Everything About AI
NextFin News - Apollo Global Management President Jim Zelter says it will take "all the markets" to fund the artificial-intelligence capital spending boom, and warns that the resulting financing gap could push hyperscalers to account for 10% to 15% of the investment-grade bond market while interest rates stay "higher for a while."
The comments, made in a television interview on Wednesday, arrive as the Federal Reserve meets this week with markets split on whether it will raise rates for the first time since July 2023. Zelter's message ties together the two dominant forces now moving through credit markets: a capital expenditure cycle with no close precedent in size or speed, and a rates regime that is no longer reverting to the post-crisis norm investors relied on for a decade and a half.
The Funding Stool: Three Legs, One Massive Gap
The scale of the AI buildout has repeatedly outrun forecasts, and the revisions are coming faster than the spending. The five largest hyperscalers - Amazon, Alphabet, Meta, Microsoft and Oracle - are expected to spend $751 billion on capital expenditure in 2026, an 83% increase over 2025, according to Goldman Sachs strategists. That figure stood at $546 billion at the start of the year and $673 billion when the current earnings season began, meaning consensus forecasts have risen by nearly $80 billion in a matter of months. The target keeps moving because the denominator keeps growing.
Zelter framed the funding math as a three-legged stool, and his description makes clear why no single source is sufficient.
"A portion of it will be funded by the cash flow of these companies. A few hundred billion will be funded by the public investment-grade markets. And our stance is private investment-grade capital will be the third leg of that funding stool."
Even with all three legs, a shortfall remains. Zelter put the gap between hyperscalers' projected capital spending over the next five to six years and the amount expected to be raised from equity and public debt markets at $1 trillion to $1.5 trillion. In a separate estimate, he has said U.S. data centers alone could require $5 trillion to $6 trillion over five years, while a broader AI infrastructure buildout could demand $5 trillion to $7 trillion over five to seven years. Those figures sit alongside Goldman Sachs Research's baseline estimate of roughly $765 billion in annual AI capital expenditure in 2026, growing to $1.6 trillion in 2031, with cumulative spending of about $7.6 trillion across compute, data centers and power.
The arithmetic has a direct market consequence. Hyperscalers could account for 10% to 15% of investment-grade issuers, placing increasing strain on a market whose issuer mix was built for a different era. To put that concentration in context, Goldman Sachs Research estimates roughly $250 billion of direct hyperscaler debt supply in 2026, assuming about one-third of capex is debt-financed. That is incremental supply landing in a market already absorbing elevated Treasury issuance and a corporate refinancing wave as maturities roll into a higher-rate world.
This is not a cyclical funding squeeze that unwinds when the next earnings cycle turns. It is a regime change in the composition of corporate credit - the moment an asset-light technology sector becomes an asset-heavy infrastructure owner, with the balance sheets to match.
Why Rates Stay Higher for a While
Zelter's "higher for a while" rate stance is the other side of the same coin, and the bond market has already priced much of it. The 10-year Treasury yield is trading close to 5%, and the 30-year yield touched 5.31% in mid-August, its highest level since June 2007. The two-year note has risen about 20 basis points since Fed Chairman Kevin Warsh's hawkish Jackson Hole speech last month, and crude oil futures have climbed back above $100 a barrel on renewed escalation in the Middle East - adding a cost-push element to an inflation picture that has stayed above target for more than five years.
The near-term policy path is genuinely contested, which is itself the point. In a survey of 93 economists, 70% expect the federal funds rate to remain in the 3.50%-3.75% range at the September 15-16 meeting, down from 90% in August. Only 56% see rates on hold for the full year - less than half the margin that existed in recent months - while the share expecting at least one hike has more than doubled. Among primary dealers the split is tighter still: half expect no change this year, ten expect at least one increase, and one, Jefferies, expects a cut.
But the Fed decision is the smaller story. The deeper driver is a synchronized global repricing of the price of long-duration government borrowing. Long-dated yields have risen in lockstep across the United States, Japan, Germany and the United Kingdom: Britain's 10-year gilt yield rose to 5.25%, its highest since 2008, Japan's 10-year yield crossed 3% for the first time since 1996, and German bund yields hit their highest level since 2011. When every developed-market curve steepens at the long end in the same month, the explanation is not one central bank's error. It is a repricing of fiscal supply, term premium and the inflation risk premium - and all three point to rates staying elevated well beyond the next FOMC meeting.
The mechanism matters. A term premium is, in effect, a fear tax on holding long-duration risk: it is the extra yield investors demand for the uncertainty that inflation, deficits or supply will erode the real value of a 30-year bond. That tax does not fall quickly. Once investors have been burned by a decade of below-consensus inflation forecasts, they do not return to duration on the strength of a single soft print. They require a higher structural premium, and that premium shows up in every corporate bond priced off Treasuries - including the hyperscaler debt that must fund the AI buildout.
The Second-Order Effect: Who Gets Crowded Out
The first-order read of Zelter's warning is straightforward: hyperscalers will issue more bonds. The second-order effect - the one that matters for the rest of the market - is who loses access. A corporate bond market that must absorb $250 billion of incremental hyperscaler supply has less room for everyone else. Investment-grade issuers outside the technology complex face wider spreads, tighter allocation windows and a higher bar for entry, because the same pool of IG-mandated capital is now being asked to fund a larger, more concentrated borrower base.
That dynamic is already visible at the margin of the credit market. Apollo has raised its internal hurdle rate over the past year, even as it positions for the AI buildout through assets such as its majority stake in Stream Data Centers. Higher hurdle rates are the transmission mechanism in action: capital does not vanish, it is repriced and reallocated, and the repricing hits the lowest-quality borrowers first. The firms that lose out are not the megacap technology names with fortress balance sheets. They are the mid-tier IG issuers - the industrials, utilities and regional financials - that priced their capital plans on a deep, cheap market that is becoming shallower and more expensive.
Zelter nonetheless sees the financial system as resilient. He cited healthy banks and an active securitization market as reasons a deep recession is less likely - a framing that separates the credit channel from the growth channel. In that view, the risk is not a 2008-style freeze in which funding disappears entirely. It is a slower, more selective squeeze: funding remains available, but only at a price that kills marginal projects. That distinction is crucial for investors reading the signal. A freeze produces defaults. A squeeze produces missed guidance, delayed capex and multiple compression - a slower injury, but one that is harder to time.
The historical analog is imperfect but instructive. The dot-com buildout of the late 1990s was funded largely through equity, so the pain was concentrated in stock prices when the cycle turned. The cloud buildout of the 2010s was smaller in absolute terms and occurred in a falling-rate world, so debt was cheap and abundant. The AI buildout is different on both dimensions: it is debt-funded at a scale an order of magnitude larger, and it is landing in a market where the cost of that debt has structurally reset higher. The combination - asset-heavy balance sheets plus structurally higher funding costs - is what makes this cycle's crowding-out risk distinct rather than merely larger.
The Counter-Thesis: The Market Can Absorb This
The strongest case against Zelter's crowding-out view is that the market is larger and more flexible than the warning implies. Goldman Sachs Research estimates roughly $250 billion of direct hyperscaler debt supply in 2026, alongside a global AI investment base approaching $1 trillion. Project finance, private credit and securitization provide funding channels that did not exist at prior capex peaks, and they can absorb supply that the public IG market cannot. Apollo itself is betting on that architecture - private investment-grade capital as the third leg of the stool, not an afterthought.
The counter-argument also runs through returns. If AI infrastructure generates the cash flows that justify today's spending, the debt is self-liquidating and the IG market simply grows to accommodate it. New issuance does not crowd out existing borrowers if the market's capacity expands with the new borrowers' creditworthiness. In that scenario, hyperscalers are not parasites on the IG market; they are its next anchor tenants, and the concentration is a feature of a maturing financing ecosystem rather than a bug.
But that optimistic case rests on a hinge that Zelter has explicitly questioned.
"Just because companies need capital, doesn't mean they're all great investments."
Need is not the same as creditworthiness, and a market can fund a boom without the boom earning its keep. The dot-com era proved that capital can flow to projects with compelling narratives and negative returns for years before the reckoning arrives. The difference now is that the reckoning, when it comes, will be carried by bondholders as well as shareholders - and bondholders have less tolerance for narrative than equity investors do.
The falsifying signal is observable and will be published quarterly. If hyperscalers' share of investment-grade issuance stays below about 5% through 2027, and IG spreads for non-technology issuers do not widen by more than roughly 20 basis points relative to comparable peers, the crowding-out thesis fails. The concentration data will appear in issuance league tables; the spread data is available daily. Either way, the market will hand down a verdict within four quarters - no model required.
What Comes Next
In the short term, all eyes are on the Fed's September 15-16 decision and Friday's August CPI print, which is expected to show a 0.4% monthly rise after a 0.1% increase in July. A hotter number would lift the odds of a quarter-point hike - the first since July 2023 - and push the 10-year yield toward the 5% line that administration officials have signaled they do not want to cross. Year-over-year inflation is expected to hold steady at 3.4%, well above the Fed's 2% target, and PCE inflation is forecast at 3.5% for 2026 and 2.4% for 2027, unlikely to return to target before 2028.
Over the medium term, the key watch item is issuance concentration. Every hyperscaler bond deal through 2027 is a data point on whether the IG market is being reshaped or merely stretched. Private credit's ability to take the "third leg" of the funding stool - and price it correctly - will determine whether the squeeze is absorbed within the private markets or transmitted to the broader corporate market. Apollo's own hurdle-rate discipline is a leading indicator here: if the largest private credit platforms keep raising hurdles, the marginal borrower is already being priced out.
In the long run, this is a structural story, not a cyclical one. The AI buildout is turning an asset-light technology sector into an asset-heavy infrastructure owner, and that transformation does not reverse when the next earnings season disappoints. Rates may pause, spreads may tighten, but the composition of the investment-grade market has shifted - and the borrowers who depended on a cheap, deep IG market are the ones who will not get it back.
The AI capex boom is not a funding problem waiting for a solution. It is a repricing event, and the price being reset is the cost of capital for everyone else.
Explore more exclusive insights at nextfin.ai.
