4AUG

Hyperscalers post record $110.75bn Q1 capex

4 min read

10:53UTC

Microsoft, Alphabet, Amazon and Meta together spent $110.75bn on capital projects in the first 90 days of 2026, with two of the four raising full-year guidance even as OpenAI trimmed its own compute commitment.

← Back to Ofgem prices the grid queue by the megawatt Jump to analysis ↓

IndustryDeveloping

Key takeaway

Alphabet and Meta raised guidance after OpenAI's retreat; Q2 earnings will test whether the divergence holds.

Amazon, Alphabet, Microsoft and Meta posted combined Q1 2026 capital expenditure of $110.75bn when results landed on Wednesday 29 April ¹. Amazon led at $44.2bn, followed by Alphabet at $35.7bn (up 107% year-on-year) and Microsoft at $30.88bn. Meta did not break out a separate Q1 number but raised its full-year range to $125-145bn, citing 'higher component costs and expanded data-centre capacity'. Alphabet raised full-year guidance to $180-190bn; on Amazon's earnings call Andy Jassy committed the company to roughly $200bn for the year.

The four are the cloud platforms on which OpenAI, Anthropic, xAI and the bulk of the enterprise AI market run their training and inference workloads. Q1 capex of this scale is the cleanest signal that boards approved their 2026 build plans before OpenAI's revision became visible, and chose not to walk them back when it did. Sundar Pichai told Alphabet's earnings call the company is 'compute constrained in the near term' with cloud backlog above $460bn, a figure that gives the raised guidance an underwriter rather than a speculator: the backlog is contracted demand, not modelled forecast.

The aggregate now firms above the $725bn annual figure flagged at the end of April . What it cannot yet answer is whether the spend reflects confidence in the demand profile or fear of being short on capacity when transformer slots close in 2027. Meta's component-cost reference points to the second reading; Pichai's backlog number points to the first. The two are not mutually exclusive, but the test arrives in the late-July Q2 earnings cycle, when boards will reset guidance against a quarter that has fully digested the OpenAI retreat.

The Microsoft figure carries a methodological caveat that matters for sectoral totals: Microsoft does not split data-centre capex from broader Azure spend in its 10-Q, so the $30.88bn includes an undisclosed non-DC portion. The combined $110.75bn is therefore an upper bound on the data-centre slice. Even on the more conservative read, the quarter is the largest concentrated build commitment the sector has posted, and it travels into Q2 against a softer OpenAI demand signal.

Deep Analysis

In plain English

Amazon, Google's parent company Alphabet, Microsoft, and Meta all reported how much they spent on data centres and cloud infrastructure in the first three months of 2026. Together they spent $110.75 billion, roughly the GDP of a mid-sized European country, in a single quarter. Two of them, Alphabet and Meta, also raised their spending targets for the whole of 2026. This is happening at the same moment that OpenAI, which relies on their infrastructure, just cut its own spending plans significantly. The companies building the data centres and the company that was supposed to fill them are now moving in opposite directions.

Deep Analysis

Root Causes

Alphabet's 107% year-on-year capex increase reflects a structural inflection that began in Q3 2024: the decision by Fortune 500 boards to commit to multi-year cloud AI contracts drove Google Cloud's backlog from roughly $250bn to $460bn in under eighteen months. That contracted demand obligates Alphabet to build the capacity; the Q1 capex is the execution of commitments already on the books, not speculative positioning.

Meta's component-cost reference exposes a second structural driver: GPU server lead times running 12-18 months mean boards that want 2027 capacity must commit 2026 capex now, regardless of what OpenAI's revised trajectory signals about 2030 demand. The two dynamics, contracted cloud backlog and hardware lead-time arbitrage, explain why no hyperscaler revised guidance downward despite OpenAI's revision.

What could happen next?

Risk
If Q2 guidance is held or raised again, capital markets will price hyperscaler equity against a demand curve that may be thinner than the build programme assumes.
Short term · 0.7
Consequence
Amazon's $200bn FY commitment, the largest single-company annual capital programme on public record, sets the ceiling against which all grid and equipment suppliers will price their 2027-2028 order books.
Medium term · 0.85
Opportunity
Alphabet's $460bn contracted cloud backlog provides a demand underwriter for its capex that other hyperscalers cannot match on a disclosed basis, making it the most defensible large-capex story into 2027.
Medium term · 0.75

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: The 2000-2001 telecom capex cycle saw WorldCom, Global Crossing and MCI commit $80bn+ in a single year to fibre build-out, justified by traffic doubling projections that were real but whose timing was wrong by 24-36 months.

The hyperscaler Q1 2026 aggregate is structurally parallel: demand is real, but the build cycle is committing capital against a 2028-2030 demand curve in a market where two-year shifts in inference architecture or model efficiency can change the required compute per revenue dollar.

Counter-parallel: Unlike the telcos, hyperscalers have $460bn of contracted cloud backlog (Alphabet alone) and diversified revenue not exposed to a single bet. The 2000 fibre glut had no contracted demand beneath it; the 2026 hyperscaler build has enterprise AI commitments that are multi-year, not speculative.

The combined $110.75bn Q1 print puts the four on a trajectory of roughly $440bn for 2026 if Q1 is the base rate, already above the $725bn annual aggregate flagged in late April on a pro-rated basis. Amazon's commitment of roughly $200bn for the full year is the single largest annual capital programme any company has publicly disclosed, exceeding Saudi Aramco's 2022 programme of $175bn.

For equipment vendors, the aggregate firms their order books through 2028 at minimum. Wood Mackenzie estimates a 30% power transformer deficit in 2026; at Q1 capex rates, every percentage point of that deficit costs the sector roughly $13bn in deferred commissioning value. The Microsoft caveat on methodology matters for analysts: the $30.88bn includes non-DC Azure spend, so the true DC-specific aggregate is somewhat below $110.75bn, though still a record.

Consensus view: Synergy Research Group's John Dinsdale tracks quarterly hyperscaler capex as the most reliable leading indicator of colocation and power demand; the $110.75bn Q1 combined print, up from roughly $62bn in Q1 2025, confirms the sector passed an inflection point in late 2025 that board cycles are now executing against regardless of tenant-level signals. Alphabet's $460bn cloud backlog is contracted demand, not projected demand.

Counter-view: Asianometry (Jon Y) notes that hyperscaler capex at this scale is disproportionately concentrated in GPU server procurement and campus land banking, both of which have 18-to-36-month commissioning tails. The Q1 spend figure tells you what boards approved in 2024, not what will be energised in 2026.

The Malaysian and Johor moratorium shows the same capex wave is meeting resistance in Southeast Asia, where supply-chain concentration in TSMC and Samsung fabs creates a second bottleneck behind the transformer queue.

Key tension: Whether Alphabet and Meta's raised full-year guidance signals that hyperscaler demand has genuinely absorbed the OpenAI retreat, or whether the guidance was set before Q2 models could price the revision and will be walked back in the late-July earnings cycle.

Sources:The Next Web

Mentions:Microsoft →Alphabet →Amazon →Meta →Andy Jassy →Sundar Pichai →Google →Synergy Research Group →Johor →Anthropic →Samsung Electronics →TSMC →OpenAI →Azure →

First Reported In

Update #3 · OpenAI cuts $800bn; rivals double down

The Next Web· 16 May 2026

Read original →

Causes and effects

Caused by

Hyperscalers raise 2026 capex to $725 billion

The prior $725bn combined hyperscaler capex figure firmed into a higher number as Meta and Alphabet raised full-year guidance in Q1 earnings.

Occurred 30 Apr 2026

Read story →

OpenAI cuts compute target by $800bn

OpenAI's retreat from owned build positions redirected demand toward leased colocation, reinforcing hyperscaler capex at the same moment OpenAI's own commitment compressed.

Occurred 28 Apr 2026

Read story →

This Event

Hyperscalers post record $110.75bn Q1 capex

The four largest cloud operators are now visibly building for a demand curve their largest individual customer has just shortened, exposing the gap between hyperscaler quarterly capex cycles and tenant compute commitments.

Different Perspectives

Government of Odisha

Odisha signed a $1.5bn agreement with HCLTech and Sarvam AI on 24 July for a sovereign AI data centre in Bhubaneswar, targeting 2028 operations. While regulators elsewhere price developers out of the queue, Odisha is bidding state financial assistance to bring capacity in, betting the AI infrastructure boom is worth courting rather than rationing.

Switch Datacenters and Uithoorn

The developer, the municipality, Liander and dozens of greenhouse growers signed a waste-heat agreement letting a 60 MVA facility proceed under a provincial rule that bans sites above 5 MVA. The public-interest exception argues the site relieves rather than adds to grid congestion, working around a standing ban from inside rather than waiting for it to change.

Iskandar Puteri residents

Residents protested in February over falling water pressure they attribute to ZDATA's 300 MW construction; ZDATA disputes causation and has not yet resolved the dispute with its wastewater-only pivot. The complaint sits unadjudicated while the operator advertises compliance to state authorities the residents cannot address directly.

Microsoft

DC Byte reports Microsoft looking to the Nordics for new capacity while Dublin's regional share falls to 15 per cent, though this is a market-intelligence inference rather than a company statement. Microsoft keeps its existing Dublin cloud region, which is costly to relocate, while directing new investment where power is cheaper and the queue is shorter.

ISO New England

ISO-NE proposed excluding new large loads from its Installed Capacity Requirement entirely, forcing data centres to bring their own generation or accept curtailment, even though New England's demand has been flat or falling for a decade. The reform is pre-emptive, written before the region has felt the strain PJM and MISO already report.

Ofgem

Ofgem opened its Data Centre Commitment Fee consultation on 29 July, arguing that free options on grid capacity let developers bank places on ten sites and abandon nine, crowding out projects that would build. It expects the deposit to filter the queue before the rule even takes final form.