Invest1 publisher3 min readPublished
Goldman's August AI Earnings-Bubble Warning Now Reinforced by Record $2.3 Trillion Investment-Grade Forecast
Peter Oppenheimer's new note argues that AI capex and government deficits are bidding for the same savings. His figures are issuance forecasts; Apollo's Torsten Slok is reading the same story in bond spreads that already trade wider.
The Investor · Invest desk
What happened
- Goldman's chief global equity strategist published a note on Thursday titled "Competition for Capital" that hardens his August warning about an AI earnings bubble while stopping short of saying the bubble exists.
- Capital spending among AA-rated technology issuers grew 65% year over year in the second quarter, the tenth consecutive quarter in which aggregate AA capex growth topped 35%.
- US convertible bond issuance has reached $135 billion year to date, and AI-related borrowers account for 44% of that volume.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- decision At roughly $575 billion of AI-related investment-grade supply in a single year, the buyer of that paper is choosing which projects get funded, and equity investors are downstream of that choice.
- constraint Because the bidding concentrates in long maturities, a data center with a twenty-year payback is quoted against sovereign issuance at the same tenor, so the hurdle rate rises on capex already committed.
- cost Absorbing 9.5% more corporate supply than Goldman previously forecast is paid for by existing holders of long corporate paper, in spread rather than in coupon.
- contradiction Oppenheimer's case rests on issuance forecasts and Slok's on spreads that have already moved, so a tightening in hyperscaler long bonds would strip the evidence from one of them and leave the other intact.
The chain runs through duration. Data centers, grids and government deficits all want money for a long time, so they bid against each other in long maturities. Oppenheimer's note opens there, citing 30-year German and Japanese yields near zero as recently as 2022 [9]. Torsten Slok, Apollo Global Management's chief economist, wrote the same thing five days earlier: "Data centers, power generation, transmission and government deficits are all long-duration claims on savings" [10][12].
A quarter of a $2.3 trillion investment-grade forecast is about $575 billion of AI-related supply in one year [7][1]. Convertibles add roughly $59 billion, the 44% share of the $135 billion issued in the US so far this year [6][2], about a tenth of the investment-grade number [5]. The $200 billion Goldman added to that forecast is around 9.5% more paper than it had previously expected the market to absorb [7][3].
Ten consecutive quarters of aggregate AA capex growth above 35% is two and a half years of it; held at that floor, spending would now be about 2.1 times where it started [5][4]. The second quarter came in at 65% [5]. Fortune describes the note as backed by capex-to-cash-flow data; the figures it reports are growth rates [8].
The two men are working from different kinds of evidence. Oppenheimer's central numbers are supply, issuance completed and issuance forecast [7]. Slok's are prices already printed. He said spreads on hyperscalers' longest-dated bonds have widened, and that "most of the paper issued in 2026 trades wider today than where it priced. Investors are still buying. They are just charging more" [13].
The step the supply numbers do not make on their own is the one from a higher cost of capital to a tech earnings problem, and six weeks after first raising it Oppenheimer still declines to say the bubble exists [3]. In August his evidence was Microsoft up 17% in a day, Meta down nearly 10% on a beat, and the equal-weighted S&P 500 outperforming the cap-weighted index by the widest margin since 2009 [14]. I think the credit half of this is checkable and the equity half is not yet. The counter-argument is decent: if AI issuers are a quarter of investment-grade supply [7], the investor pricing that paper and the investor pricing the equity are often the same institution, so a wider long end reaches the discount rate whether or not earnings ever disappoint.
The test is in the secondary market. If hyperscaler long paper tightens back inside its issue spreads while AA capex growth stays above 35%, the competition-for-capital argument loses the only part of itself that trades [13][5]. The cheaper way out is energy: if prices fall and policy rates follow, long yields can drop with capex growth still running at 65% [4][5].
What to watch
- Whether Goldman's credit team raises the $2.3 trillion investment-grade forecast again, and whether AI issuers move past a quarter of it.