Meta beat, got upgraded, and fell 7%. The cash flow statement says why.

Introduction

On October 21 Meta reported a quarter that, on the lines most people watch, was good: revenue up 26% year over year and earnings above what analysts expected. Brokers responded by raising price targets as high as $995. The next day the stock fell 6.98%, its fifth-worst session of the year, and took the companies that build data centres down with it.

That combination — a beat, upgrades, and a 7% fall — is the whole subject of this post. A chart alone cannot explain it, so we start with what the company reported and what was said about it, and only then read the levels.

Q3 digest: the number that mattered

Revenue was never the problem. Meta booked $64.6bn in the September quarter against $51.2bn a year earlier, growth of 26.1%, and trailing twelve-month growth is 27.5%. Gross margin is 81.0% and operating margin 32.9%. The advertising business did not wobble.

The problem is two lines further down. Here is the cash the business generated against the cash it spent on property and equipment, by quarter.

Subtract one from the other — and the finance-lease principal Meta also pays — and you get free cash flow: the money left for buybacks, dividends and a rainy day. It has gone from $14.1bn a quarter to below zero in three quarters.

Several outlets reported this as Meta’s first negative free cash flow since it went public in 2012, which we cannot verify from the filings we pulled — our series starts in 2025. What we can say from the filings is that it is the first negative quarter in that series, and that the swing came from spending rather than from any deterioration in the business.

What was said about it

Three things arrived alongside the numbers, and the market reaction is hard to read without them.

A raised capital-spending outlook. Coverage on the day attributed the fall squarely to this: investors “fret over negative cash flows amid raised capex”, as one summary put it. The chart above is why that lands.

A new non-voting share class, and a stock split. Meta announced a structure that lets it issue equity to fund the AI buildout without diluting Mark Zuckerberg’s control. Read generously, that is prudent financing. Read sceptically, a company with $82bn of cash pre-arranging a way to sell more stock is telling you something about how long the spending runs.

The credit market noticed. Reporting that day described Meta’s credit risk gauge hitting a record high and the bond market sending a warning on its borrowing. That is striking for a company whose debt-to-equity is 0.30 — the concern is evidently the trajectory, not the current balance sheet.

A $25bn consumer-protection judgment in New Mexico was also reported that afternoon. We have not independently verified the figure or its status, and flag it as reported rather than settled.

What makes the reaction legible is that the sell-side disagreed with the tape. Citigroup raised its target to $905, UBS to $892, DA Davidson to $800; Wells Fargo stayed Overweight while cutting to $995. The stock fell 7% through all of it. And it was not alone: GE Vernova fell about 10% and Vertiv about 7% the same day — the suppliers of the buildout, repriced on the possibility that the buildout gets financed more carefully.

Nothing in the advertising business broke. What changed is the price the market will pay for a dollar of growth that has to be bought with borrowed money.

That connects this post to our brief on mortgage spreads: the cost of capital rose for a homebuyer and for a trillion-dollar advertiser in the same month, for recognisably the same reason.

Start here: what the chart terms mean

The rest of this post reads the price action. Four terms cover all of it, and none of them know anything about cash flow.

A moving average is the average closing price over a stretch of days, recalculated daily. The 50-day summarises the last couple of months; the 200-day, the last year. Above both means the stock has been rising; below both, falling.

Support and resistance are prices where a stock has repeatedly stopped falling or stopped rising. They are not laws. They matter because enough people watch the same levels to give them a mild self-fulfilling quality, and because breaking one tells you something changed.

RSI compresses fourteen days of gains and losses into one number from 0 to 100. Below 30 is conventionally oversold, above 70 overbought. Neither is a verdict on value.

Bollinger bands draw a channel two standard deviations either side of the 20-day average, so it widens when the stock gets jumpy. Closing outside the lower band means the fall is large even by the standard of how much this stock normally moves.

What the trend looks like

The honest description is sideways, not falling. Meta spent the summer between roughly $550 and $630, underneath both averages. In September it went nearly vertical — one session, September 21, added 11.3%, the largest single-day gain of its past year — peaked at a closing high of $777.59 on September 24, and has spent the month since giving it back. A year of this leaves the stock down 2.65% in 2026 while SPY is up 11.45%.

The 50-day average crossed above the 200-day on September 23: the “golden cross”, the most widely taught bullish signal there is. The closing high came the next day.

The two levels that matter

The ceiling is 0.3% away and the floor is 18.4% away. That gap is not a prediction of a fall — it is a statement about where the evidence runs out. Between $642 and $524 there is no price at which Meta has recently stopped going down, because it has not traded in that range since March.

Two things stop that being as ominous as it sounds. Meta has closed below its 200-day average on 180 of the last 318 sessions — 56.6% of the time. For this stock, being under the long-term line is the normal state. And being 17.4% below the one-year closing high is similarly ordinary: 29.4% of sessions qualify.

What to watch next

We do not forecast prices, and none of the below is a prediction. These are the numbers that would settle the questions this post raises.

$644.26 — the 200-day average. Meta closed $1.64 under it. Reclaiming and holding it would mean the fall stopped at the level that has historically held; losing it decisively is what opens the gap below. It is the only level in play, which makes the next few sessions unusually legible.

$678.55 — the 50-day average. Every rally this month has stalled beneath it. Until Meta closes above that line the one-month trend still points down, whatever any oscillator says.

Free cash flow, next reported January 26, 2027. This is the number that caused the fall, so it is the number that resolves it. One negative quarter during a buildout is a decision; a second and a third are a condition. Watch whether operating cash flow re-accelerates or capex guidance comes down — either closes the gap, and nothing else does.

39.6% annualized volatility. At a 4.06% average daily range Meta can cover the distance to its 200-day average in one session. Levels that look meaningful on a chart are, at this volatility, about a day apart.

Valuation, on the two measures that now disagree

The spending shows up as a split between the earnings multiple and the cash multiple. On profits Meta looks reasonable; on the cash left after the buildout it does not.

Twenty-two times earnings against 26% revenue growth is not demanding. But 12.3 times operating cash flow becomes 39.9 times free cash flow, and 59.6 times on an enterprise basis, purely because of capital spending. Which of those two numbers is the real one depends entirely on whether the spending is temporary — and that is a judgement about Meta’s AI programme, not something any ratio or chart decides for you.

What the oversold signal has actually paid

Across five years and 1,255 sessions, RSI closed below 30 on thirteen separate occasions. One month later Meta was higher in ten of the thirteen, by a median of 9.65%, beating SPY by 6.5 percentage points. Measured that way, oversold Meta looks like a gift. Then we changed the window.

The five-year window contains the 2022 bear market, when Meta fell 74% and oversold signals fired all the way down; eight of the thirteen episodes live in the stretch the shorter window throws away. Only the one-month row survives both cuts, and thirteen observations is not a sample — moving the threshold from 30 to 29 reshuffles the medians. We found the same decay when we tested 637 episodes across 55 large stocks, a far better sample, and there too the edge disappeared past a month.

Risks to reading it this way

The spending may be the right call and still hurt the stock for a year. Nothing in a cash flow statement tells you the return on money spent on data centres. If the AI programme works, this quarter is a footnote; the market is simply unwilling to pay for the option while free cash flow is negative.

Our cash flow series is short. It starts in Q1 2025, so “first negative quarter” is a statement about seven quarters, not about Meta’s history. The widely reported claim that this is the first since the IPO is plausible and we have not verified it.

One day is not a verdict. The 6.98% fall came on 40.6M shares against a 50-day average of 20.2M. Heavy, but a single session, and the same week carried a market-wide AI-spending selloff that would have pulled Meta down regardless.

The news is reported, not audited. The share class, the credit gauge and the New Mexico judgment come from press coverage on October 22. The cash flow figures come from the filings.

What this means for your money

This is a cash flow question wearing a chart’s clothes. Anyone deciding about Meta today is really deciding whether the capital spending is an investment or a new permanent cost. The levels tell you what the market is doing while it makes up its mind; they do not answer it.

Beware the upgrade. Four brokers raised targets into a 7% fall. Price targets describe where an analyst thinks a stock should trade, not where it will, and this is the cleanest example you will see all year.

For option sellers, 39.6% volatility with no catalyst until January is a specific trade. You are being paid for ordinary turbulence rather than event risk, which is a different proposition from selling into an earnings date. Our market weather report covers the distinction and the Radar scores it on live quotes.

If the one-month bounce is the thesis, write the exit down. It is the only horizon both windows agree on.

Data appendix

Every figure above with the date it belongs to. Price studies are the ones this site runs on every symbol it scans — Wilder RSI(14), Bollinger Bands(20, 2), Wilder ATR(14), Chaikin Money Flow(20), the 50- and 200-day EMAs and the support/resistance logic in src/signals/technicals.ts — so they match what the scanner would show you for META.

Method: Meta’s filings report cash flow year-to-date, so each quarter is the difference between consecutive filings, and Q4 2025 is the annual report less the Q3 year-to-date figure. That reconstruction reproduces the published quarterly free-cash-flow series to the dollar for every quarter we could check against Meta’s own reporting, which is why we trust the −$2.1bn. Highs and lows are closing highs and lows, so they will not match a broker’s intraday 52-week range. A backtest episode is the first session a condition becomes true; forward returns are close-to-close over 5, 21, 63 and 126 sessions with the matching SPY window subtracted. Current levels come from Tradier; the five-year backtest needs more history than Tradier returns for this symbol, so it uses Yahoo Finance closes cross-checked against Tradier where they overlap. Pulled October 25, 2026.