The AWS Q2 number requires context to land correctly. AWS had been growing in the high-20s for several quarters while Google Cloud printed 48%, 63%, and then 82% in sequence. The narrative that had built was one of AWS losing ground to Google Cloud and Azure inside AI workloads. Q2 challenges that framing directly.

A 9-point acceleration in a single quarter — from 28% to 37% — is unusually large for a business at $42 billion quarterly revenue. The explanation Andy Jassy gave on the call: demand for AI infrastructure continued to strengthen, and AWS capacity constraints that had been limiting new workload starts in Q1 began to ease as new data center capacity came online. The $54.2 billion in Q2 capital expenditure that investors flinched at is also what enabled the capacity release that produced the 37% number.

The AI and chips units disclosures are the most forward-looking signal in the print. Both businesses eclipsed $25 billion annualized revenue run rates — more than doubling from a year ago, per Jassy. AWS’s in-house chips (Trainium for training, Graviton for general compute) give it a margin lever that neither Google Cloud nor Azure has at the same scale. Custom silicon at $25B run rate makes AWS’s infrastructure economics structurally different from a pure Nvidia-dependent stack.

Meta’s Q2 2026 print contains two separate earnings reports that almost every headline merged into one. The advertising business delivered: $59.4 billion in ad revenue, up 27%, on 14% impression growth and 12% average price per ad growth — both metrics accelerating versus Q1. Family daily active people reached 3.60 billion, up 3%. Revenue of $60.8 billion beat the $60.2-60.3 billion analyst consensus. This was not a weak business print.

The other report: $31.1 billion in Q2 capital expenditure consumed 98% of the $31.86 billion in operating cash flow Meta generated in the quarter. Free cash flow fell 91% year over year to $784 million. Meta paid $1.35 billion in dividends — more cash than it had in free cash flow. The company issued approximately $25 billion in new long-term debt in Q2 to fund construction. This is a company borrowing money to pay a dividend while spending nearly all its operating cash on data centers.

The full-year 2026 capex guide narrowed to $130-145 billion from $125-145 billion, meaning management tightened the floor — this is more spending certainty, not less. At the midpoint of $137.5 billion, and assuming operating cash flow runs at the H1 pace for the full year, Meta is structurally FCF-neutral or negative for 2026 as a whole. Zuckerberg’s message was that AI is accelerating the core business and opening enterprise opportunities. The credit market’s answer was higher yields on Meta’s debt issuance — its own way of pricing AI infrastructure risk.

The 45-point growth rate gap between Google Cloud (82%) and AWS (37%) is the single most strategically important number in the Q2 cloud cycle. AWS is still considerably larger in absolute revenue — $42.2 billion versus $24.8 billion. But if Google Cloud sustains 70%-plus growth while AWS runs in the mid-30s, Google Cloud closes the absolute revenue gap from roughly $17 billion today to less than $10 billion in four quarters. That is a different competitive map than anything the market was pricing in early 2026.

The operating margin comparison adds texture. AWS Q2 operating margin was 39.4%, up from the mid-30s a year ago. Google Cloud Q2 operating margin was 35.6% — its first time above 35%. Azure blends into the broader Intelligent Cloud segment, which ran at about 43% operating margin. All three hyperscalers are now generating meaningful profit from AI cloud services, not just revenue. The loss-leader era for cloud infrastructure is functionally over.

Backlog tells the next chapter. AWS backlog reached $496 billion, up from $263 billion a year ago — nearly doubling. Google Cloud’s comparable RPO figure was $514 billion, a $50 billion sequential add. Microsoft’s commercial RPO was $678 billion. All three have visibility profiles measured in years rather than quarters. The AI infrastructure demand that is driving current growth is not spot demand — it is contracted, multi-year, and growing.

Add up the annualized capital expenditure guidance from the four companies that reported last week and the number is staggering. Alphabet: $195-205 billion for FY2026. Microsoft: approximately $175 billion for FY27. Amazon: approximately $220 billion for FY2026 (raised from $200 billion). Meta: $130-145 billion for FY2026. The combined midpoint is roughly $720-730 billion — a figure that would rank as the 18th-largest economy in the world if it were GDP.

This scale of spending has no historical precedent in peacetime corporate capital allocation. The closest comparison is the late-1990s telecom buildout, which ended badly for most participants. The difference this time is that the hyperscalers are building on their own balance sheets rather than with leveraged debt, cloud demand is already monetizing at scale rather than speculative, and all three major cloud businesses are operating profitably. The risk profile is not 1999 — but the scale of commitment means any demand slowdown will be felt deeply and quickly across the entire tech supply chain.

For enterprise technology buyers, the capex commitment signals something important: cloud capacity is coming at massive scale over the next 24 months. Data center construction that is funded today begins delivering capacity in 2027 and 2028. Enterprise customers negotiating multi-year cloud commitments in 2026 are doing so into a supply environment that will be materially looser by the time their contracts begin running at full scale. That negotiating dynamic will reshape cloud pricing in ways that are not yet visible in the reported numbers.

Andy Jassy disclosed on the Q2 call that AWS backlog — contracted work that has not yet been recognized as revenue — reached $496 billion during the quarter, up from $263 billion a year ago. An 88% year-over-year increase in backlog is a signal that enterprise customers are not window-shopping AI cloud capacity; they are signing long-term commitments. At Q2’s $42.2 billion quarterly revenue pace, $496 billion represents nearly three years of forward revenue coverage.

The backlog growth also validates the capex investment case in a way that no amount of quarterly revenue growth can. Quarterly revenue can be pulled forward from future commitments, inflated by one-time deals, or seasonal. A backlog that nearly doubled year over year is structural contracted demand — customers have signed legal commitments to spend, typically with multi-year minimum consumption thresholds, and AWS will recognize that revenue as they consume the capacity. The $220 billion in FY2026 capex that Amazon is building toward has $496 billion in contracted demand behind it.

The comparable figures — Google Cloud’s $514 billion RPO and Microsoft’s $678 billion commercial RPO — form a collective picture of the enterprise AI demand landscape. The three hyperscalers together have contracted forward revenue of roughly $1.7 trillion in remaining performance obligations. Even if growth rates normalize toward 25-30% over the next two years, the runway is visible and the underwriting is done.

J.P. Morgan Research published analysis earlier this year calling the Software Selloff driven by “broken logic” — two AI fears that cannot simultaneously be true. Fear one: AI will eat SaaS revenue (selling Salesforce, Adobe). Fear two: AI capex generates no ROI (selling Microsoft, Alphabet). If AI is powerful enough to disrupt SaaS, it is by definition generating returns. If capex generates no returns, AI is not powerful enough to disrupt SaaS.

The Q2 earnings cycle resolved the contradiction in favor of the capex-is-working thesis. Google Cloud grew 82%. AWS grew 37% at its fastest pace in 18 quarters. Azure grew 43%. All three hyperscalers expanded operating margins while investing at historic capex levels. The demand is contracted, not speculative. The unit economics are improving, not deteriorating. The Evercore analysts who had worried about Google Cloud raising concerns of share shifts walked back that concern as AWS confirmed co-acceleration. The tape says AI capex is ROI-positive at the hyperscaler level — which means Fear Two is wrong, which should stabilize the SaaS repricing overshoot as well.

Microsoft stock jumped 7% after its Q4 earnings, back above $400. Amazon jumped 9.15% after its Q2 print. Alphabet recovered from its post-earnings dip as the broader cloud narrative consolidated positive. The Software Selloff that dominated coverage through mid-July has lost its primary thesis. What replaces it — in the absence of a macro shock — is a more differentiated story where per-seat SaaS with weak AI value-add gets repriced, and cloud infrastructure with AI demand confirmation gets re-rated higher.