In 2025, artificial intelligence moved from tech circles into everyday life. It showed up everywhere, from conversations at the dinner table, in questions about whether the next phone or washing machine would run on AI, and in emails from managers urging employees to learn how to use it.
Suddenly, the technology was no longer abstract. It was part of a daily conversation.
The companies driving that wave have seen their valuations surge at a pace rarely seen in modern markets.
Giants such as AMD, Meta and Tesla have become central pillars of the AI economy, while Nvidia alone now carries enough market weight that even small movements in its stock can ripple through global financial markets.
Yet amid the excitement over trillion-dollar valuations and breakthrough technologies, one question often goes unasked: who is actually financing the infrastructure behind the AI boom?
Interestingly, regulators have begun noticing too, though the warning line is easy to skim past.
“A few participants commented that the financing of the AI-related infrastructure buildout in opaque private markets warranted monitoring.”
The above line is taken verbatim from the minutes of the Federal Reserve meeting in January 2026.
While the financial markets celebrate massive valuations and traders book profits, the central bank is asking a simple, thought-provoking question: Who is actually financing the physical backbone of the AI boom, and what happens if that debt goes bad?
And it’s not just the Federal Reserve. Over the past six weeks, three powerful institutions — the Fed, the Financial Stability Board (FSB) and the US Treasury — have independently started probing that quieter side of the boom.
Taken together, their actions point to the same emerging worry: a growing share of AI‑infrastructure debt is being created in corners of the credit system that are hard to see, hard to value and hard to unwind under stress.
This investigation looks at what regulators have already put on the record, and why the supposed “secondary market for AI loans” appears, for now, to exist more in theory than in actual trades.
The ‘opaque corners’ of the AI boom
The Fed’s January minutes are the clearest public document linking AI infrastructure directly to private‑market financing risk.
In a section on financial stability, officials flagged elevated asset valuations and “vulnerabilities associated with the private credit sector,” including its growing role in lending to riskier borrowers and its ties to insurers and banks.
Against that backdrop, the line about “the financing of the AI‑related infrastructure buildout in opaque private markets” marked a shift: the central bank was explicitly connecting the AI capex wave to parts of the credit system that do not trade on screen.
At the global level, the FSB has been moving on two converging tracks.
In its 2026 work programme, the body said it would complete a dedicated report on private‑credit vulnerabilities as part of its work on non‑bank finance, and separately develop “sound practices for AI adoption, use and innovation by financial institutions.”
When Invezz reached out to the FSB for further details, the organisation pointed to risks on both fronts.
On AI, it highlighted a cluster of issues that “stand out for their potential to increase systemic risk”:
“AI‑related vulnerabilities that stand out for their potential to increase systemic risk include: third‑party dependencies and service provider concentration; market correlations; cyber risks; and model risk, data quality and governance.”
“GenAI also increases the potential for financial fraud and disinformation in financial markets. Misaligned AI systems that are not calibrated to operate within legal, regulatory, and ethical boundaries can also engage in behaviour that harms financial stability,” the global stability watchdog added.
On private credit, the same statement was blunt about data gaps:
“Private finance plays an increasingly significant role in the financial system by providing financing to corporates… Significant data gaps and the opacity within the sector have hindered a thorough assessment of the potential financial stability risks posed by private finance, and private credit in particular.”
“Concerns remain about the potential for a sudden stop in credit provision to corporates and the propagation of stress to the banking system or to institutional investors, given their interlinkages with private finance funds,” the FSB said.
The FSB added that it is “completing work on assessing vulnerabilities in private credit” as part of its 2026 agenda.
Taken together with the Fed’s minutes, the picture is clear: AI is driving a historic investment boom in infrastructure, and an increasingly important slice of that financing is flowing through a part of the system regulators admit they cannot yet fully map.
In simple terms, regulators are starting to pay attention to how the massive AI boom is being financed.
The Federal Reserve warned that some AI infrastructure is being funded through private markets that are difficult to track.
At the same time, global regulators say private credit markets lack transparency and reliable data.
Together, these signals suggest authorities are beginning to examine the hidden financing behind the rapid expansion of AI infrastructure.
A boom built off‑balance sheet
The scale of the AI infrastructure build‑out is not in doubt.
S&P Global Ratings’ 2026 liquidity outlook notes that technology and communications issuers, including the largest “hyperscalers,” have driven a surge in bond and loan issuance, much of it tied to data‑centre and AI‑related capital expenditures.
S&P estimates that maturities of US corporate debt rated “B‑” and below will climb from about $56.6 billion in 2026 to roughly $215 billion in 2028, creating what it calls “a sizeable refinancing wall” just as AI‑linked spending peaks.
At the same time, private credit has become one of the main channels through which smaller, non‑investment‑grade borrowers tap into that boom.
S&P’s analysis shows private‑credit lending to ‘B‑,’ and below‑rated borrowers reached nearly $146 billion in 2025, compared with about $85 billion in broadly syndicated loans to the same risk tier, and has exceeded syndicated issuance for four consecutive years.
Those numbers capture lower‑rated borrowers as a whole; they do not isolate loans tied specifically to AI infrastructure.
But their timing overlaps with what S&P describes as “tech issuance fuelled by AI and increasing leverage,” and with a rapid expansion of private‑credit assets under management more broadly.
Jeff Hooke, a senior lecturer at Johns Hopkins Carey Business School, recently shared with Invezz his peer‑reviewed study of 262 North American private‑credit funds together with Xiaohua Hu and Michael Imerman, which traces that expansion in detail.
The authors find that global private‑credit AUM has grown from roughly $375 billion in 2015 to about $1.6 trillion in 2023, with estimates from State Street suggesting it reached around $2 trillion in 2025 and could rise to $2.6 trillion by 2029, a nearly six‑fold increase in a decade.
The key difference from public bond markets is where this activity sits.
Instead of bonds trading daily on public exchanges, much of this credit is held in closed‑end funds whose valuations are updated infrequently and based largely on internal models.
In plain language, the rapid build-out of AI infrastructure is increasingly being financed through private credit rather than traditional public debt markets.
Much of this lending sits inside private funds where loans are rarely traded and pricing is not visible in public markets.
Because these investments are held privately and valued infrequently, it can be harder for regulators and investors to clearly see how risks are building inside the system.
Performance built on unrealised marks
Hooke and his co‑authors set out to understand what is actually driving reported returns in this world.
Using cash‑flow data from the Preqin database, they decompose private‑credit performance into Distribution to Paid‑In (DPI), cash returned to investors, and Residual Value to Paid‑In (RVPI), which captures the unrealised value of still‑held positions.
Their findings underscore why regulators worry about opacity.
For senior direct‑lending funds launched in 2015, roughly 30% of total reported value was still unrealised as of late 2024.
For the 2016 vintage, the share of unrealised value rises to around 50%, and for more recent vintages, RVPI accounts for over 80–90% of performance.
“We find that… a significant portion of private credit fund value is tied up in residual value even for older vintages,” the authors write, adding that this structure “represents a potential major risk” if those marks prove too optimistic when loans are ultimately liquidated.
The study also compares private‑credit funds to liquid market benchmarks.
Net of fees, senior and mezzanine private‑credit funds “barely outperform, or in some cases underperform” publicly traded floating‑rate ETFs such as the Invesco Senior Loan ETF (BKLN) and the VanEck Investment Grade Floating Rate ETF (FLTR), suggesting the higher opacity of private credit has not obviously translated into superior returns for investors.
More troublingly for an emerging, capital‑intensive sector like AI infrastructure, the authors point out that nearly half of direct‑lending borrowers have negative free operating cash flow, citing International Monetary Fund research, and that payment‑in‑kind income made up about 8% of interest income for business‑development companies in 2024, based on Fitch Ratings data.
Both features make it easier to smooth reported performance during stress.
Put simply, much of the reported performance in private-credit funds is not based on cash returned to investors but on the estimated value of loans that are still being held.
In many funds, a large share of returns remains unrealised even years after launch.
Researchers warn this could mask risks if those valuations prove too optimistic when loans are eventually repaid or sold.
‘Little secondary market for AI loans so far’
That background matters for AI because it shapes how quickly problems in a specific segment show up in reported numbers.
AI has already entered the private‑credit conversation once this year.
In February, UBS credit‑strategy analysts warned that AI‑driven economic disruption could contribute to between $75 and $120 billion in new defaults across leveraged loans and private‑credit markets by the end of 2026, with private‑credit default rates potentially climbing toward 4% in a baseline scenario and doubling under a more severe outcome.
Their focus was largely on what generative‑AI tools might do to the revenues of existing software borrowers, the demand side of the AI story.
The Fed’s minutes, by contrast, are about the supply side: the financing of the data centres, compute clusters and digital‑energy systems that make those tools possible in the first place.
When Invezz asked Jeff Hooke whether the liquidity and valuation risks he has documented in private credit extend to loans financing AI infrastructure, he did not hedge.
“Yes… there is little secondary market for AI loans so far. Time will tell if the AI loans for infrastructure go into PIK mode or loan extension mode,” Hooke told Invezz.
It is only one sentence, but it crystallises several of the issues regulators have hinted at.
If there is “little secondary market” for AI‑infrastructure loans, as Hooke says, then the standard release valves in credit, selling positions to other investors, and hedging via liquid indices are largely absent.
If borrowers begin to struggle, the primary tools left are bilateral: extending maturities, relaxing covenants, or switching to PIK interest to avoid crystallising cash defaults.
Hooke’s own research shows how those tools have been used elsewhere in private credit to defer loss recognition and sustain headline IRRs.
Latent stress can sit inside funds whose reported performance is still dominated by unrealised marks.
From the outside, that can look a lot like stability, until it doesn’t.
In practical terms, if problems start emerging in AI-related loans, they may not appear immediately in the numbers investors see.
Experts warn that many AI infrastructure loans have little active secondary market, making them hard to sell or reprice quickly.
If borrowers run into trouble, lenders may extend loan terms or allow interest to accumulate rather than declare defaults, delaying when financial stress becomes visible.
Who is ultimately providing the money?
A second set of questions regulators are asking has less to do with specific loans and more to do with who is standing behind them.
On 6 February, the US Treasury, which chairs the Committee on Foreign Investment in the United States (CFIUS), issued a Request for Information on a proposed “Known Investor Program.”
The idea is to create a framework under which certain investors with established track records could be pre‑cleared, potentially streamlining reviews of future transactions while maintaining national‑security oversight.
For private credit, the RFI goes straight to a long‑standing blind spot: foreign limited partners (LPs) in private funds.
Under current rules, foreign investors who commit capital as LPs without obtaining control or special governance rights are not always subject to mandatory CFIUS review, even if the funds they back go on to finance sensitive US businesses or infrastructure.
Treasury’s consultation, open for comment until 18 March 2026, explicitly asks how the Known Investor Program should apply to such structures.
A policy analysis by the CELIS Institute notes that draft eligibility criteria under discussion would require a prospective “known investor” to have filed at least three covered transactions with CFIUS in the past three years and to meet strict criteria around sanctions and ties to “foreign adversary” jurisdictions, a bar that many passive LPs in private funds may not meet.
For now, the program remains under development.
No final rules have been adopted, and Treasury declined to comment beyond its published materials.
But the thrust of the questions it is asking, about visibility into fund investors, not just fund managers, shows that foreign‑investment screening is starting to look upstream, towards the capital that ultimately supports private‑credit vehicles.
In simple terms, regulators are not only looking at the loans funding AI infrastructure but also at who is providing the money behind those loans.
US officials are exploring new rules that could give them better visibility into foreign investors backing private funds.
The goal is to understand where the capital ultimately comes from, especially when it may finance sensitive infrastructure or technology projects in the United States.
The questions still hanging over AI’s invisible debt
None of the documents published so far by the Fed, the Financial Stability Board, or the US Treasury suggests that AI-infrastructure financing in private-credit markets is an immediate systemic threat.
Instead, regulators are raising a set of unanswered questions.
If more of the AI infrastructure build-out is being financed in “opaque private markets,” as Fed officials noted, how liquid are those loans in practice?
Jeff Hooke’s observation that “there is little secondary market for AI loans so far” suggests that liquidity may be limited.
Another concern is how stress would appear in fund valuations if borrowers run into trouble.
With some private-credit funds still reporting a large share of their value as unrealised, tools such as loan extensions or payment-in-kind interest can delay when losses become visible.
There are also broader questions about transparency. What portion of the trillions committed to private-credit strategies is actually financing AI infrastructure, and who ultimately provides that capital?
Until new initiatives such as the proposed Known Investor Program take shape, regulators may only have partial visibility into that chain of financing.
The post Nvidia, Meta, Tesla are worth trillions, but who funds this AI boom? appeared first on Invezz