Quarterly Earnings: Amazon Borrowed the Gap, Meta Contracted It Away
Amazon raised $79B of debt as free cash flow fell to $1.2B; Meta signed $107B of non-cancelable obligations. Inside the $238B commitment stack, halted buybacks, and the $364B AWS backlog.
Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.
TL;DR
Amazon is funding its AI buildout with debt: it raised $79 billion in 2026, lifting long-term debt to $119.1 billion, while free cash flow fell 95 percent to $1.2 billion. Amazon is the only hyperscaler financing the capex gap through the bond market; Alphabet self-funds and Microsoft relies on operating cash flow. AWS backlog nearly doubled to $364 billion, concentrating demand visibility in a handful of frontier AI customers.
Meta added roughly $107 billion of non-cancelable purchase obligations in one quarter, bringing total commitments to about $238 billion across GPUs, cloud capacity, and energy. Meta halted buybacks after repurchasing $12.75 billion a year earlier, redirecting shareholder returns to infrastructure. The obligations remove Meta’s option to spend less if advertising, its only revenue engine, weakens.
Capital structure, not capex scale, is now the deciding variable in AI infrastructure: Amazon borrowed its funding gap while Meta contracted its away, and the market priced the divergence — Amazon gained less than a point while Meta fell 8.6 percent despite both beating expectations. Combined 2026 hyperscaler capex approaches $700 billion; from 2027 that spending hits income statements as depreciation, a cost the market has not yet priced.
For two years the AI infrastructure debate fixed on a single axis: how large the hyperscaler capex numbers could get before the market flinched.
Amazon at $200 billion, Alphabet at $180 to $190 billion, Microsoft near $190 billion, Meta at $125 to $145 billion. Scale was the whole narrative.
That framing is now exhausted.
The cohort has converged on comparable spending, combined 2026 capex approaches $700 billion, and backlog has answered the demand question definitively.
What remains unresolved is more structural: not how much these companies will spend, but how they will fund it and which balance sheets can absorb a multi-year build.
Amazon and Meta answered that question in opposite ways.
Amazon financed the gap through the bond market, adding leverage.
Meta contracted it away through non-cancelable obligations, sacrificing flexibility.
Neither preserved its optionality.
That divergence, not the capex totals, is what the market priced when Amazon gained less than a point while Meta fell 8.6% despite both beating expectations.
Amazon: Free Cash Flow Went to Zero by Design
On the surface, Amazon delivered a standout quarter.
Revenue rose 17% to $181.5 billion, operating margin hit a record 13.1%, and diluted EPS of $2.78 beat expectations, boosted by a $16.8 billion pre-tax gain on its Anthropic stake.
Excluding that gain, operating income grew 29.6%.
The real signal was cash flow. Free cash flow fell 95% to $1.2 billion even as operating cash flow rose 30% to $148.5 billion, reflecting $44.2 billion in quarterly capex. Nearly all operating cash was reinvested into infrastructure.
The differentiation is at the silicon layer.
Graviton, Trainium, and Nitro reached a $20 billion annualized run rate, while Anthropic and OpenAI committed roughly 7 GW of future Trainium capacity, reinforcing Amazon’s proprietary-chip strategy.
Here the cohort diverges. Alphabet is self-funding, Microsoft relies on operating cash flow, and Amazon alone is funding the gap with debt. It raised $79 billion in 2026, increasing long-term debt to $119.1 billion.
Amazon has traded free cash flow flexibility for build speed.
AWS revenue grew 28% to $37.6 billion, while remaining performance obligations nearly doubled to $364 billion.
The backlog strengthens demand visibility but also increases reliance on a handful of frontier AI customers.
Meanwhile, AWS operating margin fell to 37.7% as accelerated infrastructure depreciation outpaced monetization.
Read the full Amazon analysis: Zero Free Cash Flow, $79B in New Debt and the Financing of AI Infrastructure
Meta: The Signal Is in the Commitments Footnote
Meta reported GAAP diluted EPS of $10.44, boosted by an $8.03 billion one-time tax benefit.
Excluding it, normalized EPS was $7.31, a 9% beat driven by advertising.
Revenue rose 33% to $56.31 billion, its fastest growth in nearly five years, as stronger ad pricing and engagement not user growth drove monetization.
The more consequential disclosure was contractual.
Meta added roughly $107 billion in non-cancelable purchase obligations, bringing total commitments to about $238 billion across GPUs, cloud capacity, and energy.
The significance is not higher spending but the loss of optionality to spend less.
The commitment structure highlights Meta’s priorities.
It is expanding custom MTIA accelerators with Broadcom while securing off-grid power and long-duration storage, signaling that power not capital is the binding constraint.
In AI infrastructure, energized megawatts determine deployment speed.
Capital expenditures rose 45% to $19.84 billion as full-year guidance increased to $125–145 billion.
Operating cash flow funded the spending, but free cash flow narrowed to $12.39 billion.
More telling, Meta halted buybacks after repurchasing $12.75 billion a year earlier, redirecting capital from shareholder returns to infrastructure while maintaining modest leverage.
Meta occupies a higher-variance position. Unlike Microsoft, Amazon, and Alphabet, which fund infrastructure with diversified cloud revenues, Meta relies almost entirely on advertising.
That creates a faster AI monetization loop but no fallback. If ad pricing weakens, there is no second revenue engine to support a $238 billion contractual load.
The investment case rests on a single demand curve.
Read the full Meta analysis: The $107B Quarter That Turned Buybacks Into a Liability
The Pattern: Optionality Is Being Contracted Away
The two quarters rhyme, but the common thread is not spending it's the loss of flexibility.
Alphabet turned to self-generated power, Microsoft capitalized component inflation, Amazon replaced free cash flow with leverage, and Meta locked in $107 billion of obligations.
Each turned uncertainty into fixed costs before returns were visible.
The constraint stack is the same for all four: advanced-packaging silicon, memory, liquid cooling, transformers, and grid interconnection.
Power remains the binding constraint, dictating deployment order.
As hyperscalers compete for the same scarce inputs, costs rise, driving repeated capex revisions, including Meta's memory-driven increase. Each capacity commitment makes the next increment more expensive.
For emerging markets, the implication is exclusionary.
Amazon's model adds a capital filter, requiring stable rates and deep bond markets, while Meta's adds a demand filter, relying on sustained pricing power from a single monetization engine.
As a result, buildout is concentrating in the United States, select European hubs, and parts of Asia-Pacific, where power, water, regulatory stability, and capital markets align.
Regions without secured grid capacity and financing access are increasingly priced out, reinforcing concentration rather than broadening global expansion.
What Resolves Next
Three signals will define the next several quarters.
First, whether Amazon’s free cash flow recovers against its debt trajectory.
Free cash flow should recover as AI workloads monetize and the build cycle normalizes; whether that lands in 2026 or 2027 determines how long Amazon carries elevated leverage.
If capex stays near $200 billion and debt issuance resumes, the market will treat the balance sheet, not demand, as the constraint.
Second, whether Meta’s monetization layer expands fast enough to service commitments that are now contractually fixed.
Roughly ten million weekly conversations across its business AI agents is a usage signal; a revenue line tied to them would be thesis confirmation, and its absence through 2026 would compress the multiple further.
Third, the depreciation curve from 2027 forward, when the combined effect of the cohort deploying roughly $700 billion in 2026 capex hits income statements simultaneously.
That accounting expression of the current cycle has not yet been priced.
The Bottom Line
The demand story is settled. A $364 billion backlog at Amazon and a $238 billion contractual load at Meta confirm it.
What remains is whether capital structure compresses returns faster than AI workloads monetize.
Capital structure was the first deciding variable in the Microsoft and Alphabet prints.
Amazon and Meta showed the two ways a balance sheet absorbs it: borrow the gap, or contract it away. Neither is free, and neither travels.
Funding is key. Also worth noting that the data center debt that is rated based on the hyperscaler long-term commitments, and typically sits in the notes to their financials is vulnerable to the continued strong performance of the obligors. Pressures on the grid, souring public opinion, and the perceived growing gap between capex and ROI are storm clouds gathering. I publish The AI Grid Report (the latest edition discusses these pressures https://www.theaigridreport.com/p/the-overbuild-is-the-product?r=af3i2&utm_campaign=post&utm_medium=web).
Hyperscalers have raised $200-300B billion in debt to fund AI capex but they're leveraging free cash flow and equity raises. Free cash flow is significant across the hyperscalers (~$500B) but the financing gap is still quite large ($400-500B). I expect to see hyperscalers continue to tap public and private markets to fund their AI capex requirements over the next 3-5 years.