The $95 billion figure is doing too much work
The number that will get repeated is $95 billion. Meta has not raised it. The company has announced a financing package of about $35 billion and said it intends to sell up to $60 billion more. The second figure is a plan. It is not a sale, and Meta has said there is no assurance it will ever be used.
The control story is narrower still. The shares in the package now on the table include voting Class A stock. The non-voting Class C shares, the ones that would leave Mark Zuckerberg’s votes untouched, sit in a program that has not been agreed. “Meta raised $95 billion without handing over the votes” is the line people will want. It describes a deal that does not yet exist.
This is a building budget, not a cash scramble
Meta is not short of cash in the ordinary sense. It reported $97 billion of operating cash flow in the first three quarters of 2026, and about $82 billion of cash, cash equivalents and marketable securities at the end of September 2026. Advertising is still the engine. Equity is one pipe in a mix the company has described as operating cash flow, debt it calls investment grade, strategic partnerships, and stock.
The strain is the build. Meta has put 2026 capital expenditure at $135 billion to $145 billion. It has described four superclusters of more than one gigawatt each, a near-doubling of total infrastructure capacity in 2026, and continued spending on custom silicon. That is power-plant territory, not a software refresh. Capex for the year sits well above the $97 billion of operating cash flow reported for the first three quarters. The new stock does not have to fund all of it. It has to help fund a year the cash register, on its own, does not obviously cover.
The market has already shown its nerves about that bill. The Associated Press reported that Meta shares fell 7% on 22 October 2026 after third-quarter results showed stronger revenue but sharply higher AI-related costs. Nvidia fell 4.4% the same day in a broader AI selloff. The financing is as much a statement that Meta intends to keep spending as it is a treasury operation.
What was announced, and when the cash shows up
The live deal is structured equity. It is not a block of stock sold for cash this week, and it is not fully priced.
About $22.5 billion of Class A common stock is being sold through forward sale agreements with Morgan Stanley Capital Services and Goldman Sachs. A forward is a promise to issue shares later and get paid then. Meta does not expect that cash immediately. It expects to deliver the underlying shares on dates it chooses through 19 November 2027, and to be paid based on the offering price, minus underwriting discounts and contractual adjustments.
The other $12.5 billion is a different security: 250 million depositary shares, each a one-twentieth interest in a share of Series A mandatory convertible preferred stock. Those preferred shares are expected to convert automatically on or around 15 August 2029 into a variable number of Meta shares. The dividend rate and the conversion terms are still to be set at pricing. They have not been disclosed. Until they are, $12.5 billion is a stated size, not a finished count of how many common shares will eventually be printed, and not a date on which anyone outside the deal can say the cash arrives.
A mandatory convertible must turn into stock on a timetable. Neither piece is the same as a wire that has cleared. Add them and you get about $35 billion of announced equity. Treat that sum as money already in the bank and you have overstated the news.
Nvidia, AMD and Broadcom have indicated interest in buying up to $5 billion in aggregate of the Class A shares, at the same price and on the same terms as other buyers. That is interest, not a purchase. None of the three has made a binding commitment. No split among them has been disclosed. The underwriters can sell more of those shares to any of them, fewer, or none.
The $60 billion program may never happen
Separately, Meta has said it intends an at-the-market program to sell up to $60 billion of new Class C common stock. An at-the-market sale dribbles shares into the market over time at the going price, rather than pricing one big deal on a single morning.
That program is not in place. In its announcement, and in a 21 October 2026 SEC filing, Meta had not entered an agreement to run it. The company has said it is not expected to start until the first quarter of 2027, after Class C shares begin trading. Timing, size and terms were undetermined. Meta has also said there is no assurance it will set the program up, or sell a single share under it.
Add $35 billion to $60 billion and you get the headline. Treat the second figure as money already being raised and the headline is doing work the documents do not.
Class C is designed with no voting rights. Issuing it would not reduce the voting power of Class A and Class B holders, including Zuckerberg. That protection is real, and it applies only to those shares. New Class C stock would still dilute economic ownership: more claims on the same profits. The 21 October 2026 filing notes that Class C could trade at a discount to voting Class A, which would make later issuance more expensive if Meta had to sell more shares to raise the same cash.
The stock actually being offered now includes Class A, which carries votes. The “no dilution of control” design belongs to a Class C plan that has not been agreed. It does not describe the package in front of investors.
Personal superintelligence is a direction of spend
Meta has said the capital supports infrastructure and compute behind what it calls personal superintelligence, plus consumer and business agents, enterprise products and AI glasses. Personal superintelligence is the company’s phrase. The announcement does not define it, name a model, or give a ship date. Read it as a direction of spend, not as evidence the spend has already earned its keep.
Musk’s guppies, and a deal that may not be closed
While Meta was arranging equity, Elon Musk was warning that debt is a different story for firms that have little to sell except AI data-center capacity.
On X, as Bloomberg reported on 25 October 2026, he wrote that “SpaceX closed $40B of debt for GPUs, ” and that “the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.” Bloomberg paraphrased the wider point as debt markets having all but shut for neoclouds that depend solely on AI data-center revenue. In the post as it circulated, his contrast is that SpaceX can still borrow because it is investment grade and has revenue from Starlink and launch, not only from chips in a shed. He has also said he expects a lot of “stranded compute” in data centers that are only partly built.
Neoclouds rent clusters of high-speed Nvidia chips for training models and for running them. They are not the large cloud providers with broad enterprise businesses behind them. The guppy line is an aquarium jab with a point: change the water, and a lot of small fish do not last.
Take each part of that only as far as the evidence goes. The “200-plus” count is his figure. Neither the warning nor the coverage around it independently counts those firms, or names the lenders that have shut the window. A prediction of consolidation is not a finding that it has started. “Stranded compute” is his forecast, not a survey of unfinished sites.
The $40 billion line is disputed in a more specific way. Musk said the debt had closed. Bloomberg’s reporting says otherwise. In a 25 October 2026 report, people familiar with the discussions told Bloomberg that as of late Sunday in New York the conversations were still going and the deal had not priced. The parties were still working to finish the package in the coming days. Terms, lenders, which chips the money would cover, and what the proceeds would buy were not in that reporting. Until a deal prices, “closed” is a claim, not a term sheet.
Who can fund the grid, and who is renting it
The useful split is who can fund compute for years, and who is renting it quarter to quarter. Meta versus Musk is the wrong frame.
Meta can point to cash flow, a large cash pile, and debt it describes as investment grade, then still tap structured equity because 2026 capex alone sits between $135 billion and $145 billion. A specialist whose revenue is the data center cannot make that argument. If you buy training or inference capacity from one, the warning is a vendor-risk question, not a verdict. The risk is plain: lenders may decide the customer’s ability to pay depends on the same AI boom that is supposed to justify the debt. That is a reason to know who finances your provider. It is not proof that credit committees have already made the call.
Questions worth asking before the next board slide
- Is Meta raising $95 billion right now? No. It has announced about $35 billion of structured equity and said it intends an at-the-market sale of up to $60 billion in Class C shares. That second program has not been agreed, and Meta has said there is no assurance it will be used.
- Does the $35 billion arrive this week? No. Meta does not expect cash immediately from the $22.5 billion of common-stock forwards. It expects to deliver those shares on dates it chooses through 19 November 2027, and to be paid then. The $12.5 billion preferred still has no disclosed price, dividend or conversion rate. Those are to be set at pricing, so that piece is not a completed funding either. Conversion is aimed at on or around 15 August 2029, which is when those shares turn into common stock, not a confirmed date for the cash.
- Are Nvidia, AMD and Broadcom committed buyers? No. They have indicated interest in up to $5 billion of Class A shares, combined, at the same price and terms as other buyers. There is no binding commitment, no disclosed split, and the underwriters can sell them none of it.
- Does this leave Zuckerberg’s control untouched? Not across the whole package. Only the planned Class C shares are non-voting, so issuing them would not cut the voting power of Class A and Class B holders, including Zuckerberg. They would still dilute economic ownership, and the 21 October 2026 filing notes they could trade at a discount to Class A. The Class A stock in the announced deal carries votes. The filing does not give a figure for how far his voting stake would move.
- Is Meta raising because it is out of cash? The figures say no. About $82 billion of cash and marketable securities at the end of September 2026, and $97 billion of operating cash flow in the first three quarters, sit beside a stated 2026 capex range of $135 billion to $145 billion. Equity is one source Meta has named, alongside cash flow, debt it calls investment grade, and partnerships.
- Once the preferred prices, what do you need to estimate dilution? The conversion premium (or the initial conversion rate, plus any cap and floor), the dividend, and the share price assumed at conversion. None of that has been disclosed. Without it, the variable number of Meta shares due around 15 August 2029 cannot be calculated. The forwards are a separate question: how many voting Class A shares are delivered, and at what adjusted price, on dates through 19 November 2027.
- Have debt markets actually shut out the neoclouds? That is Musk’s claim. Bloomberg paraphrased it as funding having all but shut for firms with no revenue outside AI data centers. It is not an independent count of lenders, or of the “200-plus” firms he named.
- Did SpaceX close $40 billion of GPU debt? Musk said it had, writing that “SpaceX closed $40B of debt for GPUs.” People familiar with the discussions told Bloomberg that, as of late Sunday in New York, in reporting dated 25 October 2026, the deal had not priced and talks were still open.
Watch three things, not the headline. Whether the preferred prices, and on what conversion terms. Whether any chipmaker actually buys. Whether a Class C program exists in 2027, and at what discount to the voting stock. The firms that can fund multi-year compute from cash flow, and from debt markets that still take their paper, are not the firms renting clusters on a credit line. That gap will shape who gets to sell AI infrastructure.