Meta spent Monday doing something the world's second-largest advertising business has never done: selling equity to pay for its ordinary course of business.

The company launched two concurrent underwritten offerings totalling about $35bn. One is $22.5bn of Class A common stock, sold in connection with forward sale agreements with Morgan Stanley Capital Services and Goldman Sachs — meaning the banks borrow and sell existing shares now, and Meta is paid when it settles, on dates it chooses up to 19 November 2027. The other is 250 million depositary shares, each representing a 1/20th interest in a share of a new Series A Mandatory Convertible Preferred Stock with a $1,000 liquidation preference — roughly $12.5bn, with a Nasdaq listing planned under the symbol METAM. Underwriters get 30-day options on a further $3.375bn of common and 37.5 million more depositary shares.

On top of that sits the part that matters for years rather than weeks: a planned at-the-market programme of up to $60bn in Class C non-voting stock, expected to start in the first quarter of 2027 once Class C begins trading. No agreement for it has been signed, and Meta's statement is explicit that the timing, size and terms are undetermined. Add it all up and the number in the headline is $95bn.

The preferred is a bond wearing equity's clothes

Mandatory convertible preferred is the instrument a company reaches for when it wants debt's certainty and equity's accounting. Holders get a fixed dividend until roughly 15 August 2029, when the shares convert automatically into a variable number of Class A shares — or into Class C, once those exist. The dividend rate and conversion terms are set at pricing, which means the market, this week, gets to name a price for Meta's AI risk. A high coupon would be the market saying the buildout looks like a credit story; a low one would say it looks like a growth story that happens to be financed oddly.

The circularity is the story

Buried at the bottom of the release, under "Cornerstone Investors": Nvidia, AMD and Broadcom have, "severally and not jointly", indicated an interest in buying up to an aggregate $5.0bn of the Class A shares at the same price and terms as everyone else. Meta's own wording is careful — these are indications, not commitments; any of the three may buy more, fewer or no shares.

Five billion dollars against a $95bn programme is a rounding error. As a signal it is the entire 2026 AI financing debate in one paragraph. Three companies whose revenue is a direct function of how much Meta spends on accelerators and networking have been invited to put capital into the customer that pays them, on the same terms as public investors, days after that customer admitted it cannot fund the purchases from cash flow.

The arithmetic Meta is trying to outrun

The company guided 2026 capital expenditure at $135–145bn. It generated $97bn of operating cash flow in the first three quarters and held roughly $82bn in cash and marketable securities at the end of September. Third-quarter capex, including principal payments on finance leases, was $34.9bn against operating cash flow of $32.8bn — free cash flow of about minus $2.1bn, a swing of roughly $12.7bn from the $10.6bn positive a year earlier. Revenue was $64.6bn, up 26%, and net income $9.9bn; the advertising engine is healthy, and it is no longer sufficient.

Meta's defence of the move is that it holds Aa3/AA- credit ratings, can borrow cheaply, and prefers a "balanced mix of operating cash flow, investment-grade debt, strategic partnerships, and equity capital". That is precisely why the equity sale is the interesting part. Debt is available to Meta on excellent terms; choosing permanent capital instead says something about how large the number after 2026 looks.

Dilution you cannot vote against

Every share sold through the ATM will be Class C: no votes, full economic claim. Meta filed the amended certificate creating the class in Delaware on 21 October. Existing holders receive one Class C share per Class A or Class B share, a distribution that roughly doubles the share count and behaves like a two-for-one split — except that afterwards the board can issue new Class C shares, for capital, acquisitions or employee pay, without asking shareholders again.

Mark Zuckerberg holds about 60.6% of the votes on a 13.4% economic stake, and would receive roughly 342.4 million Class C shares. Eighty per cent of them are stapled: he cannot sell without surrendering an equivalent number of voting shares, with 2.5% released each qualifying quarter — about eight years to full release. The special committee described negotiations with him as "protracted and, at several points, contentious". He had told the board he was weighing a move out of California over Proposition 40, a one-off 5% wealth tax on billionaires resident in the state on 1 January 2026, which Californians vote on 3 November; he has pledged to stay for eight years.

Alphabet ran this experiment in 2014, after a Delaware shareholder suit and a settlement that compensated Class C holders if the new class traded more than 1% below Class A in its first year. That discount risk is now Meta's to manage.

What the market has already said

When the recapitalisation was disclosed with third-quarter results last week, Meta fell about 7% in a session, roughly $124bn of market value, and the Nasdaq dropped 2.6%. Its five-year credit default swap spread hit a record intraday high of 114.8 basis points on 23 October — equity and credit repricing the same fact from opposite sides. The stock was down about 9.4% on the week.

What to watch

Three things, in order. The preferred's dividend rate when the offering prices: that number is the market's estimate of Meta's AI risk, denominated in basis points. Whether the three chipmakers' indications become orders. And whether Class C, once it lists in January, trades at a discount to Class A — the price of dilution without a vote, quoted daily.

Sources