Meta needs compute and it is going to pay for it with paper. The Menlo Park company announced on Monday, October 26, a raise of up to $95 billion to finance its artificial intelligence infrastructure, and the detail that best captures the moment is on the other side: Nvidia, AMD and Broadcom, the manufacturers it buys hardware from, have expressed interest in buying up to $5 billion in new shares.
The money does not come in all at once. Meta is seeking $95 billion split into two tranches, with $35 billion in underwritten offerings: $22.5 billion in class A common stock and $12.5 billion in mandatory convertible preferred stock. The rest, up to $60 billion, comes through an at-the-market sales program that will not start until the new class C non-voting shares begin trading, something expected in the first quarter of 2027, according to the press release sent to investors.
The bill behind it explains the haste. Meta expects to spend between $135 billion and $145 billion in capital this year, and the round exceeds by $12.55 billion all the cash and marketable securities it held on September 30. The stock fell as much as 1.6% on Monday morning before paring losses. The question is no longer whether Meta can pay for the AI party, but how much of that party those who sell it the shovels are willing to finance.
The first tranche is already in the market and is split into two instruments. Meta is placing $22.5 billion in class A shares with forward sale contracts, so it collects nothing on the day of the placement and will receive the cash when they settle, which it expects to close before November 19, 2027.
The remaining $12.5 billion comes in 250 million units, each equivalent to one-twentieth of a mandatory convertible preferred share. The company retains the usual over-allotment options, with $3.375 billion and 37.5 million extra units if demand is strong.
| Tranche |
Maximum amount |
Instrument |
When |
| Underwritten offerings |
$35 billion |
$22.5 billion in class A and $12.5 billion in mandatory convertible preferred |
Now, with options for $3.375 billion and 37.5 million units |
| At-the-market program |
$60 billion |
Class C non-voting shares |
From the first quarter of 2027 |
| Total transaction |
$95 billion |
Equity, not debt |
Between 2026 and 2027 |
The second tranche changes the scale of the operation and also its uncertainty. Meta has not signed any agreement for the at-the-market program and acknowledges that the size and terms are not decided, so that $60 billion could arrive later or never arrive at all.
Suppliers cover at most 5% of the paper
The $5 billion from Nvidia, AMD and Broadcom sounds like resounding backing and is a modest portion: it equals 5.3% of the $95 billion Meta is seeking. The company speaks of interest, not commitment, and specifies that they may buy more, less or nothing, at the same price and on the same terms as the rest of the investors.
The gesture has industrial logic. Nvidia already invests in rented-compute providers that buy its chips, and now its three big silicon sellers sit in the capital of a customer. If Meta cuts orders, the punishment hits them two ways at once: less revenue and a stake that loses value the same day.
Anchor investors have existed in stock market listings for years. What stands out here is the context, with a buyer asking its three suppliers to finance the share sale with which it is going to pay them.
The quarter that broke the cash
The third-quarter accounts explain why Meta is seeking capital instead of waiting. Revenue rose 26% to $64.6 billion, but capital spending was $34.9 billion and free cash flow came in at negative $2.12 billion, compared with positive $10.63 billion a year earlier, according to the results published last week.
Net income of $9.86 billion includes a $9.9 billion charge tied to litigation. At the end of September the company had $82.45 billion in cash and marketable securities against $80.92 billion in long-term debt, and its future spending commitments total about $825 billion.
The credit market is already pricing it in. Five-year credit derivatives hit all-time highs last week, the same Thursday the stock fell almost 7% when negative free cash flow was disclosed, and a vehicle tied to its El Paso data center paid 7.53% for $12.5 billion.
Debt is not free through the conventional route either. In April it placed $25 billion in bonds and the order book came in at $96 billion, 23% below the previous issue, a sign that buyers are starting to demand a premium for AI risk.
Muse, or the argument that the spending already pays off
The official narrative says the investment is already returning something. Meta claims Muse surpassed 10 million users in the United States and Canada in the weeks after its September launch, that shopping is the main use for more than 25% of those who open it each week, and that two million businesses use its business agent.
It is worth noting that those figures are published by the party asking for the money and that none comes with attributed revenue. Management itself places the contribution of AI businesses in 2027, not in the quarter just closed, and the company's first frontier model, Muse Wonder, is only weeks old.
The agent is today a valuation argument rather than a line in the results, as was seen when Meta presented Muse Spark Open as its largest open model. Advertising grew 26% and sustains the cash, while the new spending has no date of return.
Who pays for the shovels when capital is scarce
The backdrop is a sector-wide bill that no one disputes. Goldman Sachs estimates that AI-related capital spending will total about $7.6 trillion between 2026 and 2031, counting compute, data centers and energy, a figure that explains why even the sector's most profitable company is turning to the capital markets.
The same Monday, Elon Musk added pressure from another flank: more than 200 rented-compute providers face outright death and consolidation, he said, because debt markets have shut the tap on any company without revenue beyond AI data centers.
Shares of CoreWeave, Nebius and Iren are down between 30% and 40% over a month. Meta has advertising that pays the bills and they do not, so the same capital drought does not hit equally those who sell compute without end customers.
By financing itself with shares instead of bonds, Meta shifts part of the bet to the shareholder and avoids inflating a balance sheet that is already nearing parity between cash and debt. The suppliers that join the capital take on that exposure from the other side of the counter, while the company keeps buying chips for four superclusters of more than a gigawatt.
The paradox is that Meta runs much of its production AI on Google's TPUs, not only on the silicon of those now buying its shares. The money from the share sale does not buy technological loyalty, it buys capacity to sustain a pace of spending that has already eaten up free cash flow.
What will be known when the 2027 program starts
Three data points remain to be confirmed and none depends on Monday's press conference. The first is how much dilution the shareholder will tolerate if the at-the-market program is executed in full, because $95 billion is a lot of new paper for a company that until now financed itself with cash.
The second is whether the manufacturers' $5 billion becomes a firm order or remains a gesture of sympathy when the price does not cooperate. The third concerns the heart of the narrative, with management promising AI revenue for 2027 and no quarter yet to back it up.
That is where it will be known whether the round was prudent financing or the first chapter of a longer correction. The difference is not decided by the size of the share sale, but by the speed at which Muse and the business agents start generating revenue.