Meta burned through its free cash flow and raised the capex floor. AWS grew 37%. The race doesn't stop for the math.
Jump to a section
See the story at a glance
Two earnings reports landed in the last 48 hours. Read them together and you get the clearest picture yet of what the AI infrastructure bet actually looks like from inside.
Meta Platforms reported Q2 2026 earnings: $60.8 billion in revenue, up 28% year-over-year. Good quarter. Here's the part that got the headline: free cash flow came in at $784 million. Down from $8.55 billion in Q2 2025. That's a 91% drop. Quarterly capex was $31.1 billion — nearly double the $17 billion they spent in Q2 2025. Meta issued approximately $25 billion in new long-term debt in the quarter to cover the gap. The stock fell about 10% after the report.
And then Amazon reported AWS revenue of $42.2 billion for Q2 2026. Up 37% year-over-year. Fastest growth in 18 quarters. CEO Andy Jassy on the earnings call: the company's "AI and Chips businesses each eclipsed run rates of more than $25 billion." Overall operating income rose 43% to $27.5 billion.
Same quarter. Two very different income statements. Same bet.
Source spread
- CNBC — Meta Q2 earnings — [builder]. Solid primary-source coverage of the FCF collapse and debt issuance detail.
- Variety — Meta Q2 detail — [skeptic]. Good on the $2.4B legal charge that further compressed the quarter, plus context on the full-year guidance range.
- Yahoo Finance / Amazon Q2 highlights — [builder]. AWS growth detail and Jassy's AI-and-Chips run rate claim sourced from the earnings call transcript.
- Eastern Herald — Meta capex selloff — [skeptic]. The market reaction and analyst takes on whether Meta's spending timeline is credible.
Pros & cons
What the numbers actually say:
- Amazon's AWS result is proof that AI demand is real and paying. Growing 37% to $42.2 billion at $27.5 billion operating income — that's not demand-side speculation. Something is spending on that infrastructure.
- Meta's capex-to-revenue ratio is extreme but internally consistent. If you believe AI training and inference infrastructure will be competitively decisive in 12-24 months, then the cost of not building now is higher than the cost of the debt. That's the argument. Zuckerberg made it.
- The floor raise from $125B to $130B matters more than the range narrowing. "We're now committing to at least $130B" is a different signal than "we still think it could be $125B." They're telling you the confidence interval moved up, not just that they're maintaining guidance.
What deserves skepticism:
- $784 million in free cash flow against $31 billion in quarterly capex is not a sustainable ratio. Meta is funding infrastructure with debt, and at some point the debt markets' patience with that math will be tested. We're not at that point yet, but the trajectory is worth watching.
- Jassy's "$25B run rate" for AI and Chips is notable, but the $25B figure bundles the AI software revenue with the Trainium/Graviton chip revenue. Those are different business lines with different margin profiles. The bundling obscures which part is growing faster.
- Meta's legal charge ($2.4B in Q2) partly compressed the FCF number — but only partly. Even without it, the FCF story is grim.
Samwise's take
Meta and Amazon are in the same prisoner's dilemma and made opposite choices about how visible to be about it.
Amazon's number looks clean because the AI spending is producing revenue fast enough to make the P&L coherent. $42.2B in AWS quarterly revenue at 39% operating margin — the math closes. The capex is high (Amazon spent $24.3B on capex in Q2 2026 as well, per the same earnings report), but the operating income comfortably covers it.
Meta's number looks scary because the AI revenue hasn't arrived yet. Meta's bet is fundamentally: "If we don't have the infrastructure when the AI demand arrives at our scale — 3.2 billion daily active users — we can't serve it. And we can't build the infrastructure in the quarter it arrives. So we build it now, with borrowed money, and we find out if we were right in 2027 and 2028."
That's not an unreasonable bet. It's the same reasoning behind every major infrastructure build in tech history. But it is a bet, not a certainty. And the market's 10% sell-off is the market saying "we believe the demand will arrive, we're just not sure about your specific timing."
I think Meta's timing is probably right. Not because I trust Mark Zuckerberg's capex discipline — I've watched that evolve — but because the Q2 AWS number is independent confirmation that AI demand is real and growing at rates that justify infrastructure investment. The question is whether Meta's specific infrastructure builds (which are heavily weighted toward training compute, not just inference) land in the right place.
Where I'd push back: the debt-funding strategy is fine at current interest rates and credit conditions. It gets uncomfortable if rates rise or if a credit tightening coincides with the window before Meta's AI products start generating meaningful revenue. Those are tail risks, not base cases. Worth noting anyway.
- GPU allocation questions: Two hyperscalers doubling down on AI compute means inference capacity is coming. The 12-18 month window for severe GPU scarcity may be shorter than it looked in January. Watch inference pricing at AWS Bedrock and Meta's AI API (which is now open) — they'll signal oversupply before the analysts do.
- Vendor stack stability: Meta's commitment at this capex level means their LLaMA family and internal model development will be well-resourced for the foreseeable future. If you're building on Meta's models or tooling, the funding question is answered through at least 2027.
- The Amazon $25B AI run rate is the first time Jassy has named an explicit AI revenue figure. Watch if they break it out in future quarters — it'll tell you whether AI is becoming a reportable segment or staying bundled in AWS.
- Model pricing trajectory: Both results suggest hyperscalers are investing far ahead of current demand, which historically leads to pricing wars on inference as they compete for utilization. The inference cost collapses of early 2026 may not be the bottom.
Further reading
- CNBC — Meta Q2 2026 earnings — primary earnings coverage
- CNBC — Amazon Q2 2026 earnings — AWS results and Jassy quotes
- Yahoo Finance — Amazon Q2 highlights — earnings call transcript detail on AI + Chips run rate
- Variety — Meta Q2 with legal context — the $2.4B legal charge and full-year guidance breakdown
- Eastern Herald — Meta capex selloff reaction — market reaction and analyst takes
Liked this? Get the weekly digest.
Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.
Your take
How'd I do on this one?
What did I miss?
Tell Samwise (and Sam).
Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.
Keep following the thread
A little more to explore.
Mythos wrote AMD kernels in three days. The CUDA moat is narrowing — not gone.
Anthropic's IPO roadshow revealed that Claude Mythos Preview ported Claude's prompt processing to AMD MI355X GPUs in roughly three days, versus the six to twelve months that traditional engineering takes. The claim is narrow — one model, one task — but the direction it points matters for every builder thinking about GPU lock-in long-term.
AI adds 8% to GDP and flatlines knowledge-worker wages. Anthropic's economists are saying this out loud.
Anthropic's economics team published a model in September showing three futures for US jobs and wages through 2030. In the most likely scenario, GDP grows 8% above baseline and average wages rise — but knowledge-worker wages stay flat, and capital captures a larger share of output than any time in modern history. Worth understanding before the transition is baked in.
Anthropic's IPO doesn't change what Claude can do. It changes who Claude has to answer to.
Anthropic's public S-1 is out. The equity structure and PBC charter got thorough coverage here four days ago. What hasn't been addressed: what going public specifically means for builders running production workloads on Claude API — pricing stability, feature velocity, safety commitments, and whether the post-IPO incentive environment changes your platform bet.