Six AI hyperscalers issued USD 244 billion in bonds in the first half of 2026, fourteen times the 2024 total. Goldman's derivatives desk called conditions carnage. The pain threshold dropped from USD 75 billion to USD 25 billion. Leverage doubled in two quarters and surpassed the entire energy sector. Goldman's own math shows the market has USD 510 billion of remaining capacity against USD 5.8 trillion in projected AI capex. The window may be closing. Here is what that means.
Goldman Sachs internal traders and strategists issued warnings this week that are now public. ZeroHedge, BigGo Finance, Futunn, and Macrostream all confirmed the same underlying Goldman data within the past 24 hours. This article is Fides Polonia's analysis of those disclosures. All quoted figures are sourced from Goldman Sachs research reports as reported by the outlets cited.
Brian Garrett, head of derivatives trading at Goldman Sachs, confirmed that credit trading desks have seen what he described as carnage. Goldman Sachs investment-grade bond trader Jeffrey Papai stated bluntly in a report that the USD 75 billion of AI-related bond issuance over the past month has left the market struggling to absorb the supply, with the spread on Goldman Sachs' AI bond basket widening by 22 basis points in a single week.
A 22 basis point spread widening in a single week in investment-grade credit is a significant event. Investment-grade bonds are supposed to be the most stable, least volatile part of the fixed income market. Widening spreads mean prices are falling and yields are rising on existing bonds as the market reprices the risk of holding them. When that happens to investment-grade paper in a week, something structural is happening, not just a temporary digestion issue from one large issuance. Papai further highlighted that market absorption capacity is shrinking rapidly. Previously, over USD 75 billion in supply was needed to stress the market. Now just USD 25 billion is sufficient to put the market on the defensive. That is a threefold reduction in the market's tolerance for new supply. The same volume of bonds that would have been absorbed comfortably six months ago is now causing spread widening and subscription ratio deterioration.
According to the Wall Street Journal, six companies classified as AI hyperscalers, Alphabet, Amazon, Meta, Oracle, Nvidia, and SpaceX, have collectively issued approximately USD 244 billion in bonds across global debt markets this year, more than double the USD 108 billion issued throughout last year and over fourteen times the USD 17 billion issued in 2024. The pace of that escalation is the most important data point. USD 17 billion in 2024. USD 108 billion in 2025. USD 244 billion in just the first six months of 2026. The trajectory is not linear. It is exponential. And exponential growth in bond issuance from a concentrated group of issuers has a mathematical endpoint determined by market absorption capacity, which is finite.
The share of AI-related bond supply in the overall investment-grade market has surged from approximately 1% in 2024 to around 7% in 2025, and reached roughly 18% in the first half of 2026. For long-duration issuance of 15 years or more, this proportion reaches as high as 42%. These concentration figures are extraordinary for a sector that effectively did not exist as a credit market category two years ago. When 18% of all investment-grade bond supply comes from six companies in a single theme, and those companies are all simultaneously competing for the same pool of institutional investor capital, the market is structurally vulnerable to a supply shock from any one of them. When Amazon alone issues USD 37 billion in a single transaction, it consumes the marginal buying capacity of every major investment-grade bond fund simultaneously.
| Year | Hyperscaler Bond Issuance | Share of IG Market | Market Pain Threshold |
|---|---|---|---|
| 2024 | USD 17B | ~1% | USD 75B+ needed to stress |
| 2025 full year | USD 108B | ~7% | Tightening |
| H1 2026 | USD 244B | ~18% (42% of long-duration) | USD 25B now stresses market |
| Next 12 months need | ~USD 1 trillion | Would be 30%+ of IG market | Window described as shut by Goldman |
It is hard to remember a larger disparity between price and sentiment within investment-grade credit. The messaging from credit investors is increasingly clear that it will be very difficult to fund another USD 360 billion in the next twelve months in the same manner. That observation from Goldman's Papai sets the immediate financing constraint: the bond market can probably absorb another USD 360 billion from hyperscalers over the next twelve months at most. The hyperscalers need approximately USD 1 trillion over the same period to fund their declared capital expenditure programmes.
Goldman Sachs credit strategy head Amanda Lyman explicitly stated that if the debt levels of hyperscale compute providers were raised to match those of large US banks, theoretically only about USD 510 billion of incremental debt space could be generated. Meanwhile, Goldman's equity research team estimates that the top five hyperscale compute providers will spend a combined USD 5.8 trillion on AI capital expenditures between fiscal years 2025 and 2030. The gap between USD 510 billion in maximum theoretical additional debt capacity and USD 5.8 trillion in projected AI capex is USD 5.29 trillion. That gap has to be funded somehow. If bond markets cannot absorb more debt at acceptable rates, the alternatives are equity issuance that dilutes shareholders, cutting capex programmes that disappoint AI narrative investors, or accepting higher borrowing costs that reduce the return on AI investment.
Hyperscaler net debt reached USD 239 billion in the first quarter of 2026, surging approximately 190% year over year. Morgan Stanley data shows that the aggregate leverage ratio of hyperscale computing providers has surged from 0.9x in the third quarter of 2025 to 1.8x currently, doubling in just over two quarters, and has now surpassed the leverage levels of the entire energy sector, continuing to rise at a pace of roughly 0.3x per quarter. At 0.3x leverage addition per quarter, the hyperscalers will reach 2.4x leverage by end of 2026 and 3.0x by mid-2027. Energy sector companies at 3x leverage are typically flagged by credit rating agencies as candidates for downgrade review. Investment-grade companies with BBB ratings that face downgrade to high yield, the so-called fallen angel scenario, are forced sellers by institutional investors whose mandates prohibit holding below-investment-grade paper.
Goldman Sachs data shows that AI infrastructure-related stocks now account for 42% of the S&P 500's total market capitalisation and are expected to contribute 50% of S&P 500 earnings growth in 2026. That concentration means that a slowdown in AI capex spending does not stay inside the technology sector. It transmits immediately to semiconductors through Nvidia, TSMC, and ASML. It transmits to data centre construction and REITs. It transmits to power infrastructure through utilities and grid equipment companies. It transmits to cooling system manufacturers, fibre optic companies, and the entire supply chain that has been building capacity to serve hyperscaler demand. Half of the S&P 500's projected earnings growth in 2026 is predicated on AI capex continuing at its current pace. If the bond market forces a reduction in that capex, that earnings growth assumption does not materialise.
Apollo and Bank of America warn that if bond market liquidity is cut off, it will trigger a chain reaction from equity financing to layoffs, potentially dragging the S&P 500 into a correction. The specific mechanism is straightforward. Hyperscalers reduce capex to manage leverage. Nvidia's data centre revenue misses estimates. Nvidia stock falls 20%. Nvidia is 6% of the S&P 500 by weight. The S&P 500 falls 1.2% from that single event alone. Multiply across AMD, Broadcom, Marvell, and the infrastructure stack and the S&P 500 correction from a hyperscaler capex slowdown could be 8 to 15% before the multiplier effects through consumer sentiment and corporate capital spending are counted.
A global credit market dislocation of the kind Goldman's warnings describe would not leave Polish equities untouched. The WIG20 would experience risk-off selling alongside other emerging and developed market equity indices in a broad credit event. That is the mechanical channel. But the specific investment question for a Polish-focused investor is whether the underlying industrial demand story for the AI buildout, copper for wiring and cooling, precision bearings for robotics, energy for data centres, is damaged by a hyperscaler capex slowdown or merely delayed.
The honest answer is that a meaningful capex reduction by hyperscalers would delay rather than cancel the infrastructure buildout. AI data centres are being built because the demand for AI inference capacity is real, growing, and not going away regardless of whether Microsoft, Amazon, and Google can issue bonds at investment-grade spreads or must pay high-yield pricing instead. The cost of capital going up does not eliminate the economic rationale for the buildout. It delays some projects and raises the hurdle rate for others. KGHM, whose copper is the most fundamental input into AI infrastructure alongside silicon, is affected by the timing of that buildout rather than by its existence. A six-month delay in Google's next data centre campus is a six-month delay in copper demand from that specific project, not the elimination of that demand.
The Goldman disclosure is significant precisely because it is Goldman. This is not a bearish hedge fund writing a short thesis or a contrarian newsletter arguing that AI is overvalued. This is the firm that has been one of the primary underwriters of hyperscaler bond issuances using the word carnage to describe its own credit trading desk. When the underwriter of the bonds is telling you the market is struggling to absorb the supply, the information asymmetry that normally exists between bond issuers and the public has temporarily collapsed. You are getting the inside view from the people sitting closest to the transaction flow.
The structural math that Lyman and Papai lay out is the most important element. USD 510 billion in maximum additional debt capacity. USD 5.8 trillion in projected AI capex through 2030. That gap is not closable through bond issuance alone. The hyperscalers know this. They have been front-loading issuance precisely because they know the window was going to close. The question is what happens next. Three scenarios are possible. One: bond markets recover, spreads tighten, and the financing machine restarts at lower volumes. Two: hyperscalers reduce capex to manage leverage, accepting slower AI buildout in exchange for balance sheet stability. Three: equity issuance supplements debt, diluting shareholders but keeping capex programmes intact. Scenario one is what every AI bull is assuming. Scenarios two and three have meaningfully different implications for the equity market.
The credit-equity divergence that Goldman's Garrett identified, carnage on credit desks while equity volatility remains calm, is the most actionable observation in the entire disclosure. Credit markets have been the leading indicator in every major market dislocation of the past two decades. The equity market's calmness in the face of credit stress is not reassurance. It is delay. How long that delay lasts before the equity market reprices the same information the credit market is already pricing is the central timing question for anyone managing risk across asset classes right now.
This article is produced by Fides Polonia Capital Management for informational purposes only. All Goldman Sachs quotes and data are sourced from BigGo Finance, Futunn, Macrostream, and ZeroHedge reporting of Goldman Sachs research notes and internal trader communications as cited, all published July 9-10 2026. This article does not constitute investment advice or a recommendation to buy or sell any security. Fides Polonia Capital Management may hold positions in companies referenced. Nothing in this article should be relied upon as the sole basis for any investment decision.