The AI Boom Has a Pricing Problem
Data centres are being built faster than anyone can agree on what the debt behind them is actually worth.
Much of that build-out is financed by hyperscalers, the small group of technology giants running cloud and AI infrastructure at global scale. Their debt is now reshaping credit markets far beyond their own balance sheets.
Hyperscaler-linked debt now accounts for more than 10% of U.S. investment-grade issuance. In the high-yield index, data centre issuance has gone from close to zero to over 3% - heading toward 5% - in under two years.[1] Almost all of it traces back to the same handful of hyperscalers funding the AI capex buildout.
This dynamic is most advanced in the U.S., where the AI buildout and private credit have grown together fastest. European data centre investment has grown too, but it's more diversified across tenants and less tightly bound to the AI capex cycle specifically.
Increasingly, that funding isn't coming from public markets at all. It's routed through asset-based finance and private credit, with asset classes built for diversification and yield now underwriting some of the most concentrated, capital-intensive infrastructure spending in a generation. One 2026 prediction from Credit Invest USA 2026 put it in stark terms: an “AI super cycle” sustaining 5% GDP growth, with private credit as preferred capital provider, could put the asset class on a path to $10 trillion over ten years. Another, from the same room, was more blunt: data centre and AI-related assets face a “correction.”
Both predictions can't be fully right. But both are being made by people underwriting this debt right now. This is exactly why valuation, concentration risk and reporting have become the pressure points investors and managers can't avoid.
[1] Source: Credit Invest USA - research call and on-site polling.
Hyperscalers explained
Hyperscalers are the small group of technology companies - cloud and AI infrastructure providers operating at global scale. Their data centre buildout is now the single biggest driver of credit issuance covered in this piece. Their capex spending is increasingly funded through debt, not cash reserves, which is what's pulling private credit and asset-based finance into the story.
The valuation dilemma
The AI capex cycle is creating private credit exposures faster than traditional valuation and reporting practices can reveal them.
Ask how a private credit deal gets priced, and you'll hear more than one answer. Some describe a shortcut version. Take a comparable public bond, apply its pricing matrix, add a spread. Others push back hard on that method, arguing real underwriting has to be built bottom-up from the asset's own fundamentals, not benchmarked against public bonds that may not be genuinely comparable in the first place.
Either way, the harder problem sits underneath: however the initial price is built, it's rarely tested against a live market.
It's compounded by a structural lag. Some experts have noted[2] that less liquid structured credit trades can take months to put together..., so when a market-wide repricing event hits, prices in these products can lag the move by two to three months. Not because the risk hasn't changed. But because there has been no new deal to reveal it.
There's a more subtle trap too. Two portfolios can look identical on paper while one is quietly built on payment-in-kind (PIK) loans rather than cash-pay ones. One manager's response has been to exclude PIK-heavy assets above a set threshold from eligible collateral entirely. A portfolio that stops generating cash is a fundamentally different risk than one that doesn't, even if the headline numbers match.
[2] Source: Credit Invest EU — panel discussion.
We spoke to Luca Blasi, Global Head of Private Markets & Regulatory Solutions at S&P Global, about why AI has raced ahead in credit origination and due diligence, but stalled when it comes to reporting and monitoring. He also shares the one question he thinks every allocator should be asking their manager about valuation methodology right now.
Concentration risk hiding in plain sight
Data centre issuers are tapping CMBS, ABS and corporate markets simultaneously, often backed by guarantees from the same small pool of hyperscaler names. Add it up across markets, and true exposure can run considerably higher than any single deal suggests.
It is not confined to public and structured markets either. Some direct lending strategies now carry as much as 20-30% exposure to software and AI-related credit[3], a concentration level several managers flagged as a live source of stress.
Even structured credit, designed to diversify and hedge exposure, isn't immune. Banks have started using significant risk transfer (SRT) transactions specifically on data centre lending portfolios. Less to free up regulatory capital, more to free up risk limits so they can keep lending into data centres at all. The tool built to manage concentration is now itself a channel for extending it.
The reporting gap
Investors polled on rebalancing between public and private credit in a sharp sell-off ranked mispricing risk between the two markets far ahead of every other answer - more than double the next-placed response, “selling liquid assets too quickly to raise cash.”
Asked separately what should be included in private credit stress-testing, the top answer wasn't any single factor - it was all of them: slower repayments, higher financing costs, lower secondary bids, and correlated redemptions together.
Meanwhile, a separate poll asked where AI is creating the most real value in credit today. Seventy-five percent said underwriting and due diligence. Just 17% said reporting and operations. AI is already reshaping how these deals get made. It hasn't yet reshaped how investors see what's happening inside them, once they're on the books.
In a sharp market sell-off, what’s the biggest mistake investors make when rebalancing between public and private credit?
Over-rotating into private credit for yield without reassessing risk
Selling liquid public assets too quickly to raise cash
Underestimating liquidity constraints in private portfolios
Failing to rebalance at all due to uncertainty
Mispricing risk between public and private markets
(ranking poll, showing weighted ranking score)
Where is AI creating the most real-value in credit today?
What comes next
None of this means the AI capex buildout shouldn't be financed through private credit. Diversification, granularity and access to deals otherwise out of reach are genuine advantages, and structured credit tools exist precisely to tailor risk for the investors who want it. The problem is that those same tools, built for a diversified, granular market, are now underwriting something narrower and more correlated, priced by methods that can obscure the risks investors most need to see.
That puts the burden back on allocators to treat valuation methodology as a live question for managers rather than background detail: how deals are marked, how PIK exposure is screened, how concentration is measured across markets rather than within a single one. Managers, in turn, need reporting and monitoring that can keep pace with underwriting decisions AI is already making faster than ever.
The AI capex cycle isn't waiting for either side to catch up. The advantage, going forward, may belong less to whoever can access the deal, and more to whoever can actually value it, monitor it, and explain it with independence and discipline.
S&P Global Market Intelligence provides independent valuation support, robust methodologies, and portfolio-level insight to help investors and managers assess hard-to-value private market assets with greater confidence. In a market where AI infrastructure debt is becoming larger, more complex and more interconnected, transparency is not optional. It is the layer that allows capital providers to understand what they own, how it is changing and where the risks may be building.
To learn more, visit our website
