Market Watch · MARKET × Sector Review · SECTOR

AI Capex Validation Night: When Four Giants Simultaneously Raised Spending, Markets Punished the Increase—for the First Time

Four of the Mag 7 reported earnings on the same night—capex guidance rose across the board—but for the first time, markets treated higher investment as bad news. The AI capital cycle has shifted gears.

Approx. 2,800 words · ~8-minute read

On April 29, within 80 seconds after U.S. market close (ET), four of the Mag 7—Microsoft, Alphabet, Meta, and Amazon—released their Q1 earnings in a tightly clustered window. All four beat expectations—but after-hours reactions diverged sharply:Two sold off; two rose。

Markets no longer reward capex with a single brushstroke. Instead, they now grade based on clarity of return pathways. Summing up their 2026 capex guidance yields a range of ~$620–680B—one of the largest single-industry, single-year capital expenditure clusters in history. Yet after-hours moves signal that capital markets’ attitude toward this spending isshifting from ‘faith’ to ‘accountability’。

CHAPTER 01

Selling Off After Beating Estimates

None of the four earnings reports was weak.Microsoft revenue $82.89B, beating consensus of $81.46B; its Azure cloud business grew +40%(+39% ex-FX), above its 37–38% guidance ceiling.Alphabet revenue $109.9B, EPS $2.82 vs. $2.63 expected; Google Cloud $20.03B, YoY +63%。Meta revenue $56.31B, adjusted EPS $7.31 vs. $6.82 expected; ad pricing +12%, ad impressions +19%.Amazon revenue $181.5B vs. $177.3B expected; AWS $37.59B, growth +28%—a 15-quarter high on an ex-FX basis.

Under last year’s script, such results would have triggered broad after-hours gains.This time, the script didn’t run.

EARNINGS NIGHT · After-Hours Reaction Summary

Source: CNBC / investingLive / Yahoo Finance · After-hours data as of April 30, Beijing time

Microsoft fell ~2% after hours (CNBC closing price $424.46, briefly dropping to $413.52). Meta plunged to -6% within 30 seconds—essentially mirroring its upward revision of 2026 capex from $115–135B to $125–145B,adding $10B at both ends.Amazon swung wildly overnight—its after-hours low hit -7%but AWS acceleration plus two long-term contracts with Anthropic and OpenAI held the floor, allowing it to rebound to near +2.2%、$268 by pre-market on April 30. Alphabet gained outright—Cloud +63% was decisive, and its capex guidance remained unchanged.

CHAPTER 02

Where the Divergence Lies

Laid side-by-side, the logic behind the divergence isn’t whether they beat or missed—it’s‘What exactly does this money buy?’—markets are grading, for the first time, on clarity of return pathways.

CAPEX GUIDANCE · 2026 Full-Year Capital Expenditures

Bar height = upper end of capex range; orange segment = incremental increase announced this time.

Sum of all four ~$620–680B, one of the largest single-industry, single-year capex clusters in history.

Alphabet was the only one rewarded. Cloud +63% YoY directly linked ‘capex doubling’ to tangible enterprise revenue. Its $175–185B guidance is nearly double last year’s $91.4B—a typical red flag—but GCP’s $20.03B quarterly revenue and accelerating growth mean cash inflows and capex outlays are now synchronized—at least on the cloud front.

Amazon’s narrative is the most complex. AWS growth accelerated to +28%, a 15-quarter high. That same night, it disclosed two long-term contracts:Anthropic committed over $100B to AWS over 10 years, locking in up to 5 GW of Trainium compute power(Project Rainier already deploys more than 500,000 Trainium2 chips); OpenAI expanded its AWS contract from $38B to $100B over eight years, consuming 2 GW of Trainium capacity—spanning both Trainium3 and Trainium4 generations. This marks the first time Amazon’s in-house chips secured two flagship-scale contracts of equal magnitude. Its $200B annual capex plan compressed free cash flow to $1.2B(down from $25.9B twelve months earlier)—a lingering concern—but markets ultimately voted confidence in the two contracts.

Microsoft is the textbook case of ‘beating—and still falling short.’ Revenue, EPS, and Azure all beat expectations; its AI business annualized run rate climbed to ~$37B; Microsoft 365 Copilot paid seats jumped from 15 million last quarter to 20 million. Yet quarterly capex (including finance leases)$31.9B, $3.4B below the $35.3B consensus—a number markets disliked.The reason is straightforward: Microsoft’s prior bottleneck was ‘insufficient capacity—Azure couldn’t absorb demand.’ Falling short on capex signals that supply constraints remain unresolved—even as demand roars. In that context, spending less isn’t prudence—it’s lost sales.

Meta took the heaviest punishment. Revenue +33%, adjusted EPS well above expectations, ad pricing +12%, ad impressions +19%—this set of metrics would have delivered a +5% after-hours pop last year. But its $10B capex hike triggered a -6% plunge within 30 seconds. The difference lies in the return pathway behind that spend: ‘longer-term AI assistant narrative,’ not Alphabet’s direct, immediate ‘cloud customers paying today.’

CHAPTER 03

Sector View—The Hidden Split Within the Mag 7

Zooming out from individual companies to the sector reveals a subtler, more durable fault line. For the past two years, the Mag 7’s sector narrative boiled down to one sentence—‘Who spends more on capex is the AI winner.’NVIDIA supplied the chips; the others absorbed them. Markets priced that story into PE multiples above 35x. That sentence no longer holds after tonight.

In the same evening, Meta—raising capex—sold off; Microsoft—raising capex—was modestly punished; Alphabet—raising capex but with strong Cloud performance—was rewarded; Amazon—raising capex but backed by long-term contracts—received cautious validation.The AI sector is now stratifying by ‘clarity of return pathways.’

SECTOR STRATIFICATION · Three-Tier Sector Split

Even amid rising capex, markets assign three distinct valuations.

This stratification will deepen over coming quarters. Any company with large capex but ARR growth lagging behind will face valuation pressure; any with large capex and accelerating cloud growth can sustain its multiple.

Beyond intra-sector splits, another outward ripple emerges: the four companies’ combined ~$650B capex will channel significant portions into in-house chips and power infrastructure—AWS Trainium has already locked in Anthropic and OpenAI, generating over $20B in annualized revenue at triple-digit YoY growth. NVIDIA’s singular-beneficiary status is being sliced up. This thread aligns with the logic laid out in our April 25 piece ‘CPU’s Return’—The heterogeneous arms race has already pivoted at the hardware-allocation level.。

CHAPTER 04

Powell’s Farewell and the Fed’s Fracture

Another key thread of the earnings night was the Fed’s added weight on capital-cycle judgment. At 2:00 p.m. ET on April 29, the FOMC held rates steady at 3.5-3.75% —a move fully priced in by markets. What moved the needle was the vote split.

FOMC, April 29 · Largest dissent since October 1992

Powell steps down as chair on May 15 but remains on the Board as a governor · Kevin Warsh has cleared the Senate Banking Committee’s party-line vote (13-11), full Senate vote expected mid-May

Powell said this was his final press conference as chair ("This is my last press conference as chair"). Markets broadly expect Warsh to ease policy sooner than Powell, under political pressure.

Linking this thread to the earnings night: the Mag 7’s combined ~$650B capex plans rely heavily on financing and operating cash flow (Amazon’s FCF collapse to $1.2B is an early warning). If the Fed enters a phase of ‘cutting—but deeply divided,’ lower funding costs would ease capex-cycle financing pressure—one of the few positive variables in Capex Phase Two.But that hinges on Warsh narrowing the current 8–4 split to 9–3 or 10–2; otherwise, Fed messaging and actual path will oscillate.

CHAPTER 05

Capital-Cycle Shift: From ‘Building Capacity’ to ‘Capacity Must Deliver ROI’

Placing tonight on a longer timeline marks a clear gear shift in the AI capital cycle.

Phase One (2023–2025 core theme):Investors paid for capex itself. Logic: ‘Compute is scarce—the earlier you spend, the earlier you secure position.’ So whoever spent hardest was deemed the winner. NVIDIA was the biggest beneficiary; hyperscalers were secondary beneficiaries; PE multiples soared to historic highs.

Phase Two (starting in 2026 — tonight is the inflection point): Investors now demand a clear ROI path tied to capex. The same capital expenditure earns market credibility if it secures long-term contracts — e.g., Anthropic for 5 GW and OpenAI for 2 GW — though skepticism remains; it earns a premium if it directly maps to near-term cloud revenue acceleration — e.g., GCP +63%, AWS +28%; but it incurs a penalty if it only supports vague ‘future AI assistant’ narratives.

The least-penalized positions are: ① capex directly locked into long-term contracts (Amazon’s model); ② capex that simultaneously accelerates cloud revenue (Alphabet’s model). The most-penalized position is large capex with no direct linkage to current revenue — i.e., distant, story-driven bets (Meta’s current case).

This shift will manifest more concretely in second-order financial metrics. The first to surface is free cash flow — Amazon’s 12-month rolling FCF dropped from $25.9B to $1.2B, the sharpest early warning signal of this phase. Next comes the depreciation curve: $650B in capex, amortized over a 5–6 year depreciation cycle, will systematically compress hyperscaler reported margins over the next 2–3 years. At that point, markets will scrutinize each company’s earnings quality far more rigorously.

This logic spills over into China along two distinct channels. One is valuation — multiples pressure on U.S. hyperscalers will transmit repeatedly via ADRs, Hong Kong tech stocks, and A-share AI stocks, especially those propped up by ‘China’s Magnificent Seven’ narratives. The other is supply-chain timing — hyperscaler capex won’t contract immediately in response to tonight’s reaction (contracts are already signed), but downstream order fulfillment will grow more selective. Compounded with the x86 server CPU shortage — which began circulating on April 25 —capacity holders locked into contracts will capture upside, while generic ‘AI-themed’ labels will become increasingly difficult to price.。

As noted earlier, AI companies are shifting from the model layer down to the application layer."Claude Design: AI Companies Are No Longer Building Tools — They’re Eating Applications". Tonight’s 80-second earnings deluge confirms that capital markets have adjusted their pricing framework accordingly —not ‘all AI gets a premium,’ but ‘all AI must first show ROI.’。

Disclaimer

This article is based on publicly available information; all data sources are cited in-text and in the source list below. It does not constitute investment advice nor reflect the position of any institution. Readers assume all risks associated with any actions taken based on this content.

Sources

CNBC · investingLive · Microsoft Investor Relations · Yahoo Finance · Seeking Alpha · DataCenterDynamics · OpenAI Blog · GeekWire · Motley Fool · Tech Insider · Fox Business · Fortune · StockTitan

Data as of April 30, 2026, Beijing time. GDP and Core PCE data from the FOMC meeting — not yet released as of publication — are excluded from analysis.