.png)
In 2008, a few analysts spotted the US housing market’s hidden risk, subprime mortgages were bundled, mislabelled as safe, and sold worldwide. When defaults surged, the system collapsed. The lesson? When everyone believes something is “too big to fail,” that’s often when danger emerges. Today, artificial intelligence is the new “AAA” a shiny asset attracting massive capital. Venture capital, corporate spending, and infrastructure investment are pouring in, all under the shared belief that AI will transform everything, the returns on data centre infrastructure, GPU capacity, and cloud build-out will justify the capital being deployed, echoing the psychology that fuelled the housing boom.
Just as in 2005-2006, when investors like Burry and Paulson had to build bespoke CDS to bet against subprime mortgages, JPMorgan’s ready-made hyperscaler basket signals that taking a bearish view on AI debt is now tradable.
The most obvious parallel is the sudden emergence of a market for credit derivatives on companies once considered almost risk-free. Data from the Depository Trust & Clearing Corporation (DTCC) shows that single-name credit default swaps (CDS) tied to majorUS tech firms surged 90% between early September and December 2025. A year earlier, CDS on companies like Meta and Alphabet barely traded at all. This rapid market development mirrors the mid-2000s, when mortgage-backed securities went from perceived safety to complex, layered instruments that amplified risk.
Source: Depository Trust & Clearing Corporation
The scale of debt driving this growth is equally remarkable. In autumn 2025, Meta, Amazon, Alphabet, and Oracle collectively raised $88 billion to fund AI infrastructure. Hyperscalers issued over $121 billion in US investment-grade bonds in 2025, compared with roughly $28 billion annually between 2020 and 2024. JPMorgan projects investment-grade tech companies could raise $1.5 trillion by 2030. Concentrating such large amounts of debt on a single theme, AI infrastructure, echoes the frantic collateralised debt obligations (CDO) issuance of 2005-2007, when capital flowed into one asset class without clear pricing of risk.
JPMorgan’s CDS basket highlights concentrated risk, with equal exposure to five companies Alphabet, Amazon, Meta, Microsoft, and Oracle each closely tied to AI investment outcomes. f the AI thesis weakens, there is limited diversification to offset losses, as performance across these firms is likely to move together. This dynamic echoes pre-crisis CDO tranches, where shared exposure to the same housing markets amplified losses.
In 2005-2006, the ABX index made shorting subprime possible, and falling levels triggered bank losses, forced sales, and wider spreads a self-reinforcing loop. In the hyperscaler CDS market, rising spreads could similarly increase borrowing costs for AI capex, challenging ROI assumptions. Narrative momentum amplifies the risk: just as housing relied on the belief that prices could not fall, today’s AI boom is driven by expectations of transformative returns, estimated by Goldman Sachs at up to $20 trillion, which reinforce investor confidence.
Circular financing loops and narrative risk. Tech giants are investing heavily in AI startups, which in turn become major customers for their cloud and chip services. This revenue reinforces the appearance of AI success, attracting further capital and fuelling a self-reinforcing cycle. While profitable today, the loop echoes pre-crisis CDO dynamics, the numbers can look strong until underlying assumptions break down. Narrative momentum also amplifies this effect, as confidence in generative AI’s transformative potential drives investment, much like the belief that housing prices could never fall.
Source: Bloomberg News reporting
Investors are taking notice. Specialist asset managers note that credit risk, once deemed negligible, now varies by company and warrants hedging. Some acted decisively, Benedict Keim of Altana Wealth shorted Oracle’s CDS in October 2025. Oracle’s five-year CDS later rose fourfold to roughly 198 basis points, reflecting a repricing reminiscent of early 2007, as sophisticated hedging emerged before the market fully grasped the risks.
Source: LSEG, Markit
Comparing today’s hyperscaler CDS market to the 2008 subprime crisis can be analytically useful, but the differences are substantial and they matter when assessing systemic risk.
Credit quality is far stronger. Unlike the subprime mortgages during the 2008 crisis, which were extended to borrowers with poor credit histories and little documentation, the debt referenced in JPMorgan’s CDS basket comes from five of the world’s most profitable companies. Alphabet, Microsoft, and Meta generate hundreds of billions in projected cash flow, while even Oracle, the weakest name, is a mature, profitable enterprise with recurring revenue. The probability of an outright default, what would trigger a CDS payout, is far lower than in 2008.
Source: MacroTrends
The regulatory environment today is fundamentally different. After 2008, Dodd-Frank required central clearing for most CDS contracts, dramatically reducing the counterparty chain risk that had allowed AIG’s $440 billion notional exposure to nearly collapse the global financial system. Basel III significantly increased capital requirements, the Volcker Rule limited proprietary risk-taking by banks, and stress testing became routine. These reforms directly addressed the mechanisms that amplified a housing correction into a systemic crisis. Today, counterparty risk in the CDS market is largely contained through central clearing a safeguard that simply didn’t exist in 2006.
Transparency is another major difference. Pre-crisis CDOs were notoriously opaque, investors often didn’t know which mortgages were in the tranches they owned, and rating agencies assigned AAA ratings based on flawed models. In contrast, the five companies in JPMorgan’s basket file detailed quarterly earnings, follow SEC disclosure rules, and are among the most closely monitored corporate borrowers in the world. Investors buying protection on this basket know exactly what they own. The opacity that caused systemic mispricing in 2007 is largely absent today.
Leverage dynamics are also less extreme. In 2008, AIG and a network of shadow banks sold huge amounts of credit protection without sufficient collateral, turning the CDS market into a hidden leverage amplifier. Today, the hyperscaler CDS market, while growing rapidly, is relatively contained. Oracle’s CDS alone averages over $830 million in weekly trades (Bloomberg, March 2026), substantial, but tiny compared with the synthetic CDO structures that nearly brought down the global banking system.
The launch of JPMorgan’s AI Hyperscaler CDS basket in early 2026 marks a significant shift in how financial markets are pricing AI-related risk. Historically, AI investments were largely an equity story, reflected in the valuations of technology companies, GPU manufacturers, and data centre operators. The emergence of a standardised credit instrument indicates that AI execution risk is now being systematically incorporated into credit markets, including leveraged loans, CLO tranches, and CDS baskets. This expansion in risk pricing means that if AI monetisation falls short, the impact won’t be limited to tech stocks it could spread across the entire capital structure, affecting credit spreads, loan markets, and structured products linked to hyperscalers.
The basket responds to several underlying factors. Hyperscalers, once largely debt-free and funded by cash flows and equity, have become major issuers in the investment-grade bond market, with JPMorgan projecting they could account for roughly 30% of net IG supply in 2026. This concentration creates pressure on spreads and drives genuine institutional demand for hedging instruments. Single-name CDS on companies like Meta and Alphabet were largely illiquid until late 2025, and the basket fills this gap, allowing banks and asset managers with large tech exposures to manage portfolio concentration risk efficiently.
Source: Facset (Long Term Debt)
At the same time, the product reflects a growing recognition of real AI credit risk. Massive AI infrastructure spending projected to exceed $600 billion industry-wide, with hyperscalers alone potentially accounting for $400 billion creates structural funding gaps that cash flows cannot fully cover. If AI returns fall short due to demand shocks, competitive disruption, or regulatory challenges, these companies’ debt-funded expansion becomes vulnerable. The basket acts as insurance against such scenarios rather than signalling imminent default. JPMorgan’s scenario analysis assigns a 40% probability to AI success, 45% to a moderated slowdown, and 15% to a systemic shock, demonstrating that even the originator does not view AI as risk-free.
In the end, the AI Hyperscaler CDS basket is not a warning of impending crisis, it is a clear marker of market maturity. Hyperscaler debt now represents a measurable share of major investment-grade indices. A sharp rise in protection demand could therefore trigger index rebalancing and forced selling, pushing up borrowing costs, slowing capital expenditure, and transmitting mild headwinds to broader economic growth. The existence of this liquid, standardised instrument shows that credit markets are now approaching the AI infrastructure boom with rigorous, professional scrutiny. They recognise its immense potential but are equally focused on managing the risks tied to such a capital-intensive cycle. In other words, while the gold rush is still in full swing, sophisticated investors have already started quietly building their exits.