The $570 Billion Footnote: Why AI's Real Story This Year Is in the Bond Market, Not the Earnings Call
Nvidia just raised $25 billion with $85 billion of orders. Morgan Stanley sees $570 billion of AI-related debt by year-end. The trade is not in the equities the bonds finance — it is in the spreads the bonds are setting.
The headline last Monday was that Nvidia had returned to the corporate bond market for the first time since 2021, pricing $25 billion of investment-grade paper in seven tranches stretching to 2056. The headline beneath the headline is more interesting. Orders for the deal hit roughly $85 billion — more than three times the size of the bond, the second-largest U.S. corporate investment-grade offering of 2026 by issue size, and a clear signal that institutional credit markets are no longer pricing the AI buildout as a hyperscaler equity story. They are pricing it as a credit story. The pivot is the trade.
The data was already pointing to it. Morgan Stanley estimates that AI-related debt issuance had reached $236 billion globally by May 31, four times the same-period figure for 2025. The bank's forecast, published on June 10, sees total 2026 issuance approaching $570 billion. AI-tied paper now constitutes roughly 14% of the JPMorgan U.S. Liquid index — the high-grade benchmark — up from 11.5% in 2020 and now larger than the U.S. banks sector, historically the index's heaviest weight. The arithmetic is simple: AI is no longer being financed at the margin. It is being financed at the index level.
The Mount Rushmore of Credit
Bloomberg Intelligence's tech credit team has been the loudest voice arguing that the issuance wave is structurally sound. Senior analyst Robert Schiffman made the case directly at the Bloomberg Global Credit Forum in New York on June 3, in a panel moderated by Paula Seligson.
"There is no AI bubble. The borrowing spree to fund the AI buildout is coming from the Mount Rushmore of credit."
— Robert Schiffman, senior tech credit analyst at Bloomberg Intelligence, speaking at the Bloomberg Global Credit Forum in New York, June 3, 2026.
Schiffman's "Mount Rushmore" framing is the structurally correct read. The Nvidia deal priced at AA-rated spreads of roughly T+35 to T+65 across the tranches, and the company's reasoning, per Reuters sources familiar with the transaction, was not capital expenditure financing but rather "to establish a liquid benchmark to its cost of credit." Translated: Nvidia, sitting on a projected $200 billion of free cash flow this fiscal year, did not need to borrow. It chose to, in order to anchor the curve at which AI-adjacent issuers can fund themselves. That is not the behaviour of a bubble issuer. That is the behaviour of an issuer setting the floor.
The JPMorgan Reframe
The macro framing has been led by J.P. Morgan. Tarek Hamid, the head of the bank's U.S. credit research and strategy team, raised the firm's 2030 AI infrastructure spending forecast in a June 16 client note to $5.5 trillion — up roughly $400 billion from the November 2025 projection. The composition matters more than the headline. JPMorgan estimates that approximately $4.1 trillion of the buildout will be debt-financed, with loans covering an average of 85% of total project costs. Hamid's team captured the analytical pivot in a single phrase that has since been quoted across the credit research community.
"Finding the right silicon financing paradigm is the billion-dollar question. The sheer scale of issuance tied to hyperscalers and data centers over the past year is rapidly reshaping bond markets."
— Tarek Hamid, head of U.S. credit research and strategy at J.P. Morgan, in a June 16, 2026 client note.
The "silicon financing paradigm" framing is doing real work. JPMorgan now forecasts $1.5 trillion of investment-grade bond issuance and roughly $150 billion of leveraged finance tied to AI and data centers over the next five years. That is not a tactical issuance window — it is a structural reweighting of the high-grade index toward a single thesis. The implication is that AI duration, AI spread risk, and AI-issuer credit selection are now core asset-allocation decisions for any institutional fixed-income mandate. The buy side is being repriced into the trade whether the buy side wants to be in it or not.
The Capacity Question
Morgan Stanley's global head of debt capital markets, Anish Shah, framed the absorbing-capacity question on the same Bloomberg June 3 panel, projecting that AI-tied issuance could approach 15% of all U.S. credit sales by year-end. The sequence of recent deals suggests the market is currently absorbing the supply without spread widening: Nvidia's $25 billion was three-times subscribed; Oracle, Meta and Alphabet have each priced multi-billion-dollar deals through Q2 at tight spreads; Amazon's most recent issue traded inside its outstanding curve on day one. None of this looks like a market struggling to digest supply. It looks like a market actively pulling supply forward.
The neocloud segment — smaller, GPU-rental-focused issuers like CoreWeave and Crusoe — is the second-order story. Morgan Stanley puts neocloud unsecured issuance at zero a year ago, $40 billion this year, with a further $20 billion expected by year-end. The neocloud tranche prices wider, in the BB-to-single-B range, but the demand-side data is consistent: investors are funding the buildout at every layer of the capital structure, not just the top.
The risk is not that the AI thesis is wrong. The risk is concentration. Nathaniel Rosenbaum and Erica Spear, JPMorgan strategists, noted in a separate June client note that AI issuers as a single thematic now exceed U.S. banks in the JULI index — and that the index, by construction, will continue to skew toward the thematic as deals price. A drawdown in AI-equity sentiment that triggered even modest spread widening in the issuer set would mark down a meaningful slice of every passive credit fund in the market. That is the asymmetry to monitor.
Our View
The trade is in the spread curve, not the equity curve. We see three discrete positions worth taking. First, long Nvidia 2031s and 2036s versus AA financials of comparable duration: the new Nvidia curve is mispriced relative to its rating quality and cash position, and we expect roughly 10 basis points of compression as it trades into the broader AA universe. Second, short the neocloud BB tranche: spreads are tight relative to operating risk and customer concentration, and any single hyperscaler GPU-procurement renegotiation would mark these wider in days. Third, watch the August earnings cycle: if hyperscaler capex guidance for 2027 is revised lower at the margin, the IG AI complex will widen 8 to 15 basis points and the trade is to be a buyer of that widening, not a seller into it.
The equity narrative around AI in 2026 has been the obvious one: spending up, earnings up, multiples up. The credit story is the one that has not yet been priced into broader portfolio construction — and it is the more important story. Nvidia did not need to borrow $25 billion. It borrowed it to fix the price at which everyone else borrows. The market's response — $85 billion of orders against a $25 billion deal — is the institutional credit market telling the equity market that the buildout is being capitalised, in the technical balance-sheet sense of the word, faster than the equity story has fully grasped.
This commentary is prepared for institutional and accredited audiences and does not constitute investment advice. Positions and views are those of Solomon Grey Capital and may change without notice.
