Meta and the day control stopped being free
Posted on: 2 November 2026
When Deliveroo came to the London market in the spring of 2021, several of the City's largest fund managers, Aviva Investors and Legal & General among them, let it be known that they would not be buying. Part of the objection concerned the riders, but the part that mattered for the price concerned the share structure, which gave the founder, Will Shu, twenty votes for every share he held for the first three years. The shares fell by roughly a quarter on the first morning and the financial press settled on a phrase that has followed the company ever since, the worst flotation in London's history. What the boycott said, underneath the headlines, was that British institutions were perfectly willing to hand their money to a founder but expected a say in return. London has spent much of the time since rewriting its listing rules so that founders with weighted votes no longer have to go to New York. This month the same question came back from across the Atlantic, at a scale that makes Deliveroo look like a rounding error.
On 21 October Meta, the owner of Facebook and Instagram, reported third-quarter revenue of $64.6bn, up 26 per cent on a year earlier and ahead of what analysts had expected. The following day the shares lost almost 7 per cent to close at $648.85, on a day when the Nasdaq fell 2.6 per cent. The explanation offered almost everywhere was the familiar one, that investors are frightened of how much the company is spending on artificial intelligence. That is only half true. Meta has been announcing that spending for some time and the shares had already slipped on the same worry in July. What changed in October was who would be paying for it.
The number that mattered was not revenue. Free cash flow, the money left over once the business has paid for its running costs and its investment, went negative to the tune of $2.12bn, something that had not happened in more than a decade. Capital expenditure on data centres and chips reached $34.9bn in three months, the forecast for 2026 as a whole moved up to between $135bn and $145bn, and on 2027 the company said only that it would continue to invest aggressively. Costs rose 68 per cent in the quarter, swollen by a $9.89bn charge linked to the settlement reached in August with American state attorneys-general over litigation about teenagers' mental health, and the operating margin halved from 40 to 20 per cent. Long-term debt, meanwhile, had risen in nine months from $58.7bn to $80.9bn.
The move that explains the fall sat in a filing published the same day. Meta is creating a new class of listed shares, Class C, which carry no votes at all and which from 19 January 2027 will be handed to every existing shareholder at the rate of one for each share already held. On the share price the effect is that of a two-for-one split. On control the effect today is nil and tomorrow it is decisive. Mark Zuckerberg, who founded the company and still runs it, owns about 13.4 per cent of the equity but controls 60.6 per cent of the votes through Class B shares that carry ten votes each. Every future capital raising done in Class C will bring in fresh money without adding a single vote, so his economic stake can shrink while his control stays exactly where it is.
He has tried this before. In 2016 Facebook proposed an almost identical scheme and dropped it in September 2017, on the eve of a trial brought by shareholders in Delaware. This time the board voted on 19 October and shareholder approval arrived the next day by written consent, a procedure that dispenses with a meeting when a majority of the votes sits in one pair of hands. The filing is also candid about the hurry. An independent committee of the board recorded that borrowing on the scale management had in mind risked damaging the company's credit rating, and that it would be prudent to raise capital before the money available was soaked up by the other giants of the industry. The road of debt was narrowing, which left the road of equity.
Five days later, on 26 October, the raising began. Meta sold $25bn of ordinary shares at $625 apiece, a figure increased during the offer from the $22.5bn first announced, and alongside it issued mandatory convertible preferred shares paying 6.5 per cent, securities that behave like a bond until 2029 and then turn into shares whether the holder likes it or not. It also announced a programme of around $60bn of Class C shares to be sold directly into the market once those shares exist. The buybacks that for years steadily reduced the number of shares in issue have stopped.
Anyone who has watched markets for long enough will recognise the instrument. In 2014 Google distributed non-voting Class C shares to its shareholders, one for each share held, so that it could use stock to pay staff and buy companies without diluting the control of its founders, Larry Page and Sergey Brin. The instrument is the same, but the circumstances are reversed. Google did it while cash was arriving faster than it could be spent, so its Class C shares were an insurance policy on the future. Meta is doing it in the quarter in which cash has stopped being enough, so its Class C shares are a tap. In one case the paper protects control. In the other it keeps control standing while money is requested from people who will never have a voice.
In forty years of founding and selling companies I have learned that control is measured on the day you need other people's money, and not a day earlier. As long as a business pays for everything out of its own profits, power concentrated in a founder's hands costs nobody anything, because whoever buys the shares knows he will have no say and receives in exchange a machine that produces cash and buys back its own stock. When tens of billions have to come from outside, the bargain changes character. Whoever puts in the new money takes on a growing share of the risk of a wager worth hundreds of billions on artificial intelligence, while whoever makes the wager carries an economic share of it that grows thinner with every issue. Nassim Taleb, the Lebanese-born essayist who has built much of his work on the asymmetry between those who decide and those who bear the consequences, would call it skin in the game wearing thin while the power stays intact.
None of this means the wager is wrong. Revenue is growing by a quarter a year, there is $82.45bn of cash and securities on the balance sheet, and the company expects fourth-quarter revenue of between $73bn and $76bn. The question is not whether Zuckerberg is right about artificial intelligence, since nobody can know that yet. The question is who pays if he turns out to be wrong.
There is one further detail in the offering. Nvidia, AMD and Broadcom, the companies that sell Meta the chips for its data centres, indicated a non-binding interest in buying up to $5bn of the new shares. Suppliers financing a customer so that the customer keeps buying is not a new sight. Around 2000 Lucent and Nortel, two of the largest makers of North American telecoms equipment, lent heavily to the operators that bought their kit and found those loans returning to their own balance sheets as losses when the operators failed; Marconi was the British casualty of the same bust. Five billion dollars in a plan worth ninety-five is small, and it does not turn 2026 into 2000. It does suggest that the circuit of capital behind artificial intelligence is beginning to close on itself.
For anyone sitting on a board, or looking after a family's money, the lesson has less to do with Meta than with control structures designed for times of plenty. London's long argument about weighted voting shares has been conducted almost entirely as a question of competitiveness, of whether the City can afford to turn away founders who would otherwise list in New York. That argument may well be right. What Meta's filing adds is a view of the moment the bill arrives, because the company states in its own words that it expects the new Class C shares to trade at a discount to the ones that carry votes. The fall on 22 October should not be read as a verdict on artificial intelligence, which the market is in no position to deliver, and the split is cosmetic. The line worth reading is the one in which a company of that size writes, in its own hand, that the shares it is about to sell will be worth less than the ones that already exist, because that line tells the buyer exactly how much he counts.