Analysis
AI capex arms race is the term for the scale of infrastructure spending that Big Tech is committing to in the race to dominate AI compute — and the numbers coming out of the April 29 earnings calls just blew past every prior benchmark. After the close on April 29, MSFT, META, AMZN, and GOOG all reported Q1 2026, all beat across the board, and then dropped a collective 2026 capex guidance that topped out at $725 billion — double the 2025 run rate — single-handedly blowing past the market's prior "all six hyperscalers combined" estimate of $700B.
The market cracked in half. Same prints, two completely different reads on what they mean.
All figures from the April 29, 2026 earnings releases (as of 2026-04-29):
| Company | Q1 GAAP EPS | Beat | Revenue & YoY | Core Cloud | After-Hours |
|---|---|---|---|---|---|
| GOOG | $5.11 | +$2.44 | $109.9B, +22% | Google Cloud $20B, +63% | +7%+ |
| META | $10.44 | +$3.78 | $56.31B, +33% | Ad revenue +34% | −8%, $50B+ mkt cap wiped |
| MSFT | $4.27 | +$0.22 | $82.9B, +18.3% | Azure +40%, AI annualized $37B+ | Flat |
| AMZN | $2.78 | +$1.13 | $181.52B, +16.6% | AWS $37.6B, +28% (15-qtr high) | +3%+ |
The prints were clean — genuinely clean, not just by lowered expectations. GOOG's Cloud grew 63% and flipped from also-ran to the room's standout, with backlog nearly doubling to $462B in a single quarter. META's EPS beat margin hit a company record. MSFT's remaining performance obligations (RPO) surged to $627B, up 2× YoY. AWS re-accelerated — fifteen straight quarters of deceleration reversed — with Anthropic's deal still sitting as dry powder.
And then the tape did something counterintuitive. The best print didn't get the biggest rally. The company with the record EPS beat — META — got smashed 8% after hours, shedding over $50B in market cap in a single session.
The pivot happened in real time: the market stopped asking does AI have demand and started asking how much are you burning to serve that demand, and is it ever coming back?
The raw capex table first. All figures from April 29, 2026 earnings calls (as of 2026-04-29):
| Company | 2024 Capex | 2025 Capex | 2026 Guidance | 2-Year Growth |
|---|---|---|---|---|
| AMZN | $83B | $132B | $200B | +141% |
| MSFT | $44.5B | $64.6B | $190B | +327% |
| GOOG | $52.5B | $91.4B | $180–190B | +252% |
| META | $37.3B | $72.2B | $125–145B | +289% |
Midpoint of the ranges: combined 2026 guidance sits at $705B. Upper bound: $725B. The consensus call heading into earnings was that all six U.S. hyperscalers combined would hit $700B this year. Four companies just blew that number out on their own.
First, the order books are real — not forward projections, not slides. MSFT's RPO is $627B. GOOG's cloud backlog doubled in a single quarter to $462B. AWS has $364B in committed, unexecuted contracts plus a nine-figure Anthropic partnership. Customers have wired money and are waiting in queue for compute capacity. The hyperscalers are spending to fulfill contracts they've already signed, not to speculate on future demand.
Second, the demand signal is bleeding through the entire supply chain. Seagate's print confirmed it: global data storage capacity is running above-full, with supply/demand imbalance well beyond consensus models. This isn't analyst narrative — it's corporate purchasing activity showing up in supplier balance sheets.
Third, there's no evidence of demand saturation at the enterprise level. GOOG went on record at the call: 2027 capex will grow "significantly" from 2026 levels. MSFT guided tight supply persisting at minimum through end of 2026. The killer enterprise AI application hasn't even shipped at scale yet. Calling the peak now means you're reading the first chapter and declaring the book over.
First, capex growth is completely decoupled from revenue growth — and that gap is widening. MSFT revenue grew 18% YoY while guiding capex up nearly 200%. META's revenue grew 33% while capex guidance jumped ~90%. No industry sustains this ratio indefinitely. Every dollar going in today eventually compresses future margins — and the compounding effect of massive depreciation charges hasn't started flowing through yet.
Second, the market already voted. META's print was arguably the cleanest of the group — record EPS beat, 33% revenue growth, 34% ad revenue growth. The stock dropped 8% the second management raised the full-year capex ceiling from $135B to $145B. That's the market encoding a simple message: earning more money and then routing all of it through a data center doesn't compound equity value for shareholders.
Third, free cash flow is getting hollowed out — and that matters more than EPS right now. AMZN's trailing 12-month free cash flow dropped 95% YoY to just $1.2B. The bull run in U.S. tech over the last three years ran on two engines: earnings growth and buybacks. Buybacks are funded by free cash flow. When the cash flow goes, the buyback shrinks, and the EPS support leg of the bull case disappears with it.
The real disagreement here isn't about AI's future. It's about whether commercial AI revenue scales fast enough to justify the burn rate on this particular timeline.
Bulls are pricing the 2026–2027 window: order book locked, compute constrained, pricing power intact, monetization starts compressing the capex-to-revenue gap within the next 4–6 quarters.
Bears are pricing the 2028+ window: no industry compound-grows capex 3× while revenue grows 20% indefinitely. The terminal outcome is overcapacity, depreciation headwinds, and ROI compression. The bigger the build-out, the harder the eventual correction.
The mistake retail investors make is trying to front-run the terminal outcome with short-dated positioning. You don't need to have a view on whether AI is the internet or the tulip mania. You need to have a view on what's confirmable in the next 12 months — and right now, the order books are real.
Valuation data from original analysis, consensus estimates (Wind, as of 2026-04-29):
| Company | Current Price | 2027F EPS | Fwd P/E | PEG | Rating |
|---|---|---|---|---|---|
| META | ~$600 | ~$36 | ~17× | ≈1 | Deep value, maximum margin of safety |
| MSFT | ~$480 | ~$21.8 | ~22× | ≈1 | Fairly priced, no obvious bubble |
| GOOG | ~$350 | ~$14 | ~28× | ≈1.5 | Stretched, poor risk/reward |
| AMZN | ~$270 | ~$9.6 | ~28× | >1.6 | Most expensive on adjusted basis |
My read is direct:
META is the only genuine value in the group. 17× forward P/E, PEG near 1 — you're buying growth-stock earnings velocity at value-stock multiples. Even if the AI cycle softens, the downside is cushioned. If ROI lands as guided, the upside leverage is asymmetric. The post-earnings 8% drop was a sentiment overshoot, not a fundamental breakdown. That's where you layer in.
MSFT is fair, for what it is. 22× P/E, PEG ≈ 1, Azure steady, the best order-book visibility of the group. Nothing wrong with the fundamentals — but $190B in capex still carries execution risk. Buy on pullbacks, not in stride.
GOOG and AMZN are sentiment-premium priced. 28× forward P/E on growth that doesn't fully justify that multiple. PEG above 1.5 on both. If capex starts biting margins, the valuation contraction will be sharp. Neither name makes sense until you see 20%+ drawdowns from current levels.
I don't think we're in innings 7–8 of a late-cycle trade. That's too pessimistic. I also don't think this is free money with no execution risk. That's too optimistic. The honest read is that we're in the middle innings — maybe 5–6 of a long game — and the first half was about whether AI was real, the second half is about who actually monetizes it.
The internet analogy isn't wrong. In 2000, you couldn't call the internet over just because too much fiber had been laid. The application layer came later. AI's application layer hasn't been built yet at the enterprise scale that justifies this infrastructure. That doesn't make the infrastructure wrong — it makes the timing of the ROI uncertain.
Three risks that aren't priced cleanly:
Core principle: don't chase highs, don't go to zero, cut the attack exposure, get into defensive quality, and earn certainty premium.
Single-name positioning:
Supply chain play: Regardless of which hyperscaler wins, the shovel-sellers win. Data storage, advanced chips, optical interconnects, data center thermal management — hard infrastructure with confirmed demand pull, highest certainty profile in the complex.
Position sizing: Cut tech allocation from full-offensive to 50–60% of portfolio. Keep 40%+ in cash. Buy NDX put options with 1–2% of total portfolio at current low-vol levels — asymmetric tail hedge that costs almost nothing while the index is at all-time highs.
It didn't suddenly double — it's being pulled higher by locked-order commitments. MSFT has $627B in remaining performance obligations. GOOG's cloud backlog doubled QoQ to $462B. AWS carries $364B in unexecuted contracts plus a nine-figure Anthropic deal. The spending is contract fulfillment, not speculative build.
Because META raised the high end of full-year capex guidance from $135B to $145B, and the market re-priced the cash flow math immediately. The signal isn't about earnings quality — it's about capital allocation. When every dollar of free cash flow routes back into a data center, buybacks shrink and shareholder returns compress. That's the regime change the market is trading.
Trailing 12-month Amazon FCF fell 95% YoY to $1.2B, driven by AWS expansion capex plus amortization on the Anthropic investment. The structural issue is that capex outlays are hitting P&L faster than cloud revenue growth can offset them. This is the bear case in raw numbers: free cash flow funds buybacks, buybacks support EPS, EPS supports multiples. Break that chain at the first link and the whole stack re-rates.
This is not investment advice. VM Genius uses Smart Money Concept frameworks plus AI agents to read institutional accumulation and distribution in real time. Follow the blog and the Discord for updates.