Summary: Meta reported with Q4 2025 earnings that advertiser-facing video generation tools reached a $10 billion revenue run-rate in Q4 2025, growing nearly three times faster than overall ads revenue. It also cited a 3.5% lift in Facebook ad clicks from GEM after training GPUs doubled and a 24% increase in incremental conversions from incremental attribution. The article cautions that a run-rate is extrapolated, not booked revenue.
On January 28, 2026, together with its Q4 2025 earnings, Meta put out a newsroom report named “2026: AI Drives Performance” that spells out what AI is genuinely doing to ads and engagement. The figure that lands hardest: taken together, the advertiser-facing video generation tools reached a $10 billion revenue run-rate in Q4, and their quarter-over-quarter pace is nearly triple that of overall ads revenue. Set against the lingering doubts about Meta’s AI capex, the report’s point is plain — AI is no longer merely a spending narrative; it has become an earning one, and the payback is surfacing first in ad delivery, the piece of the business nearest to cash.
Video Generation Tools: A $10 Billion Annualized Business
- The tools advertisers use to produce video creative and ad variants added up to a $10 billion revenue run-rate in Q4 2025, a figure the earnings call itself confirmed.
- That quarter-over-quarter pace is nearly three times overall ads revenue growth — advertisers keep adopting generative creative at a quickening rate rather than leveling off.
Next to mid-single-digit growth for the ads business overall, a ten-figure product line compounding at triple the speed is the strongest case Meta can make that its AI outlays pay for themselves. It also shows how quickly the ground moved: in about two years, generated video creative went from experiment to the default workflow for mainstream advertisers.
The Payoff From Ranking: GEM, Lattice, and Instagram’s Newest Model
- After the GPUs used for training were doubled, the GEM ads ranking model produced a 3.5% lift in Facebook ad clicks and a conversion gain above 1% on Instagram.
- A newly added Instagram run-time model pushed conversion rates up another 3%, while reworking the Meta Lattice architecture improved ads quality by 12%.
- Run each of those percentages against Meta’s ads base and you get effects worth billions of dollars a year.
The method deserves attention: none of this came from plugging in a larger model. It came from patient engineering on ranking infrastructure — more compute, architectural tuning, better choices about where models run. It is a quiet, compounding infrastructure wager rather than a launch event, and it is precisely what rivals will find hardest to copy in a hurry.
Attribution and Business Messaging: Billion-Dollar Lines of Their Own
- Incremental attribution, measured against standard attribution, lifted incremental conversions by 24% and, seven months after launch, already runs at a multi-billion-dollar annual rate.
- US click-to-message ads revenue climbed more than 50% year over year, helped along by the Website to Message ad format, and paid WhatsApp messaging passed a $2 billion annual run-rate.
- In Mexico and the Philippines, Business AIs now handle more than 1 million conversations each week — AI sales agents carrying live traffic, not a pilot program.
One logic ties attribution and messaging together: AI earns money not just by making content but by demonstrating that ads genuinely perform, and that proof counts most for keeping small advertisers around.
Where Engagement Collects Its Share
- Facebook’s ranking changes lifted views of organic feed and video posts by 7%, and its surfaces now serve over 25% more same-day Reels than they did in Q3 2025; Threads gained 20% in time spent, and 75% of Instagram recommendations in the US now point to original posts.
- The number of Meta AI daily actives generating media tripled year over year, and AI dubbing now spans nine languages.
What improves on the engagement side flows back into advertising: longer watch time and sharper recommendations enlarge the pool of impressions available to monetize.
Three Cautions Before You Cite It
First, a run-rate extrapolates to a full year — nothing about it is booked revenue — and Meta has a knack for dressing algorithmic iteration up as an “AI story,” so credit the direction and haircut the magnitude. Second, the gains are the product of ranking, attribution, and creative generation working in combination, not of any single AI product; what most companies can actually copy is the workflow, not the model. Third, the message to advertisers is blunt: generated video creative and variant testing are now standard equipment in mainstream buying, and arriving late carries a price. That matches what we expected in our 2026 opening outlook: AI’s payback is migrating from chips and cloud toward the places nearest to cash — ads and business messaging.