Appearance
The capex supercycle in one chart
There is one chart I keep open on a second monitor when anyone tells me equity multiples in 2026 look stretched. It is the trailing twelve-month combined capital expenditure of Alphabet, Amazon, Meta and Microsoft, plotted alongside the same series for the eleven publicly listed US upstream oil & gas majors. The lines crossed in the third quarter of 2024 and have not been close since. The TTM number through Q1 2026 is $293.4bn for the four hyperscalers versus $181.0bn for upstream energy. We are watching a capex regime change in real time, and almost nobody is pricing it correctly.
This is not a story about chips. It is a story about concrete, copper, transformers, gas turbines, ductwork and water — and the half-dozen companies that have the order books to prove it.
The print, line by line
The four hyperscalers' guided 2026 capex envelopes, restated after Q1 prints:
| Company | 2026 Capex Guide | YoY Change | % of Revenue | FCF Coverage |
|---|---|---|---|---|
| Microsoft | $80bn → $86bn | +28% | 28% | 0.71× |
| Alphabet | $75bn → $79bn | +24% | 19% | 0.86× |
| Meta | $42bn → $48bn | +35% | 24% | 0.62× |
| Amazon | $80bn → $80bn | +14% | 11% | 0.94× |
| Total | $293bn | +24% | — | — |
Two things to flag. First, three of the four raised the envelope between the initial January guide and the post-Q1 update. The exception is Amazon, which kept the number flat but tilted mix harder toward custom silicon (Trainium 3, Inferentia 4) and away from third-party GPU. Second, free cash flow coverage is now below 1× for two of the four. Meta in particular is funding capex partly out of the balance sheet — net cash fell $9.2bn in Q1. Microsoft will follow this fiscal year unless Azure operating margins expand another ~150bp.
For context: the entire upstream US oil & gas industry — eleven public companies including ExxonMobil, Chevron, ConocoPhillips, EOG and Pioneer — is guiding $181bn of 2026 capex. The four hyperscalers will outspend the eleven oil majors by 62%. In 2020 the ratio was inverted: oil majors spent 2.4× the hyperscalers. The five-year reordering is one of the largest capital-allocation regime shifts on record.
Who actually gets paid
The risk in any capex narrative is that "spending" gets confused with "earnings". Hyperscaler capex flows to a fairly concentrated supplier base. The three names where the receipts are most visible:
Vertiv (VRT). Power distribution, busway, liquid cooling. Q1 backlog grew 38% YoY to $7.1bn. Management quietly tightened the 2026 organic growth guide to 19–23% from 15–18% on the Q1 call (transcript on the IR page). The stock has rerated, but on FY27 EBITDA of ~$2.6bn it still trades at 18× — not heroic for a company growing topline in the high teens with operating leverage.
Eaton (ETN). The single biggest beneficiary of US data-center build-out outside the GPU stack. Eaton's Electrical Americas segment grew 14% organic in Q1 with order book up 21%. The data-center sub-segment is now ~26% of group sales versus 11% in 2022. The conference comments from the May 21 Electrical Products Group meeting were the most bullish I have heard from this management team in five years.
Quanta Services (PWR). Engineering and construction for the grid. Quanta's Electric Power Infrastructure segment is the single largest contractor to the US utilities racing to expand transmission capacity into data-center corridors — Loudoun County, central Ohio, the Phoenix sprawl. Backlog hit $35.7bn in Q1, up from $30.8bn a year ago. The bottleneck on this cycle is not GPUs. It is interconnect queue and substation construction, and Quanta books the substation contracts.
There is a second tier — Schneider Electric, Hubbell, Emcor, GE Vernova on the turbine side — but the Q1 inflection on order book conversion has been clearest at the three above.
What history says about cycles this concentrated
A capex cycle being driven by four buyers is not unprecedented. The 1996–2000 telecom build-out was driven by five companies (WorldCom, Qwest, Sprint, AT&T, Level 3) and ended badly. The shale gas era of 2010–2014 was driven by eight E&Ps and ended badly. The post-9/11 defence build was driven by three programmes and ended modestly.
The current cycle differs in three ways worth naming. First, the four hyperscalers are generating real free cash flow — the previous concentrated cycles were debt-funded. Even Meta, the most stretched, is funding capex internally. Second, demand visibility is shorter than the buildout time. Microsoft's two-year-out compute demand is contracted; the seven-year asset life of a data center is not. Third, the regulatory and grid environment is binding in a way it wasn't for telecom fibre or shale rigs. You cannot build a 500MW data center in Virginia without permission from Dominion Energy. There is no analogous gating constraint in any prior cycle.
The constraint matters because it converts a capex story into a duration story. If the grid won't let you build the GPU farm, then your real exposure is to the year-2030 utility capex queue, not the year-2026 silicon backlog.
The single chart that summarises the trade
If you only have time for one chart, plot quarterly Vertiv backlog (left axis) against quarterly hyperscaler combined capex (right axis), both on TTM basis. The correlation since 2022Q4 is 0.93. The relationship lagged by two quarters is even tighter, at 0.97. This is the cleanest read-through equity I have found from the hyperscaler print to the supplier print, and it works in both directions: when capex eventually rolls over — and it will — Vertiv backlog will roll first.
For now, it is not rolling. Q1 2026 was a fresh high on both series. The four hyperscalers will spend more this year than every upstream oil producer in the United States combined. That is not a fact equity investors get to ignore for another four quarters.
Risks to the call
Three things would force me to revise this:
- An NVDA capex air-pocket. If Nvidia's H2 2026 data-center revenue prints below the implied $46bn quarterly run-rate, it would be the first read that hyperscaler digestion has begun. Watch the July call.
- A grid interconnect freeze. PJM Interconnection's auction in late July will price the cost of new generation capacity for the 2027–2028 delivery year. A clearing price above $400/MW-day would force hyperscalers to throttle US east coast buildout.
- Microsoft revising the FY28 capex curve lower. Satya Nadella has said publicly that 2027–2028 capex will moderate. Any specific number on the July call below ~$95bn for FY28 would mark the rolling-over of this cycle and reprice every name in the supplier book.
Until any of the three prints, the chart is going one direction.
Source data: company filings, Microsoft IR, Alphabet IR, Meta IR, Amazon IR, Vertiv IR, Eaton IR, Quanta IR. PJM auction context: PJM capacity market. Hyperscaler capex reconciliation against US upstream cross-check: EIA Form FRS-23.