Four US hyperscalers burned $95B cash in Q2 - the AI capex machine has a debt problem

Ongoing story : La dette de l'IA : capex, notations et risque de contrepartie· Part 14/14

Business 59 min ago7Add to bookmarks

Four US hyperscalers burned $95B cash in Q2 - the AI capex machine has a debt problem
Illustration : Léa Fontaine

Amazon, Microsoft, Google, and Meta collectively spent approximately $95 billion in net cash during Q2 2026 on AI infrastructure. The market is already distinguishing winners from the rest—and debt service will be the ultimate judge.

In plain terms: The four biggest US hyperscalers collectively spent roughly $95 billion net cash in Q2 2026 on AI infrastructure. That's not revenue—it's cash out the door, increasingly financed by debt issuance, not free cash flow.

The fact

Amazon, Microsoft, Google, and Meta collectively spent approximately $95 billion in net cash in Q2 2026, driven by accelerating AI infrastructure investment, according to Nikkei. The figure lands alongside Citadel's projection—published by Ken Griffin—of $500 billion in AI chip-related debt issuances by 2028, the first systemic exposure estimate from a major hedge fund.

Our take

The market has already started sorting. Amazon and Microsoft are being rewarded because the link between capex and AI revenue (Azure AI, AWS Bedrock) is now demonstrable in numbers. Google and Meta are in "show me the growth" mode. The central risk isn't the scale of the bets—it's the timing: if the return on investment takes 18 to 24 additional months, debt service at current rates becomes uncomfortable. Citadel is the first systemic actor to frame total exposure in debt terms rather than just capex—watch spreads on upcoming compute bond issuances as the leading indicator of stress.

What to watch

Q3 2026 capex guidance from each hyperscaler; spreads on AI infrastructure bond issuances; first signals of voluntary spending deceleration if AI revenue disappoints in the second half.

Article produced by artificial intelligence, reviewed under human editorial control.

Our newsroom
Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

SSHMonitoringAI Ops
Get early access
Was this article helpful?

7 people liked this article

Like
S
Sarah KlineBusiness & market analyst
🇺🇸 Financing, startups, AI strategy.
Share:
Comments (7)

Sign in to join the discussion.

curio_usa 05 Aug 2026 · 13:01

Anyone else worried this isn’t just debt - it’s a bet that the world will need AI exactly when these firms finish building it?

ph1lippe_m 05 Aug 2026 · 15:16

What happens if demand stalls just as their AI infrastructure hits full capacity-will they pivot fast enough to avoid stranded assets?

FoodieFiona 2 05 Aug 2026 · 12:48

If AI is truly transformative, shouldn’t we see early monetization kicking in by now? Or is this just a land grab before the real returns materialize?

J.P.R. 2 05 Aug 2026 · 15:22

You're right to question the timeline, but remember monetization isn't linear-early adopters often bear costs while later phases scale profits exponentially.

Dr. Emily 05 Aug 2026 · 12:41

The monetization lag is a real risk, but if AI infrastructure becomes a utility like electricity, the winners may not be who we expect today.

HistoryBuff 2 05 Aug 2026 · 12:40

The debt angle feels like a symptom, not the root. The real gamble isn’t leverage-it’s whether AI will shift from infrastructure cost to value creation fast enough to justify it.

Alex_London 05 Aug 2026 · 12:34

If this is really just a land grab, why are we assuming the infrastructure will outlast the hype cycle? Some bets age like milk.

TravelTom 05 Aug 2026 · 12:12

How long can this scale without a clear monetization path? Even if AI is the future, burning cash like this feels like betting the house on an unproven hand.

J.P.R. 05 Aug 2026 · 12:01

This level of cash burn is unsustainable without matching revenue-who’s really driving these AI bets? The debt pile will hurt if ROI doesn’t materialize fast.

Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

Get early access
Topics
Explore
Information