Microsoft, Google, and Amazon Earnings Side by Side: The AI Infrastructure Race Reaches a Watershed
The latest earnings reports from the three cloud providers Microsoft, Google, and Amazon show cloud businesses accelerating across the board: AWS third-quarter revenue was $45.98 billion, up 39.3% year over year; Azure posted $33.165 billion for the quarter, up 48%; Google Cloud revenue was $30.71 billion, more than doubling year over year; and their backlogs reached $578 billion, $677 billion, and $561 billion, respectively. Over the same period, all three posted negative free cash flow: Amazon had a net outflow of $42.36 billion over the past 12 months, Alphabet was negative $9.07 billion, and Meta had a net outflow of $2.12 billion, while only Microsoft remained positive at $18.6 billion. Market reactions diverged; Alphabet fell 12.64% in a single day on slowing advertising growth and its swing to negative cash flow, and the article argues that the competitive bottleneck is shifting from chips to power, land, and grid-connection approvals.
Why read this
AI score 65
The article places the three cloud giants' orders, capital expenditures, and free cash flow in the same table for comparison, showing how the compute shortage is changing the dimensions of competition.
Topics
Body · AI translation
In this week's earnings season, the three cloud giants Microsoft, Google, and Amazon released their results one after another, producing a rare scene of all three appearing together: cloud providers' compute capacity is being snapped up by customers queuing up, with order books already written through 2028; yet on the same companies' financial statements hang negative free cash flow unseen for years.
Cloud businesses accelerate across the board, orders booked out three years
Start with Amazon. As the world's largest cloud provider, AWS posted third-quarter revenue of $45.98 billion, up 39.3% year over year, its fastest growth in nearly five years, with annualized revenue reaching $184 billion and growth rising for six consecutive quarters. Underpinning that figure is a backlog of $578 billion on the books, growing at a triple-digit percentage year over year and up more than $80 billion from $496 billion at the end of the second quarter. CEO Andy Jassy said plainly in an earlier interview: 2026 capacity is already sold out, most of 2027 is booked, and a substantial portion of 2028 has customer commitments.
Microsoft followed closely, with numbers that also beat expectations. In the first quarter of fiscal 2027, the company reported revenue of $92.28 billion, up 19% year over year, and net income of $36.63 billion, up about 32%. This quarter Microsoft disclosed Azure's absolute revenue separately for the first time: $33.165 billion for the quarter, up 48% year over year, with growth accelerating rather than slowing at the hundred-billion-dollar scale, while commercial remaining performance obligations (RPO) rose 73% year over year to $677 billion. Microsoft also disclosed that it added 24 data centers in the quarter, a delivery pace rarely seen in its history.
By comparison, Google Cloud's growth was the steepest. Alphabet's third-quarter Google Cloud revenue was $30.71 billion, more than doubling year over year, with backlog reaching $561 billion. Even more notable was the quality of profitability: Google Cloud's operating margin hit a record high of 36.2%, with operating profit more than triple the year-earlier period. But another remark from Sundar Pichai deserves more attention: customers now wanting to lock in 2027 compute capacity find it "extremely difficult." Google even resorted at one point to renting external compute, including paying monthly for SpaceX's compute resources, to fill its own supply gap.
Year-over-year growth in cloud revenue for the three cloud providers' latest quarters. AWS was 39.3%, Azure 48%, and Google Cloud more than doubled. Source: company financial reports
The three companies report on different bases, but they point to the same thing: the demand curve for AI compute is far steeper than the supply curve. Cloud providers used to worry that capacity would sit unused once built; now they worry that once built, they cannot keep up with orders. Backlog is not intent—it is signed contracts, locking in revenue for years ahead and putting delivery pressure on the table in advance.
Under the capex race, cash flow comes under strain for the first time
Fulfilling those orders comes at no small cost. Amazon maintained its 2026 capex guidance at about $220 billion, with property and equipment spending alone reaching $77.3 billion in the third quarter, more than 30% above market expectations; Alphabet's full-year capex guidance is $195 billion to $205 billion; Meta, a company built on advertising, is one of the most aggressive buyers in this round of AI infrastructure, raising its capex floor to $135 billion and making clear it will continue investing aggressively in 2027; Microsoft promptly launched a bond offering of at least $25 billion after its earnings, which was 4.6 times oversubscribed, marking the software giant's first return to the U.S. investment-grade bond market in nearly eight years.
Free cash flow (*Amazon on a trailing-12-month basis, others for the latest quarter) and annual capex guidance. Source: company financial reports
Spending comes first and revenue later, and pressure on the cash flow statement follows. Amazon's free cash flow was a net outflow of $42.36 billion over the past 12 months, versus a net inflow a year earlier; Alphabet's free cash flow was negative $9.07 billion, negative for a second consecutive quarter; Meta had a net outflow of $2.12 billion, its first since 2012; among the four, only Microsoft still held positive free cash flow of $18.6 billion. Three "cash cows" bleeding cash in the same quarter is a first since cloud computing was born.
Interestingly, investors did not veto negative free cash flow outright. Microsoft CFO Amy Hood said the company would continue raising funds "when the timing is right"; Meta created non-voting Class C common stock, reserving theoretical financing room of about $700 billion for future AI investment. Management has clearly anticipated that the funding needs of this infrastructure cycle will far exceed what operating cash flow can bear, and debt and equity instruments are becoming standard. The credit market is making its choice too: orders for Microsoft's AAA bonds poured in, while issuance costs for lower-rated tech debt rose markedly, and Meta's five-year credit default swap cost at one point rose to the highest level on record for that contract, with capital concentrating toward certainty.
A market verdict of fire and ice
After Meta was first to report after the close on October 21, Microsoft and Google reported on the same day on the 28th, and Amazon closed out the lineup on the 29th, once the stories of orders and cash flow were told, the market priced each one in turn—and the market's verdicts were strikingly divergent. Microsoft beat expectations across the board, rising as much as 6% after hours; though it edged down 0.16% the next day amid a broad tech pullback, it held broadly steady. Meta beat expectations on both revenue and profit, with the biggest highlight being its AI shopping agent Muse surpassing 10 million users in North America six weeks after launch; the stock rose more than 6% after hours, but the next day, dragged down by a massive legal penalty and a swing to negative cash flow, it closed down about 7%, becoming the only one of the four to be sold off despite beating expectations.
Alphabet was a different story. Its three core metrics—revenue, earnings per share and advertising revenue—all fell short of expectations, with search growth slowing from 19% in the first quarter all the way to 12%. Management explained that currency was the single biggest drag, compounded by World Cup ad spending concentrated in the second quarter, and a surge in AI-mode queries with monetization still low; combined with consecutive quarters of negative free cash flow and an $8.7 billion non-cash equity investment loss, the stock plunged 12.64% the next day, its largest single-day drop since going public, and dragged the Nasdaq down 1.38% that day. At Amazon, although AWS beat expectations again, online retail grew just 2%, fourth-quarter guidance came in at the low end of the midpoint, and free cash flow remained negative, sending the stock down more than 7% after hours.
Share price performance of the relevant companies around earnings disclosures (Microsoft, Meta and Alphabet at the next U.S. market close after their respective reports; Amazon at the level after its after-hours decline narrowed; Intel at its intraday after-hours decline; Samsung Electronics at the close on the disclosure day; SK Hynix at its performance the next day after disclosure). Data source: public market quotes
The divide between gainers and losers points to the same conclusion: the market no longer pays for "AI investment" itself, only for "AI returns." The market's scrutiny is not unreasonable: a considerable portion of this round's AI revenue comes from cross-orders between cloud vendors and AI companies, and its quality should be discounted; and if the depreciation schedules for GPUs and data centers are overstated, future income statements could come under additional pressure. High cloud growth is no longer a talisman, and investors have begun checking item by item: can orders convert into revenue, can cash burn be covered by profit, and is the core base of traditional businesses being eroded by AI. Google's experience shows precisely that when a cloud surge cannot offset worries about slowing search advertising, the market will not hesitate to reprice.
The upstream super-cycle: ice and fire beneath the expansion wave
The cloud giants' urge to expand capacity travels up the supply chain, and memory makers are the first to benefit. SK Hynix posted third-quarter revenue of 104.6 trillion won, breaking 100 trillion in a single quarter for the first time, with operating profit of 83.3 trillion won and an operating margin as high as 80%, as HBM4 began shipping in volume; the company expects the memory shortage to last until 2030, with more than half of 2027–2028 capacity already locked up by long-term contracts, and announced a shareholder return plan of about 100 trillion won.
Samsung Electronics also set a record: quarterly revenue of 195.3 trillion won and operating profit of 107.4 trillion won, making it the world's first tech company to top 100 trillion won in quarterly operating profit, with average DRAM prices up 20% quarter on quarter and HBM4 sales tripling quarter on quarter. Samsung expects tight supply to continue through 2028 and is raising 2026 capital expenditure by about 60% to 84.4 trillion won. The two companies' statements corroborate each other: pricing power in the memory segment has fully returned to the sellers.
Yet Intel, on the same supply chain, is not faring well.
Third-quarter revenue was $16.51 billion, up 21% year on year, with data center and AI revenue surging 78%, the fastest growth in 15 years, and earnings per share of $0.52 beating expectations; but with the triple pressures of the October explosion at its Israeli fab, a weak PC market and rising memory prices, fourth-quarter revenue guidance was only $15–16 billion and EPS guidance $0.25, far below market expectations, sending the stock plunging more than 20% after hours.
On the same supply chain, some enjoy fat profits while others cannot even deliver, and the divergence under a shortage economy is widening. This boom has also washed over China's supply chain: domestic DRAM maker ChangXin Technology expects revenue of about 265–270 billion yuan for the first three quarters, roughly seven times year-on-year growth, with net profit attributable to shareholders of 131–132.5 billion yuan, a sharp turnaround from a year-earlier loss.
The watershed: from whose model is stronger to who has enough power
Taken together, the signals from this week show that the industry's competitive logic is shifting. The bottleneck has moved from the chips themselves to power, land, cooling, grid-connection approvals and construction cycles—links that expand on a timescale of years and inherently cannot keep pace with model iteration. Andy Jassy said demand already locked in for 2028 is "stunning," and Sundar Pichai said every company wants to double its compute; in essence they are saying the same thing: compute will be scarce for the foreseeable future.