Meta's cash went negative for the first time since 2012. Why does the model still say $746?

$32.8 billion in, $34.9 billion out: the quarter the ads stopped covering the AI bill

Meta ($META) reported revenue of $64.60 billion for the quarter to September 2026, up 26%, with the average price per ad up 14% and impressions up 10%. The advertising machine is running faster than at any point since 2021. Operating income still fell 37% to $12.99 billion, because a $9.89 billion charge for the August 2026 settlement with 52 state attorneys general landed in the quarter. Strip it out and operating income would have been $22.88 billion and diluted EPS $7.40 instead of the reported $3.83.

The number that changed the picture sits lower down the statement. Operating cash flow was $32.79 billion; capital expenditure including finance-lease payments was $34.91 billion. Free cash flow - what is left after the data centres are paid for - came to negative $2.12 billion, against a positive $10.63 billion in the quarter to September 2025. Over nine months it fell to $11.05 billion from $29.51 billion, a 63% drop while revenue grew 29%.

The December quarter will not reverse this. Guidance puts revenue at $73-76 billion, 22-27% above the $59.89 billion of the quarter to December 2025. Capex guidance was narrowed to $135-145 billion for 2026; with $85.82 billion already spent in nine months, that implies $49-59 billion in the final quarter alone, more than the whole of 2024.

Zero buybacks, $24.9 billion of new debt and a share class built to be sold

Meta repurchased $0 of its own stock in the nine months to 30 September 2026, against $26.25 billion in the same period of 2025. The basic share count rose 1.4% to 2,552 million. Long-term debt went from $58.74 billion at 31 December 2025 to $80.92 billion after a $24.91 billion bond issue, while cash and securities held at $82.45 billion. Insiders made 278 sales worth $258.9 million in 12 months and no purchases.

Then comes the Class C stock. From 19 January 2027 every Class A and B share receives one non-voting Class C share, which trades like a 2-for-1 split. The information statement lists its first purpose as capital raising, followed by employee pay and acquisitions. It was approved by written consent of Zuckerberg-controlled entities holding 56.1% of the votes; 80% of his Class C shares are tied to his voting stock, 20% are freely saleable. On the 21 October call management stated there are no specific financing plans for the new class yet. The shareholder who was paid $26 billion in buybacks in 2025 now holds the instrument that would fund the next data centre.

Fair value $746.55: why the model pays 26x earnings but balks at 44x cash

The FPI fair value is $746.55, 7.0% above the $697.52 close on 21 October 2026, a verdict of fair. On earnings the stock looks reasonable: 26.4x against an industry median of 17.1x, below Microsoft at 29.6x and above Alphabet at 16.7x. On cash it is the dearest it has ever been: 43.9x free cash flow, against 24.6x at the end of 2021 and 36.8x at the end of 2025. The free cash flow yield of 2.28% means $100 of stock produces $2.28 of spare cash a year, below the industry's 3.46%.

What lifts a DCF for this company is the quality of the profit: a net margin of 29.8% against 16.2% for the industry, a return on invested capital of 17.1% against 8.1%, revenue compounding at 19.9% a year for three years, and guidance implying another 22-27% in the December quarter. What holds it back is that today all of that operating cash is being poured into servers, plus up to $18.59 billion of settlement payments still to be made. A DCF values the cash a business will throw off once the building stops; the price today is paying for a building site.

Street vs model: $775 median, $903 in the last month, $1,000 at the top

Wall Street rates the stock a buy: 2 strong buy, 54 buy, 11 hold, 2 sell. The median target of $775 from 6 October 2026 sits 3.8% above the FPI fair value, a gap small enough to be noise. The nine targets issued in the last month average $903.33, 21% above the model; Truist (13 October) and Wells Fargo (6 October) at $1,000 sit 34% above it. The low of $595 is 20% below. After the results Goldman Sachs put its target at $860 on 21 October, following Deutsche Bank at $820 on 16 October, Guggenheim at $850 on 14 October and Pivotal at $930 on 13 October.

The gap has a simple mechanism. The banks price 2027 earnings with the $9.89 billion charge removed and with Muse, which reached 10 million users in six weeks, counted as a new revenue line. Half of the FPI model is a cash-flow calculation, and the cash flow just turned negative; the 20% consensus weight is what pulls the FPI number up toward the Street rather than away from it. Every target and the fair value halve mechanically once the Class C dividend is paid in January 2027.

Three numbers before 27 January 2027: $74.5 billion, $59 billion and $746

The next results are due on 27 January 2027, where consensus expects $73.65 billion of revenue, below the $74.5 billion midpoint of guidance. A print at or above $76 billion, Reality Labs losses tracking toward the promised $15 billion in 2027, and a 2027 capex figure that does not jump far above $145 billion would give the DCF the cash it needs to move up. The stated target of a more than 10x cut in Muse compute cost within a year is the lever that would turn capex back into free cash flow.

The downside path is equally numeric: 2027 capex well above $145 billion, a first Class C issuance that dilutes without votes, revenue under $73 billion, or a youth-related trial verdict on top of the $18.59 billion already agreed. Three signals to watch: December-quarter capex against the implied $49-59 billion; the share price against the $746.55 fair value above and the $632.62 200-day average below; and the 19 January 2027 effective date, after which every per-share figure in this report is divided by two.