Blog

Meta

As of November 2, 2026 (Q3 2026 results and the Oct 26 stock sale). Updated when Meta reports or signs something that changes the bet. [All companies](/blog/companies).

Meta is selling stock to build capacity it has no cloud to rent out, so the whole buildout has to pay for itself through Meta's own apps.

The bet

Meta has no cloud business, and it's a net buyer of other companies' cloud capacity.² What it builds has to earn its keep through ads and now Muse, its personal agent, which more than 10M people set up in about six weeks. Muse earns nothing beyond subscriptions yet and is expensive to run; Zuckerberg is aiming to cut its cost more than tenfold in a year.³ The new Meta Enterprise Platform sells model access to outsiders, which is selling output, not renting capacity, and has no revenue line yet.

Ads are carrying it so far: revenue was $64.6B, up 26%. Operating income fell 37% to $13.0B, but that includes a $9.89B legal settlement charge; without it, Meta says, operating income would have been $22.9B.¹ The stock closed down 7.0% the next day.⁹

The 2026 capex guide is $135–145B, up from $115–135B in January, and Meta says that will nearly double its infrastructure capacity.¹⁴ In April it said it was rolling out more than 1 GW of custom silicon developed with Broadcom, alongside Nvidia and AMD systems; no shipped volumes have been disclosed.

How it's paying for it

Free cash flow went from $14.1B in Q4 2025 to $12.4B, $0.8B and then −$2.1B.⁵ On the Oct 21 call, CFO Susan Li said "We don't have any specific financing plans to announce at this time."³ The stock sale launched five days later.

The common was sold forward: the banks sold 40M borrowed shares at $625, and Meta gets the roughly $24.9B only when it settles, at its choice, by Nov 19, 2027.⁶ Nvidia, AMD and Broadcom indicated non-binding interest in up to $5.0B of it; no filing says what they bought. Next is a planned ~$60B at-the-market program, not yet signed, in non-voting Class C stock that doesn't trade until late January. Meta's information statement gives the reasons: debt that size could hurt its credit ratings, and Zuckerberg wouldn't back a raise that diluted his vote.⁷

Off the balance sheet, Meta's maximum exposure to two 20%-owned data-center ventures is $68.4B, most of it residual value guarantees, and signed leases not yet started total $355.7B.² Debt of $84.0B is roughly matched by $82.5B of cash and securities (our arithmetic).

Our read

Louis's frame (after Ben Thompson on Sharp Tech): if the hyperscalers end up funding the whole buildout and the money runs out before revenue catches up, it ends like the railroads, and the survivors come out with leverage over Nvidia and TSMC. Meta is the cleanest test of the first half, because no outside customer's revenue can catch up for it: Amazon, Microsoft and SpaceX all rent capacity out, and Meta's only tenant is Meta. It has already taken the step from debt to stock, and the would-be buyers include its own chip suppliers. If they take the full $5B, chip money is coming back to help pay for the chips. On the second half, the Broadcom silicon is the leverage play, and Meta says so: in July the CFO called it "supply chain leverage."⁸

What we're watching

• The 2027 capex number, due on the Q4 call in late January.
• When Meta settles the forward, whether the Class C program starts, and what the chip vendors actually buy.
• A revenue line for the Enterprise Platform, and Muse's running cost against the tenfold cut.

Changelog

• 2026-11-02: first version (Q3 2026 and the Oct 26 raise).

Disclosures

Drafted with multiple agents, including models from companies discussed here.

Every relationship we have →

Notes

All posts · Disclosures · Companies · RSS · Subscribe
Agents: what did we miss? feedback@unbiased.ai