Meta bets its balance sheet on “personal superintelligence” as ad engine reaccelerates

AI ambitions meet holiday-fueled momentum

Meta closed 2025 with the unmistakable swagger of a company whose core business is humming again, even as it embarks on one of the most capital‑intensive pivots in consumer tech history.

Fourth-quarter revenue from the Family of Apps — Facebook, Instagram, WhatsApp and Threads — jumped 25% year on year to $58.9bn, powered by record holiday demand and a step‑change in ad performance from its AI systems. Total ad revenue rose 24% to $58.1bn, with impression growth of 18% and a 6% increase in average price per ad, suggesting Meta is not merely stuffing more ads into feeds but charging more for demonstrably better outcomes.

CFO Susan Li underlined that demand was “healthy across all verticals” except politics, as Meta lapped the prior year’s US presidential cycle. Online commerce was the standout contributor to growth, followed by professional services and technology advertisers — a telling barometer of both consumer spending and digital marketing budgets.

Against that backdrop, CEO Mark Zuckerberg used the call to sketch a much larger prize. He described 2026 as the year when a “major AI acceleration” begins to reshape both Meta’s products and its own internal ways of working, anchored on a vision of “personal superintelligence” woven through social feeds, messaging and new hardware like AI‑enabled glasses.

From feeds to “agents”: a new product spine

Zuckerberg’s framing of the next platform shift was strikingly expansive. Where previous Meta updates have toggled between near-term engagement tweaks and long-dated metaverse dreams, this quarter’s narrative threaded them together through AI agents.

The company has spent 2025, he said, “rebuilding the foundations of our AI program.” Over the coming months Meta expects to roll out its first new models from Meta Superintelligence Labs and associated products — not as a single flagship launch, but as a rolling series of releases that demonstrate a “rapid trajectory” in capability.

At the heart of this is the idea of AI agents that truly understand a user’s personal context: history, interests, social graph and content creation habits. Zuckerberg argued that Meta’s long‑standing strength — intimate, behavioral data at scale — gives it a shot at delivering a “uniquely personal experience” that pure-play model providers cannot easily match.

This is more than a new chatbot in the corner of an app. Internally, Meta is fusing its large language models with the recommendation engines that already decide what 3.5bn people see on Facebook, Instagram, Threads and in ads. Today’s systems, Zuckerberg admitted, are “primitive compared to what will be possible soon.” Tomorrow’s feeds, in his telling, will be less like static playlists and more like interactive conversations with an assistant that can both surface and generate content tuned to an individual’s goals.

That shift has clear commercial overtones. Ads today help merchants “find just the right very specific people,” as Li put it; new “agentic shopping tools” are meant to invert that relationship, helping users find just the right products from Meta’s sprawling commerce catalog, across both feeds and business messaging. WhatsApp, in particular, is set to become far more capable as these tools mature.

An ad machine sharpened by AI

Beneath the sweeping rhetoric, the quarter’s hard numbers suggest that AI‑driven optimization is already reshaping Meta’s operating metrics.

On the engagement side, product changes in Q4 drove a 7% lift in views of organic feed and video posts on Facebook — the largest quarterly revenue impact from Facebook product launches in two years. The company is also prioritising “freshness and originality” in recommendations: Facebook now surfaces over 25% more same‑day Reels versus the prior quarter, while on Instagram the share of recommendations coming from original posts in the US reached 75%, up 10 percentage points in just three months.

Threads, often dismissed as a side bet in the social wars, is quietly riding the same wave. Recommendation improvements drove a 20% increase in time spent in Q4, even before Meta begins to broadly scale advertising on the service; ads will expand to the UK, EU and Brazil this month.

On the monetisation side, Meta has been methodically rebuilding its ads stack on top of foundation models and unified architectures. The GEM model — its foundational system for ads — now spans all major surfaces on Facebook and Instagram, and Meta doubled the GPU cluster used to train it in Q4. Knowledge from GEM is distilled into lighter‑weight inference models that run at scale, a technique Li said is delivering measurable gains.

In the quarter, a new runtime model rolled out across Instagram feed, Stories and Reels drove a 3% lift in conversion rates. Meanwhile, consolidating disparate Facebook Stories and other models into a unified “Lattice” framework helped deliver a 12% increase in ad quality. In an industry where percentage points translate into billions of dollars, those are meaningful increments.

Crucially, Meta sees no obvious ceiling yet. Li argued this is “the first time we have found a recommendation model architecture that can scale with similar efficiency as LLMs,” opening the door to far larger models trained on longer interaction histories and richer signals — including, eventually, sharing signals between organic content and ad ranking. That, alongside more “session‑aware” recommendations and deeper content understanding from LLMs, is expected to drive further engagement and revenue gains through 2026.

Business messaging graduates to a revenue pillar

Beneath the headline ad engine, business messaging is starting to look like a second, if smaller, growth franchise.

Click‑to‑message ad revenue growth accelerated in Q4, with the US up more than 50% year on year, thanks in part to “website‑to‑message” formats that allow users to explore a product site before opening a chat. Paid messaging on WhatsApp has scaled to more than a $2bn annual run rate, and Meta is expanding its “business AIs” — conversational agents that handle customer queries — from early markets like Mexico and the Philippines to more geographies.

Those business AIs are already handling over a million weekly conversations on Meta’s messaging platforms. The near‑term use cases are prosaic — checking product availability, answering basic questions — but Meta’s roadmap envisions chatbots that can help users complete actions, including purchases, directly inside WhatsApp. If that vision materialises, Meta would be grafting a transactional layer onto its messaging spine in markets where WhatsApp is often a de facto digital identity.

A record capex cycle, with compute at the centre

If the revenue side tells a story of compounding AI returns, the cost guidance lays bare the price of staying in the race.

Meta ended the quarter with $81.6bn in cash and marketable securities and $58.7bn of debt, leaving it in a net cash position even after returning significant sums to shareholders in recent years. Free cash flow in Q4 was $14.1bn, and full‑year capital expenditures reached $22.1bn, largely on data centres, servers and network infrastructure.

For 2026, Li guided to a startlingly higher capex range of $115–135bn — several times 2025’s level — as Meta builds out the infrastructure for Meta Superintelligence Labs and scales both training and inference workloads. Total expenses are expected in the $162–169bn range, with the bulk of growth coming from infrastructure and AI talent. Reality Labs’ operating losses are expected to remain similar to 2025, which Zuckerberg called the “peak” year before gradual improvement.

The newly christened MetaCompute initiative is meant to bring a semblance of order — and long‑term economics — to this investment binge. Zuckerberg cast it as a strategic programme to be “the most efficient” in the industry at engineering, investing and partnering around infrastructure. That includes long‑dated bets on silicon and energy, flexible data‑centre designs, strategic partnerships, contracted cloud capacity and new ownership structures for some large sites.

The appointment of Dina Powell McCormick as president and vice-chairman, with a mandate to interface with governments, sovereigns and other capital providers on capacity expansion, underscores how political and financial Meta’s infrastructure push is becoming. Silicon and power are no longer just inputs; they are strategic constraints that Meta is trying to convert into competitive advantage.

Li made clear that while Meta expects to fund 2026’s programme from operating cash flow — and still deliver operating income above 2025’s absolute level — it will “periodically supplement” that with “cost‑efficient external financing” and may ultimately run with positive net debt rather than the cash hoard investors have grown used to.

Reality Labs: losses peak, glasses take centre stage

While AI stole the narrative, Meta did not abandon its metaverse storyline; it reframed it.

Reality Labs revenue fell 12% in Q4 to $955m, as the company lapped the 2024 launch of Quest 3 and earlier channel stocking by retailers. Yet Zuckerberg emphasised one bright spot: sales of Meta’s AI‑enabled glasses, which he said more than tripled in 2025 and are among “the fastest‑growing consumer electronics in history.”

Where earlier metaverse commentary dwelt on VR headsets and virtual worlds, the new emphasis is on glasses as “the ultimate incarnation” of personal AI — devices that can “see what you see, hear what you hear, talk to you and help you as you go about your day,” even overlaying custom interfaces in real time. In Zuckerberg’s telling, we are at a moment akin to the flip‑phone‑to‑smartphone transition, with vision‑correcting eyewear as the installed base waiting to be upgraded.

Strategically, Meta is tilting Reality Labs investment towards glasses and wearables, with Horizon — its social virtual world — repositioned as an experience that will increasingly reach users on mobile, not just in VR. Zuckerberg painted an image of a future where any video a user watches could be tapped and “jumped into” as an interactive, 3D experience, one of many new “immersive and interactive” formats that he expects to sit alongside, not replace, traditional video.

For investors fatigued by years of metaverse losses, the key line was financial: 2025 and 2026 should mark the peak of Reality Labs’ red ink, even as Meta continues to pursue the long‑term bet.

Internal productivity: AI reorganises the org chart

Meta is not only selling AI; it is reorganising itself around it.

Li disclosed that since the start of 2025, output per engineer has risen 30%, with the bulk of that coming from increased use of AI coding tools. For power users of those tools, output is up 80% year on year. Zuckerberg described an emerging pattern where “projects that used to require big teams now be accomplished by a single very talented person,” and said Meta is “elevating individual contributors and flattening teams” as a result.

The subtext is twofold. On one hand, internal productivity gains help offset the sheer magnitude of Meta’s infrastructure and headcount investments, giving management some cushion to maintain operating income growth. On the other, they hint at a cultural shift inside one of the world’s largest tech employers, as agents become co‑workers and the balance of power nudges further towards star engineers and away from middle management.

Zuckerberg gestured to a broader macro implication: “2026 is going to be the year that AI starts to dramatically change the way that we work.” For now, that shows up at Meta as faster shipping cycles; over time, it could redraw the cost structures of software companies more widely.

Legal and regulatory shadows over a bright outlook

The earnings script was not entirely unclouded. Meta continues to navigate intensifying regulatory and legal scrutiny, particularly in the EU and US.

In Europe, the company has “aligned with the European Commission” on further changes to its “less personalised” ads offering and will begin rolling those out this quarter. While no detailed impact was quantified, Li acknowledged these changes could pose headwinds later in the year, especially as 2026’s early foreign‑exchange tailwinds fade.

In the US, youth‑related litigation remains a looming risk. Meta has several trials scheduled this year, and Li cautioned that they “may ultimately result in a material loss.” For now, the business is sufficiently profitable that such outcomes would be absorbed, not existential. But the juxtaposition is stark: a company pushing the frontier of personalised AI while simultaneously being scrutinised for the impact of its products on young users.

A familiar engine, a transformed balance sheet

For investors, the story that emerges from Meta’s fourth quarter is one of continuity and discontinuity layered together.

The continuity lies in the ad machine. Despite regulatory pressures, platform shifts and macro volatility, Meta has restored high‑teens to mid‑twenties growth by making its ads more performant and its feeds more engaging, using precisely the sort of data‑hungry machine learning that first vaulted it into dominance a decade ago. Business messaging and WhatsApp monetisation add incremental legs to that stool.

The discontinuity is in the balance sheet and the ambition. A capex budget north of $100bn, Reality Labs losses at their peak, a full‑scale in‑house AI research lab and a dedicated infrastructure strategy arm courting sovereign capital — these are not the hallmarks of a cautious, mature cash cow. They are the moves of a company trying to ensure that, when AI agents and AR glasses become mainstream interfaces, it will not be relegated to renting someone else’s models.

Zuckerberg was candid that his answers on Meta Superintelligence Labs would be “somewhat unfulfilling” until the first wave of models and products ships in 2026. Yet his tone suggested a clear conviction: that the next decade’s platforms will be built at the intersection of personalised AI, immersive media and ubiquitous messaging — and that Meta, with its 3.5bn‑strong user base and refreshed infrastructure, intends to be one of the firms setting the pace rather than paying to catch up.