Meta announced the largest equity financing in its history on Monday, October 26, 2026, and almost every headline described it as a $95 billion raise. The SEC filing that actually launched the deal tells a more interesting story: of the $95 billion headline, Meta receives roughly $12.5 billion in cash right now. The rest is either deferred by up to thirteen months behind forward sale agreements or contingent on an at-the-market program that, per the filing, has no signed agreement, no determined size, and no determined terms.
That distinction is not legal nitpicking. It changes how you should read the entire "hyperscalers are running out of cash for AI" narrative, and it reveals one of the most carefully engineered capital structures a large-cap company has ever put in front of the market.
What Meta Actually Filed
The company filed a free writing prospectus with the SEC on October 26 launching two concurrent underwritten offerings:
- $22.5 billion of Class A common stock sold through forward sale agreements with Morgan Stanley Capital Services and Goldman Sachs & Co. The banks borrow shares and sell them into the market today; Meta delivers its own shares and collects cash later, on dates Meta selects, no later than November 19, 2027. The filing states plainly that Meta "does not expect to initially receive any proceeds" from this tranche.
- $12.5 billion in depositary shares (250 million shares, each representing a 1/20th interest in a new Series A Mandatory Convertible Preferred Stock, $1,000 liquidation preference per preferred share). These convert automatically into common stock on or about August 15, 2029. Meta expects this tranche to pay cash upfront, listing on Nasdaq under the ticker METAM.
The over-allotment options confirm the split: underwriters get options for up to an additional 37.5 million depositary shares and up to $3.375 billion of additional Class A stock. Both are exactly 15% of their respective tranches.
The third piece, the roughly $60 billion at-the-market program, is not a transaction at all yet. Meta says it plans to launch it in early 2027, only after its new Class C common stock begins trading, and the filing includes explicit language that no agreement exists and there is "no assurance" the program happens as described. Benzinga's coverage and Anadolu's summary both lead with $95 billion; the filing's own "Financing Plan" section is where you learn that $95 billion is a sum of one done deal, one deferred deal, and one intention.
| Component | Size | Cash to Meta today | Dilution today | Notes |
|---|---|---|---|---|
| Class A common, forward sales | $22.5B | ~$0 | No | Settles on Meta's schedule by Nov 19, 2027 |
| Series A mandatory convertible preferred | ~$12.5B | ~$12.5B | No | Converts around Aug 15, 2029; ticker METAM |
| ATM program in new Class C stock | ~$60B | $0 | No | Planned for early 2027, no agreement signed |
| Chipmaker cornerstone interest | up to $5B | part of above | No | Nvidia, AMD, Broadcom; non-binding |
Why the Delayed Cash Is a Feature, Not a Bug
Read the forward sale agreements as an engineer reads an API contract, and the design intent becomes obvious. A conventional secondary offering forces a company to accept today's price and today's dilution. Meta's structure separates the three decisions that a single offering bundles together:
- Price discovery now. The banks sell shares into live demand this week, while the story is fresh and demand from cornerstone investors is on the table.
- Settlement timing later. Meta picks the delivery dates. Because the forward price adjusts with the stock, a higher share price at settlement means Meta issues fewer effective dollars of dilution per dollar raised, or equivalently collects more cash per share delivered. Meta holds something economically close to a call on its own share price for thirteen months.
- Voting control permanently. The mandatory convertible converts into Class A today, but the filing contains a switch: once the new non-voting Class C stock is issued (announced October 21, effective January 19, 2027, structured as a one-for-one dividend on Class A and B, economically like a 2-for-1 split), the preferred can convert into Class C instead, and Meta can settle the forwards in Class C shares only.
That last clause is the load-bearing wall. Zuckerberg holds about 60.6% of voting power on a 13.4% economic stake. Every dollar in this plan dilutes economics while leaving the voting stack untouched, because new supply flows into a class with zero votes. Tech Funding News notes the Founder Alignment Agreement further locks this down: Zuckerberg can freely sell only 20% of his Class C shares, must sell or convert other shares to reduce his stake, and must remain a California resident for eight years. Meta raised the maximum amount of equity a public company has ever sought while making it structurally impossible for that equity to change who controls the company.
The Funding Gap, Computed
Meta is not cash poor; it is cash-flow negative on infrastructure. Here is the arithmetic from the filing and Q3 disclosures, with each assumption labeled.
Assumption 1: run-rate operating cash flow for full-year 2026 equals the reported first nine months annualized: $97B / 9 x 12 = about $129B. (Meta reported the $97B figure in the filing itself.)
Assumption 2: 2026 capex lands at the midpoint of Meta's own guidance: ($135B + $145B) / 2 = $140B.
Gap: roughly $11B of capex above operating cash generation for 2026, before debt service, buybacks, or the 2027 ramp. The Q3 picture is sharper: capex of $34.91B against $32.79B of operating cash flow produced negative free cash flow of $2.12B in a single quarter.
Context on the balance sheet: Meta ended Q3 with about $82B in cash and marketable securities against $80.9B of long-term debt. The October 2025 $30B bond issue and the $29B Hyperion data center financing already used much of the easy debt capacity, and the company disclosed $279B of future AI data center lease commitments in its Q2 filing. So the honest framing of Monday's news is not "Meta needs $95 billion to survive." It is: Meta faces a roughly $10-15B annual funding gap now, growing into a 2027 where capex is expected to keep climbing, and it chose to pre-fund the gap across three instruments with three different cost profiles while preserving both its Aa3/AA- ratings (debt capacity stays open) and its founder's votes.
The $12.5B of real upfront cash covers slightly more than a year of the current gap. Everything else is insurance against the gap widening.
The $5B Question: Circular, or Just Vertical Integration in Reverse
Nvidia, AMD, and Broadcom expressed non-binding interest in buying up to $5 billion of the Class A offering collectively, on the same terms as other investors. $5B is under 4% of Meta's 2026 capex, so the optics matter far more than the amount. Two readings compete:
The bear reading is vendor financing dressed up: chipmakers propping up demand for their own product by buying equity in their customer, echoing Nvidia's $100B OpenAI commitment. LSEG researchers David McNay and Robin Marshall published an analysis on October 21 warning that across the industry, "ownership webs can make tracking the revenue to earnings a forensic accounting exercise."
The bull reading notices the AMD detail. In February, AMD granted Meta warrants for up to 160 million AMD shares at one cent each, contingent on chip purchase milestones. If AMD now buys Meta stock, the two companies hold claims on each other's equity, which is exactly how long-horizon supply contracts get enforced when the assets are too specific and the horizons too long for a purchase order to carry the risk. Equity is what both sides use when they want commitment device semantics, not investment semantics. Cornerstone stakes here look less like OpenAI-style vendor financing (where the vendor funded the buyer's purchase directly) and more like mutual hostages with dividends.
Neither reading is complete. The verifiable fact is small: up to $5B, non-binding, same terms as everyone else.
What This Means for Builders and Watchers
Three concrete consequences follow from the structure rather than the headline.
First, treat announced AI capex and committed capital as different unit types. SpaceX's $40B callable GPU debt announced this past weekend is closed financing; Meta's $60B ATM is a plan. The industry conversation routinely adds the two together. The record $10B GPU-backed loan is collateralized by actual chips; an ATM program is collateralized by a stock price that has to remain acceptable while $60B of new supply slowly hits the tape.
Second, watch the METAM pricing this week for the real cost signal. The mandatory convertible's dividend rate and conversion premium, to be set at pricing, are what the market actually charges Meta for deferred dilution. That number will tell you more about perceived AI capex risk than any $95B headline.
Third, the Class C mechanics set a template. Alphabet raised $80B in June with underwritten and ATM components; Anthropic is reportedly preparing its own entry to equity markets. When the largest spenders discover they can raise tens of billions without touching founder voting control, the non-voting-class trick becomes standard infrastructure for the AI buildout, and "whose company is this" drifts further from "who owns the shares."
Meta's own framing in the filing is one line worth quoting: the company intends to fund investment through "a balanced mix of operating cash flow, investment-grade debt, strategic partnerships, and equity capital." Monday was the day equity joined that mix, and the filings show a company that would rather deliver shares in 2027 than accept dilution in 2026. That is not the behavior of a borrower backed into a corner. It is the behavior of a borrower who expects its own stock price to do the negotiating. Whether that expectation holds is the single question the next thirteen months of AI capex, and the November 2027 settlement date, will answer.