages, or an oversupply of data centers could lead to sizeable credit losses for private credit investors. Those are not fantasy risks. Data centers need land, power, chips and tenants willing to pay enough to justify the buildout.
The banks are not the whole story here. Senior lenders usually sit higher in the capital stack, so they get paid before the investors holding equity or subordinated debt. Pension funds and insurers often hold those lower layers because they need yield. That is why the ECB keeps coming back to transparency, definitions and data collection. If regulators cannot see who owns which slice of the AI debt pile, they cannot judge whether a sector-specific stumble becomes a broader financial problem.
The FSB's report also pointed to Tricolor and First Brands, two private credit-backed US automotive companies that collapsed after fraud allegations. Banks including JPMorgan and Barclays took losses tied to Tricolor, while UBS and Jefferies reported exposure to the failures. Those companies were not AI firms, but they show the weakness in the market. Private credit can look calm right up to the moment the paperwork turns out to be thinner than everyone assumed.
G7 leaders are discussing AI at Évian-les-Bains this week, with AP reporting that OpenAI's Sam Altman, Google DeepMind's Demis Hassabis and Anthropic's Dario Amodei joined talks on the technology's future. Most of that conversation is about safety, dominance and regulation. The money should be on the table too. You cannot separate the technical race from the financing race when the same buildout is driving bond sales, private loans and power-hungry data center projects across several markets.
Axios cited Goldman Sachs analysts saying hyperscaler capital spending could reach $770 billion in 2026, equal to their total cash flow from operations. That is the cleanest figure in the whole story. If the largest technology companies are spending roughly everything they generate on AI infrastructure, the next dollar has to come from somewhere: bonds, equity issuance, private credit, reduced buybacks, or suppliers stretching their own balance sheets.
The ECB's current ask is dull and necessary: better data, clearer definitions and more consistent treatment of open-ended private credit funds. Dull is fine. Financial accidents usually begin in places where everyone agrees the details are too boring for the front page.
For now, the central bank is in the warning business, not the intervention business. That is the right place to start. But if AI revenues do not rise fast enough to meet the debt being written against them, the argument will move quickly from transparency to containment. Nvidia's 2056 bonds are a useful marker. The AI boom is borrowing against a future three decades away, and regulators are right to ask who carries the loss if that future arrives late.
Also read: Hyperliquid's HYPE token hits $76.70 as SpaceX futures make the case for permissionless finance • GCash's parent Mynt wins approval to file for what would be the Philippines' largest-ever IPO • Kuaishou's Kling AI is spinning out at a $20 billion valuation and the revenue numbers make that hard to argue with
