Meta’s actual Q3 2025 call transcript is a masterpiece of corporate narrative, led by figures who act like children assigning executive roles to stuffed animals.
This transcript is the translation of what Meta executives would say if they were forced to admit they can read a balance sheet without supervision and a juice box.
The unsettling truth is that nobody on this call is steering the bus; they are simply documenting the route it decided to take today.
Kenneth Dorell, Director, Investor Relations
Good afternoon and welcome to Meta’s Q3 2025 earnings call.
This is the part where I remind you that everything we are about to say is forward-looking, which means if any of it turns out to be wrong, we will simply refer you back to this sentence.
We will reference both GAAP and…
Meta’s actual Q3 2025 call transcript is a masterpiece of corporate narrative, led by figures who act like children assigning executive roles to stuffed animals.
This transcript is the translation of what Meta executives would say if they were forced to admit they can read a balance sheet without supervision and a juice box.
The unsettling truth is that nobody on this call is steering the bus; they are simply documenting the route it decided to take today.
Kenneth Dorell, Director, Investor Relations
Good afternoon and welcome to Meta’s Q3 2025 earnings call.
This is the part where I remind you that everything we are about to say is forward-looking, which means if any of it turns out to be wrong, we will simply refer you back to this sentence.
We will reference both GAAP and non-GAAP metrics. GAAP is the version that counts for the SEC. Non-GAAP is what we use when we want the story to have a happy ending.
With that, I’ll hand it to Mark, who will now describe a cost explosion as a frontier opportunity.
Mark Zuckerberg, CEO
Thanks, Ken.
We had another strong quarter, which here means the ad engine kept us afloat while we dragged an AI lab and a hardware side quest as ballast.
We are positioning Meta as the leading frontier AI lab. We acknowledge missing the mobile and won’t make the same mistake twice, which is why we now hoard GPUs like a doomsday prepper hoards canned beans. “Open source AI” is the phrase we use because it makes regulators temporarily forget their job.
We are building an industry-leading amount of compute; though it’s purely speculative, given we lack an identifiable business model. Ask ten people when superintelligence arrives and you’ll get a scatter plot shaped like the trajectory of a drunk seagull stealing a sandwich.
We believe it’s prudent to spend more on projects that have less certainty.
We stopped chasing returns years ago. We chase scale now, because scale is the only metric that matters. The spending has become the strategy.
The truth is no one here knows how to stop spending, even if they wanted to.
About 3.5 billion people use at least one of our apps every day. We still call it community because saying “inescapable virtual prison” makes people uncomfortable.
Threads claims 150 million daily actives, which is a generous interpretation of a mis-tap. Time on Facebook is up 5 percent. Video time on Instagram is up over 30 percent. The content is worse, which proves we correctly diagnosed society.
We’ve entered the era where the feed teaches itself what you’ll tolerate. Our job is simply to pretend we’re the authors when we’re really the audience.
On ads, the story is more believable. We unified dozens of smaller models into fewer, larger ones and now describe common sense efficiency gains as scientific breakthroughs. Automated tools push over $60 billion in annual spend. “End-to-end AI-powered” means the system runs the show and the entire point of your now redundant job is to articulate to your boss whatever it just did.
The company is three giant transformers: Facebook, Instagram, and the ad engine. We’re turning them into one system that governs what the world sees and what advertisers pay for access, and none of us could stop it if we tried.
More than a billion people use Meta AI each month, mostly because we grafted it onto every input surface that accepts text. Usage rises as the model improves. We call that engagement. The more accurate word is dependency.
The apps keep people inside. The AI makes the feed harder to escape. The hardware makes sure you take us with you.
Susan will now translate optimism into line items.
Susan Li, CFO
Thanks, Mark.
I’ll walk through the numbers and gently suggest that nothing about them is alarming.
Our Family of Apps community keeps growing. We estimate that over 3.5 billion people used at least one of our apps daily in September. Q3 Family of Apps revenue was $50.8 billion, up 26 percent. Ad revenue was $50.1 billion, also up 26 percent.
The machine is still very much alive and funds our corporate strategy, which is whatever Mark’s dart lands on.
Our job in Finance is to turn whatever happened into something that sounds planned.
Total ad impressions increased 14 percent. Average price per ad increased 10 percent. Impression growth was strongest where monetization is weakest.
That has been true for a decade. We are used to it.
Reality Labs brought in $470 million, up 74 percent. Some of that was retailers front loading. Some was increased demand because the headsets no longer resemble medical equipment.
Total revenue was $51.2 billion. Total expenses were $30.7 billion. A spread like that buys a lot of indulgence for whatever AI saga Mark decides to tell next.
To address the part you are all pretending not to look at: we’ve built data centers off-balance sheet through structures that maintain a delicate technical separation that allows Meta to keep roughly billions of dollars of assets and debt off balance sheet while continuing to provide all material economic support.
The key is that the rules permit it. That’s all anyone needs to know.
Headcount is 78,400, up 8 percent. This is 78,399 more people than Mark will tolerate once he figures out how to turn employees into GPUs.
Operating income was $20.5 billion at a 40 percent margin. That is the number we want you to remember.
Net income looked weak at $2.7 billion until you see the $15.9 billion non-cash tax charge we will never actually pay. Excluding that, net income was $18.6 billion. Tax law shifted, so we marked down future benefits we no longer qualify for.
Capex was $19.4 billion. The AI story we advertised showed up in the form of a bill. Once the bill shows up, we update the story to match it. That is the entire process. We’ve reached the stage where the explanation matters more than the math.
Free cash flow was $10.6 billion. We repurchased $3.2 billion in stock and paid $1.3 billion in dividends, a polite way of saying, “We know the capex scares you, so here’s a warm blanket.”
We ended with $44.4 billion in cash and $28.8 billion in debt. We are short of restraint, not liquidity.
On revenue drivers, nothing has changed. Keep people inside the apps. Squeeze more value from every minute they don’t escape. Time spent is up. Our virtual prison is working as intended.
Meta AI keeps growing, which mostly means we’ve glued it to anything that accepts text. Deep-dive answers now surface Reels on purpose. If you ask for information, we hand you a video.
Threads is growing. Ranking got better. Time spent is up 10 percent. We added direct messaging so people can now argue in smaller rooms too.
On monetization, our job is simple. Increase ad supply without anyone noticing. Improve performance just enough to explain why supply increased.
On capital allocation, our priority is to fund the AI story. When that feels heavy, we return capital to remind you the business is still absurdly profitable and to keep you from asking more questions.
Guidance: Q4 revenue $56 to $59 billion. 2025 expenses $116 to $118 billion. Capex $70 to $72 billion.
2026: compute needs rising faster than expected; expenses outpacing revenue; models commoditizing; business models lagging.
This concludes the optimistic portion of the call.
Analyst Q&A
Brian Nowak, Morgan Stanley: You’re spending enough on GPUs to destabilize a medium-sized economy. Show me one number—any number—that proves this isn’t a very elaborate group project in self-delusion. And while we’re at it, how bad is the hardware business this time?
Meta: Every compute dollar we spend just sharpens the ad auction, which is all anyone here cares about. A/B tests say, “green arrow up.” We will continue buying hardware until the grid protests under the weight of our machine’s ambition. Reality Labs is down because we convinced Best Buy to load inventory in Q3. We’ll blame Q4 on seasonality.
Mark Shmulik, Bernstein: Threads still looks like a witness protection program for Twitter refugees. Tell me what it wants to be when it grows up. Also, you’re calling this thing an inference cloud. When does that become an adult and turn into a business instead of a line item that scares accountants?
Meta: Threads exists to prove we can still ship a social product. The inference cloud is the ad engine with a new name. We price it low, scale it fast, raise it later. You know the drill. Whether the drill still works is a separate discussion.
Doug Anmuth, J.P. Morgan: Your feed models already know more about human impulse control than most therapists. Is there more juice to squeeze, or are we in the diminishing-returns phase? And please stop pretending expenses are disciplined when they are obviously running away from you.
**Meta: **Engagement goes up whenever the model gets bigger. That’s the entire business. Infrastructure is where the money goes. Headcount is what we point at so no one stares too long at the servers.
Ross Sandler, Barclays: Please tell me this spending stops. Even a lie would help.
Meta: It does not stop. The market rewards spending. We are doing what we’re told.
Justin Post, Bank of America: If advertisers automate everything, you own the entire funnel. Should I be terrified or impressed? Also, is all this compute actually necessary, or are we all just pretending the models demand it?
Meta: Advertisers prefer not to see how the sausage is made. Advantage+ spares them the view. The only constant is that each new layer of automation demands more compute. That’s the subscription they don’t know they signed up for. And once subscribed, nobody downgrades. Not us. Not you.
Youssef Squali, Truist: Your assistant is glued into every app. When does it cover even one percent of its own costs?
Meta: Our definition of monetization is elastic. The assistant stretches it further. The timeline is whenever investors stop indulging the free trial.
Closing
The official transcript will behave like a well-trained pet and remain asleep while the adults argue.
The unofficial version is a feral cat. The ad engine works. The infrastructure bill is a live grenade.
The relationship works because the market wants it to.