Debt-funded AI capex makes the hyperscaler trade riskier than the chip trade
The AI trade is splitting into two very different bets: suppliers booking revenue now and hyperscalers borrowing against a payoff that may take longer to show up. That distinction matters more this week because the market is starting to care not just about AI demand, but about who is carrying the financing risk.
The market keeps talking about AI as if it is one trade. It is not. The sharper split now is between infrastructure vendors selling into a live buildout and hyperscalers underwriting that buildout with ever-larger capital commitments and, increasingly, external financing. That is why we think the contrarian read is the right one here: leverage changes the risk profile more for the cloud platforms than for the chip and networking names getting paid today.
The key fact is not that AI spending is big; it is that the funding mix is changing. Public reporting this month tied expected AI-related global debt issuance to nearly $570 billion in 2026, while the four biggest hyperscalers are expected to spend roughly $700 billion in outlays this year. That is a different setup from a clean software compounding story. When the buildout starts leaning on debt markets, hyperscaler equity becomes more exposed to rates, spreads, and the timing of monetization — risks that sit above and beyond whether AI demand is real.
That is why recent financing headlines matter. AMZN secured a $17.5 billion loan facility to support AI infrastructure, and GOOGL expanded a massive capital raise tied to AI infrastructure and global compute. Even NVDA announced a $25 billion bond sale, which shows the debt channel is spreading across the ecosystem. But the distinction is still important: Nvidia is selling the picks and shovels into a demand wave that is already visible in its numbers, while the hyperscalers are the ones making the multiyear return-on-capital bet. Suppliers can stumble on expectations, but hyperscalers are taking balance-sheet and duration risk in a way the market had not fully priced until recently.
The comparative fundamentals reinforce that split. NVDA is not cheap at 33.17x earnings, but it is also growing revenue 65.5% with a 63.0% net margin. That is what getting paid now looks like. MSFT, by contrast, trades at 22.67x earnings with 14.9% revenue growth and a strong 39.3% net margin, but the issue is not operating quality; it is that a larger share of the AI upside still depends on converting enormous capex into durable incremental revenue. The same tension shows up in GOOGL at 25.84x earnings and 15.1% revenue growth. These are still premium, durable businesses, but they are being asked to absorb infrastructure intensity more like utilities while still being valued like compounders.
A quick snapshot of the split helps:
NVDA: 33.17x P/E, 65.5% revenue growth, 63.0% net margin
AVGO: 45.37x P/E, 23.9% revenue growth, 38.8% net margin
Yes, bulls will argue the demand side is strong enough to justify the spend. Consensus-tracked cloud revenue for the major providers rose to $84.8 billion in the first quarter, up 39% year over year, which is real evidence that this is not speculative overbuild in the old dot-com sense. But that counter misses the market question in front of us. The debate is no longer whether AI demand exists; it is whether the payoff arrives fast enough to justify debt-assisted capex at this scale without compressing multiples for the buyers first.
That is also why the tape has looked confused. Reuters tied Big Tech weakness on June 25 to concern over debt-backed hyperscaler spending, while chip names have still shown they can rip when guidance is strong, as this week's rally in AI semis demonstrated. Even so, supplier risk is more cyclical than structural right now. AVGO can get hit hard if expectations outrun guidance — the post-earnings drop proved that — but that is different from the financing problem facing hyperscalers. A supplier de-rates when orders slow. A hyperscaler can de-rate even while demand remains healthy if investors decide the capex payback period is stretching and the balance sheet is doing more of the work.
The better historical analogy is late-cycle telecom and fiber, not software bubbles. Massive infrastructure waves often look brilliant on the demand line and messy on the financing line at the same time. Today’s hyperscalers are far better businesses than the overbuilt telecom operators of prior cycles, and this is not a credit-crisis call. But once annualized AI capex starts being discussed in the $750 billion to $850 billion range, with a path toward $1 trillion next year, the market has every reason to separate the companies monetizing the buildout immediately from the companies promising to monetize it later.
The cleanest way to frame the trade now is simple: the chip and infrastructure names are exposed to spending risk, but the hyperscalers are exposed to spending risk plus financing risk plus timing risk. That does not make NVDA, AVGO, ANET, or VRT safe. It does mean the burden of proof has shifted more heavily onto MSFT, AMZN, GOOGL, and META to show that AI capex is not just strategically necessary, but economically efficient on a timetable equity investors can live with.
What would change our mind? A clear acceleration in AI monetization at the platform layer, paired with evidence that capex can be funded comfortably from operating cash rather than increasingly through debt markets. Until then, we think the market is right to split AI into two buckets — and wrong to assume the buyers deserve the same risk premium as the sellers.
Our take, not advice. This is opinion commentary — informational only, not personalized investment recommendations. Markets carry risk. Do your own research and consider your own situation before any trade.
▌The Daily Briefing · Free
A new stock idea, every evening.
One stock worth watching each weekday, plus the analysis behind it. Free, in your inbox.
