Thesis. The bellwether for hyperscaler capex on this watchlist — $130–145B guided for 2026, with ~$35B of CoreWeave and 2,609 MW of Vistra nuclear on its books — whose free cash flow is now shrinking fast enough that the market has sold three consecutive beats.
Reviewed 19 Sep 2026, after Q2 2026 results on 29 Jul. Next review after Q3 2026 earnings on 28 Oct 2026.
What it does
Meta is here as a bellwether, not a category constituent. It is one of the four largest buyers of almost everything in this file — and increasingly a competitor to parts of it. An advertising business growing 28% funds a buildout of frontier-model training capacity, and its capex guidance plus its named supplier deals are direct demand signals for the neoclouds, the power producers and the hardware stack.
Capex guidance is the number. Not revenue, not EPS. A hyperscaler's capex line is a supplier's revenue line, and Meta has explicitly said near-term capacity is worth more to it than 2028 flexibility — it wants the hardware now.
The most important question about this page is not whether Meta is a good business. It is whether $130–145B a year survives contact with a shrinking free cash flow line.
Bull case
As a demand signal.
- FY26 capex guided $130–145B. Q1 underspent ($19.84B against ~$27.6B expected) while full-year guidance went up — which implies $35–42B per quarter through H2 2026, more than double the year-ago pace
- Explicit prioritisation of near-term capacity over 2028+ flexibility, "since near-term capacity is more valuable"
- Q2 revenue $60.80B, +28% (+27% constant currency), with ad impressions +14% and price per ad +12%. The engine funding all of this is still accelerating
- Operating margin 30.9% despite total costs rising 55%
- Every named supplier commitment is multi-year and large — see Major customers below
- Financing is being supplemented with long-duration debt and partnerships such as BlackRock, which extends how long the spend can run
- Headcount down 1% to 75,472 after a ~8,000 reduction in May — cost discipline outside the buildout
Bear case
As a demand signal.
- Free cash flow is dwindling and the market has punished it three quarters running. Meta beat on revenue and EPS the prior quarter and still fell 6–7% on the capex number. A beat alone has not moved this stock in three consecutive reports — and a hyperscaler under FCF pressure is the first place capex gets cut
- CFO Susan Li declined to give a 2027 capex figure. The visibility every supplier in this file depends on stops at December
- Q2 EPS $6.18 missed $7.18 consensus by 14%, including $2.40B of legal charges and $1.18B of severance. FY26 expenses raised to $165–169B
- Q3 revenue guided $61–64B, weaker than expected
- "Meta Compute" is a competitive threat to the neoclouds it currently buys from — signalled as a future compute-rental business, directly against CRWV and NBIS
- Legal: several US youth-related trials scheduled this year that the company says "may ultimately result in a material loss"
- Reality Labs still guided to roughly $19B of annual losses, on par with 2025
Major customers
Not applicable in the usual sense — Meta is a buyer. Its commitments are other companies' backlogs:
- CRWV — $21B added in Q2 on top of a prior $14B, ~$35B through 2032. The largest single neocloud contract in this database
- VST — 20-year PPAs for 2,609 MW of nuclear capacity, deliveries from late 2026
- CEG — 20-year agreement for ~1.1 GW from the Clinton nuclear plant from 2027, Meta's largest energy deal
- AVGO — the custom MTIA accelerator, in production shipments for inference and recommendation
- AMD — multi-generation Instinct deployments (6 GW)
- NBIS — $12B of dedicated capacity from early 2027, plus a commitment to take up to $15B more
- P (Everpure) — hyperscaler storage win for internal use
- — turbine demand via its power partners
Meta's own paying customers are advertisers: millions of accounts with no meaningful concentration.
What would change the view
- Q3 results on 28 October — whether FY26 capex is reaffirmed, raised or cut. After NVIDIA, this is the most important single demand datapoint on the watchlist
- Any 2027 capex commentary, since the refusal to guide is itself the current signal
- Meta Compute launch timing and scale
- Free cash flow trajectory — the constraint that would force a capex decision
- Whether the CoreWeave, Vistra or Constellation contracts see any modification
- MTIA volumes showing up at Broadcom
Update log
19 Sep 2026 — Converted to the standard template. Added the Nebius commitment to the supplier list ($12B dedicated plus up to $15B additional, confirmed from Nebius's own disclosures), which was missing here.
29 Jul 2026 — Q2 2026 reported. Revenue $60.80B (+28%), EPS $6.18 against $7.18 consensus — a 14% miss including $2.40B of legal charges and $1.18B of severance. Operating margin 30.9%; total costs $42.03B (+55%). DAP 3.60B (+3%). Q3 revenue guided $61–64B; FY26 capex narrowed to $130–145B; FY26 expenses raised to $165–169B. 2027 capex guidance declined. Stock fell on light revenue guidance and dwindling free cash flow.
Research and education only — not investment advice.
Made with Bullet