April 29, 2026 was the day Big Tech stopped pretending to be anything other than AI infrastructure companies.
Microsoft, Alphabet, Meta, and Amazon all dropped quarterly earnings on the same Wednesday afternoon. Every single one beat expectations. Every single one credited AI. And taken together, they spent roughly $130B on infrastructure in just three months. That's roughly the entire annual revenue NVIDIA reported for fiscal 2025 ($130.5B); in other words, NVIDIA's biggest customers spent NVIDIA's full year of revenue on chips and concrete in 90 days.
Then Sundar Pichai went on Google's earnings call and said the quiet part out loud: "Obviously, we are compute constrained in the near-term... our cloud revenue would have been higher if we were able to meet that demand." Translation: the world wants more AI than the world's most valuable companies can pour concrete to deliver.
First up, the TL;DR
Microsoft, Google, Meta, and Amazon reported Q1 2026 earnings on the same day, and the headlines all sounded suspiciously similar.
Here's what happened:
- Microsoft's AI business (the bucket of revenue Satya Nadella attributes to AI specifically) hit a $37B annual run rate, up 123% YoY. M365 Copilot now has 20M paid enterprise seats; Accenture alone signed up 740,000 of them.
- Google Cloud grew 63% (its biggest jump ever) to $20B in quarterly revenue. AI products built on Gemini grew nearly 800% YoY, and the cloud backlog (signed contracts not yet delivered) doubled in one quarter to $462B.
- Amazon's AWS grew 28%, its fastest in 15 quarters. Amazon's chip division (Trainium, Graviton, Nitro) passed a $20B annual run rate, with OpenAI committing to 2 GW of Trainium capacity and Anthropic committing to up to 5 GW.
- Meta's revenue grew 33% to $56.3B, mostly from ad-targeting AI. It also raised 2026 capex guidance to $125-145B (from $115-135B).
Why this matters: Q1 capex across the four hyperscalers totaled roughly $130B; nearly 2x what they spent in Q1 2025. Google raised 2026 capex guidance to up to $190B and said it'll "significantly increase" again in 2027. Combined with Microsoft's $627B RPO (signed commercial contracts across M365, Azure, and Dynamics 365, not yet billed) and Google's $462B cloud backlog, there's now over a trillion dollars in signed enterprise commitments the hyperscalers cannot deliver yet.
Our take: Cloud growth rates (AWS +28%, Azure +40%, Google Cloud +63%) used to tell you who was winning. Now they tell you who has the most concrete poured. The real moat is custom silicon plus locked-in workloads, and Amazon may be quietly winning a war nobody's talking about: OpenAI on Trainium, Anthropic on Trainium, Meta on Graviton. AWS spent most of its existence renting out other people's chips. Now everyone wants to rent its chips instead.
Open question: when does $500B+ in annual capex stop being a demand signal and start being an arms race? Wall Street is already split. Goldman and Moody's say capex estimates are still too low; Cresset says Amazon's free cash flow collapse (from $25.9B to $1.2B in twelve months) is the canary. The Motley Fool's read is the most useful: stop watching total capex, start watching margin cushion. Google has 32.8% net margins and in-house TPUs. Amazon has 10.8% margins and the boldest bet. If AI revenue takes longer to catch up than the bulls expect, the gap between those two numbers is what determines who's still standing in 2027.
How a single quarter rewrote the AI playbook
To understand what happened on April 29, you need to look at the four reports as one document, not four. The narrative across all of them is identical: AI is selling faster than we can build for it, so we are bringing forward years of capex into a single year, and we still can't meet demand. The differences are in how each company is making that bet.
Microsoft: the agentic computing pitch
Nadella spent most of his earnings call selling a phrase, "the agentic computing era," and gave it teeth with one hard number: a $37B annual AI revenue run rate, up 123%. To put that in scale, Microsoft's AI business alone is now roughly the size of Salesforce's entire fiscal 2026 annual revenue ($41.5B).
The Copilot story is the cleanest demonstration of where enterprise AI actually sits in 2026. Microsoft's M365 Copilot now has 20M paid enterprise seats, and the number of companies paying for over 50,000 seats has quadrupled. Bayer, Johnson & Johnson, Mercedes, and Roche each have over 90,000 seats. Accenture alone signed up for 740,000.
More importantly, people use it. Nadella claimed Copilot weekly engagement now matches Outlook ("a daily habit of intense usage"), and queries per user grew almost 20% QoQ. As of last week, Agent mode (where Copilot takes multi-step actions in your documents on its own) is the default in Word, Excel, and PowerPoint.
Two strategic moves stand out. First, Microsoft made Copilot multi-model. M365 Copilot now supports Anthropic's Claude alongside OpenAI's GPT, with auto-routing between them. Second, Microsoft is decoupling its commercial story from OpenAI in the financials: the company now reports "non-GAAP" earnings that exclude OpenAI investment results, signaling that the OpenAI exposure is becoming a sideshow to the core agentic platform story.
The number that should make Microsoft's competitors nervous, though, is $627B in commercial RPO, up 99% YoY. That's signed enterprise commitments Microsoft has not yet delivered. No company in software history has had a backlog like that.
Google: the most ridiculous quarter in cloud history
Google Cloud grew 63% to $20B. AI solutions inside that grew nearly 800% YoY. Gemini Enterprise paid monthly active users grew 40% in just one quarter. Direct API calls now hit 16B tokens per minute, up 60% from Q4.
Then Pichai said, on the call, that cloud revenue would have been higher if Google could have met demand. The cloud backlog doubled in a single quarter to $462B, the number of $100M-to-$1B deals doubled YoY, and Google now signs multiple "billion-dollar-plus" cloud deals per quarter. Customer commitments outpaced initial deal sizes by 45% in just three months.
To handle that, Google raised 2026 capex guidance to up to $190B and warned it will "significantly increase" again in 2027. The company also issued $31.1B in senior unsecured notes during the quarter; that's a fancy way of saying Google borrowed a small country's GDP to keep pouring concrete.
The interesting part is how Pichai is framing the constraint. He's spinning capacity-constrained as a positive ("our backlog demonstrates how Google Cloud is different"). It probably is. But there's a less convenient reading: Google's 63% growth is a measure of how fast it can build, not how much enterprises want from it. Real demand is much higher, and it's flowing somewhere. The most likely "somewhere" is AWS.
Amazon: the chip company hiding inside a retailer
AWS grew 28% to $37.6B, its fastest growth in 15 quarters. Operating income hit $14.2B, with a 37.7% margin. Those are great numbers, but they look slow next to Google's 63%. The headline-grabbing number is buried elsewhere in Amazon's release: Amazon's chips business is now at a $20B annual revenue run rate, growing triple-digits YoY.
That number includes Graviton (general-purpose CPUs), Trainium (AI training chips), and Nitro. And the customer list is a who's who of the AI economy:
- OpenAI committed to roughly 2 GW of Trainium capacity through AWS, beginning in 2027
- Anthropic committed to up to 5 GW of current and future Trainium
- Meta signed a deal to deploy tens of millions of AWS Graviton cores for agentic AI workloads
- Uber is putting Graviton4 to work on millions of daily ride-matching decisions, with Trainium3 training the matching models
Amazon also disclosed it landed 2.1M+ AI chips over the last twelve months (more than half of which were Trainium) and is deploying 1M+ NVIDIA GPUs starting in 2026. The Bedrock model platform processed more tokens in Q1 alone than in all prior years combined, and customer spend on it grew 170% QoQ.
There's a counter-narrative worth holding here. Amazon's TTM free cash flow collapsed from $25.9B a year ago to $1.2B today, a 95% drop, driven entirely by a $59.3B YoY increase in capital expenditures. The only thing keeping Amazon's reported earnings looking healthy is a $16.8B unrealized gain on its Anthropic stake. That is not free cash flow. That is paper. If AI demand softens before the chips earn their depreciation back, this is the kind of math that looks bad in retrospect.
Meta: the superintelligence outlier
Meta is the odd one out. It is the only one of the four whose AI investments don't (yet) directly generate cloud or chip revenue from third parties. Its 33% revenue growth came almost entirely from advertising, which AI helps target more efficiently (ad impressions +19%, average ad price +12%).
The strategic bet is bigger and weirder. Mark Zuckerberg told investors Meta released its first model from "Meta Superintelligence Labs" (his renamed AI division) and that the company is "on track to deliver personal superintelligence to billions of people." That's not enterprise infrastructure language. That's consumer AI assistant language.
To pay for it, Meta raised 2026 capex guidance to $125-145B (from $115-135B prior), citing higher component pricing and additional data center costs. Reality Labs is still hemorrhaging money (-$4B operating loss in Q1), but the broader Meta business is now a 41%-margin ad machine subsidizing a moonshot bet.
The risk is straightforward: Meta is the only hyperscaler whose Q1 numbers don't yet show a customer paying for AI directly. The other three have enterprise contracts they can point to. Meta has Zuckerberg's conviction. That has worked before (mobile, Reels) and failed before (Metaverse). The next four quarters will tell us which template this fits.
The Anthropic thread running through every report
Every single one of these earnings releases mentioned Anthropic.
- Amazon booked $16.8B in unrealized gains on its Anthropic stake; Anthropic committed to up to 5 GW of Trainium; Claude Opus 4.7 launched on Bedrock
- Microsoft added Claude as a multi-model option in M365 Copilot
- Google announced the prior week it would invest up to $40B more in Anthropic in cash and compute
- And on the same day all this hit, Bloomberg reported that Anthropic is weighing a funding round at over $900B in valuation
A startup that, a year ago, was the awkward number-two model lab is now structurally embedded in three of the four largest software companies on Earth. Whether that's a sign of a healthy multi-polar AI market or a vendor concentration risk dressed up as competition is, at this point, a Rorschach test for what you already believe about AI. Either way, Dario Amodei is having a year.
What this means for you
Three things you can probably bank on for the rest of 2026:
- Cloud price pressure is going one way: up. Google can't meet existing demand; Microsoft has $627B in unbilled commitments. Don't expect aggressive enterprise discounting. The hyperscalers don't need your business badly enough.
- Multi-model is the new default. Microsoft is auto-routing between OpenAI and Anthropic. AWS Bedrock now hosts both. If your AI strategy assumes a single model provider, it's already out of date.
- Custom silicon will quietly eat into NVIDIA's near-monopoly. OpenAI alone committing 2 GW to Trainium, plus Google's TPU growth and Microsoft's Cobalt, signals that the largest AI buyers are deliberately diversifying their chip dependence. NVIDIA will still win 2026; the question is what 2028 looks like.
Where it goes next
The bull case is simple: there's a trillion dollars in signed-but-undelivered cloud contracts. Real money is on the table for whoever can build fastest. The AI economy is no longer "promising"; it's billed.
The bear case is also simple: 2026 capex across the Big Four hyperscalers is on pace for $500B+. That's roughly five times what they spent on AI just two years ago. If usage growth doesn't continue at current pace (Bedrock tokens, Copilot queries, Gemini API minutes), this becomes the largest infrastructure overbuild in tech history.
The honest answer is we don't know yet which one wins. What we do know: every Big Tech CEO now sounds the same on earnings calls, and that same-ness is itself a signal. When everyone is making the same bet at the same scale, the differentiation moves down a layer. The interesting question for 2027 won't be "who has the most capex," it'll be "who has the most efficient chips, the deepest customer lock-in, and the best margins on inference."
If you want a one-question test for which company is winning the next phase: don't watch revenue. Watch how much each one says about its custom chips and its agent platforms next quarter.
