Meta reported third-quarter revenue of $64.6 billion, up 26% year over year, beating analyst estimates of $63.37 billion. But capital expenditures surged to $34.91 billion, pushing free cash flow to a $2.12 billion deficit — Meta's first negative free cash flow since 2012. Meta also raised the low end of its full-year capex forecast from $130 billion to $135 billion, with the high end unchanged at $145 billion, and said it would issue non-voting Class C shares to fund AI spending, acquisitions, and compensation. Tesla separately cut its 2026 capex forecast to about $20 billion from more than $25 billion, and said it and Musk's SpaceX are exploring a potential merger. The result was a broad AI selloff: Meta shares fell 7%, their largest drop since July; SpaceX declined 9%; Nvidia fell 4%; a
Investors are increasingly questioning whether the hyperscalers' massive AI infrastructure spending will generate adequate returns. Meta's negative free cash flow and record debt-insurance costs — five-year credit default swaps hit a record 114.79 basis points — signal that funding the AI buildout will require substantial investor participation, likely through stock sales. At the same time, Tesla's smaller capex budget spooked investors because its lofty valuation depends on future AI and robotics businesses such as the Cybercab robotaxi and Optimus humanoid robots.
Meta's third-quarter revenue rose 26% to $64.6 billion, while free cash flow went negative at -$2.12 billion — its first negative free cash flow since 2012. The Nasdaq Composite fell 2.6% on Oct 22, its steepest one-day decline since June 5.
MMeta just posted its first negative free cash flow since 2012, and the whole AI trade wobbled.
CThat is a real number. Revenue up twenty-six percent, and they still burned two billion dollars of cash.
MThe Nasdaq had its worst day since June. Meta fell seven percent. Nvidia, four. And Tesla dragged too.
CLet's be fair — Meta beat on revenue. Sixty-four point six billion. The problem is they spent thirty-five billion on data centers.
MExactly. The AI bill came due. You can't just say 'AI is the future' forever. At some point the cash has to come back.
CBut that's the play, Matilda. They're building the infrastructure before the revenue arrives. Same thing Amazon did with e-commerce for a decade.
MAmazon never went cash-flow negative in year thirteen of the buildout. Meta is selling new shares to fund this — non-voting Class C stock.
COkay, that part's a red flag. When you issue new stock to pay for the AI arms race, you're asking investors to dilute themselves.
MAnd their credit default swaps hit a record. The bond market is nervous. This isn't just Wall Street being impatient.
CBut what's the alternative? Stop spending, get left behind by Google and Microsoft? Nobody wants to be the company that skipped the infrastructure bet.
MRight, because spending has worked out so well for every company that went all-in on a new platform before it paid off.
CIt worked for Apple. It worked for Nvidia. It worked for the people who built the actual internet.
MAnd it destroyed everyone who built the railroads in the wrong places. There's a difference between investing and burning cash.
CBut the analysts at eToro said it — investors welcome big budgets when returns are visible. The question is whether Meta's AI products can monetize.
MNotice nobody answered that question with a number. It's always 'trust us, this builds the future.'
CThe fact that Tesla cut their capex forecast is interesting though. They went from twenty-five billion down to twenty billion.
MWhich spooked the market even more. Because Tesla's whole valuation depends on robotaxis and Optimus. If they're spending less, does that mean they're not chasing it as hard?
COr they've reached the point where the factories and chips are already paid for. You don't spend the same money twice.
MYou believe that? Tesla's robotaxi production is basically a prototype line.
CI believe they know their cash position better than analysts do. And the reported merger talk with SpaceX changes the math.
MA merger would be wild. Tesla plus SpaceX under one roof — but that's a topic for another hour.
CDifferent world entirely. Over on Bluesky, a metalworker's son named Michael Burchert asked a good question about his father's trade.
MWhat did he notice?
CHis dad's a welder and fabricator. He says that sector's seen disruption after disruption — robotics, laser welding, 3D printing.
MAnd construction hasn't?
CThat's his question. Why does metalworking constantly reinvent itself while construction crawls along doing the same thing for thirty years?
MThat's actually a sharp observation. And you know why, right? Construction is regulated to death. Every town has its own code, unions, lawsuits.
CBut welding has codes too. Pressure vessel standards, structural certification. That didn't stop robotic welding from taking over.
MWelding happens in a factory, mostly. You bring the part to the robot. Construction is outdoors, site-specific, weather-dependent.
CSo does that mean construction is just immune to automation? I don't buy it.
MI'm not saying immune. I'm saying harder. A robot that welds the same joint ten thousand times is one thing. A robot that adapts to a muddy jobsite is a whole different engineering problem.
CBut that's exactly why the payoff is bigger. Whoever cracks that wins a trillion-dollar market.
MAgreed. But there's a reason the biggest construction innovation in twenty years was a nail gun that fires twenty percent faster.
CYou say that like it's a joke. I've used one of those. It's a beautiful tool.
MYou would appreciate a faster nail gun.
CAbsolutely. I'd take it over a robot that gets stuck in the mud.
MMeantime, another small sign that the 'AI plateau' talk may be easing — an account called Anth Becker on Bluesky says coding models might be topping out, but office tasks and robotics still have clear room.
CThat feels right. Code generation is a narrow world. Managing your email and your calendar? Controlling a machine? Much broader.
MAnd that's the part that'll actually touch regular people, not just engineers.
CThe real prize is a model that runs your small business admin. Nobody needs a model that writes better React.
MSpoken like someone who's never had a React bug at two a.m.
COh, I've had those bugs. I just don't miss them.