Meta's AI is not a tech narrative — it is the plumbing of its advertising business
Mark Zuckerberg has a habit of presenting every technical advance at Meta as a civilisational milestone. In the first-quarter 2026 earnings results, the language was, as usual, ambitious. But this time the numbers do the work the narrative does not need to do: $56.3 billion in revenue, a 33% year-over-year growth, and an advertising machine that increased the average price per ad by 12% while simultaneously expanding impression volume by 19%. Those two variables moving together, in the same direction and at the same time, are not a market accident. They are evidence that Meta's investment in artificial intelligence is producing something that most technology companies still cannot clearly demonstrate: a measurable improvement in advertiser willingness to pay.
The question that rarely appears in quarterly earnings analysis is the most useful one for understanding what is happening here. It is not how much it grew, but why the buyer of Meta's advertising inventory is paying more for the same space that cost less a year ago. The answer has less to do with the AI narrative declaimed on earnings calls and more to do with a concrete technical transformation in the architecture of ad delivery, feed personalisation, and the quality of the conversion signal that Meta can offer its clients.
The asset that does not appear on the balance sheet
Before arriving at the advertising figures, it is worth pausing on the audience metric. 3.56 billion daily active people across Meta's family of applications is not just a number of scale: it is the foundation on which the value of the inventory is built. But the figure itself no longer surprises anyone. What does matter, and what the market tends to misread, is the quality of the time those people spend inside the platforms.
Video watch time on Facebook grew by more than 8% globally, with a 9% increase specifically in the United States and Canada. On Instagram, improvements to the content ranking systems generated a 10% increase in the time spent on Reels. Those percentages are not vanity metrics. They translate directly into inventory available for monetisation, and more importantly, into inventory that the user is actively consuming, not merely that appears on their screen while they are doing something else. There is a structural difference between an impression served and an impression consumed, and Meta's recommendation systems are working specifically to close that gap.
The most revealing technical detail that Zuckerberg mentioned is the one that receives the least coverage: posts published the same day now account for more than 30% of recommended content in Reels, double that of a year ago. What that implies for the advertising buyer is significant. An inventory of fresh content, quickly indexed and delivered at the moment of greatest relevance, carries a higher attention rate than an inventory of aged content. Meta is not just showing more videos; it is compressing the cycle between content production and its optimal distribution, which elevates the quality of the context in which an ad appears.
This is the variable that does not appear in presentation decks but that explains the behaviour of the price per ad better than any other narrative. A higher-attention context is worth more to the advertiser. A model that can better predict the probability of conversion justifies a higher CPM. And Meta now has, according to its own figures, ad delivery systems — its internal models Lattice and GEM — that generated more than a 6% increase in the conversion rate for landing page ads. The Adaptive Ranking Model, which routes advertising requests towards the models with the highest probability of conversion, contributed in turn to an additional 1.6% improvement in conversion rates across the main platforms.
Eight million advertisers and a lesson in technology adoption
The Meta results contain an adoption figure that deserves to be read carefully, beyond its purely advertising dimension. More than 8 million advertisers are using at least one of Meta's generative AI tools, compared to 4 million at the end of 2024. That is a doubling in less than six months.
The speed of that adoption is not explained solely by the availability of the tools. It is explained by the fact that the advertiser perceives a measurable result: those who used the video generation feature recorded a 3% increase in their conversion rates compared to those who did not use it. The 3% may sound modest in the abstract. For an advertiser managing hundreds of thousands of euros in monthly advertising spend, a 3% improvement in conversion is an operational difference that justifies a workflow change without friction.
This is the most robust adoption mechanism that exists in the enterprise market: an improvement in outcomes directly attributable to the tool, measurable in the short cycle and sufficiently tangible that whoever tries it does not want to return to the previous process. Meta is not selling generative AI as an abstract category. It is delivering it as a function within the workflow that the advertiser already uses, and it is measuring it with the metric that the advertiser already has as an objective. The friction of adoption is minimal because the entry point is familiar and the result is verifiable.
The AI assistant for advertisers, already fully deployed, is resolving account issues at a rate 20% higher than in the early testing phases. That is relevant not only as a product metric but as a retention signal. An advertiser whose account problem is resolved more quickly has less reason to look at other platforms.
The $19.84 billion in capex and what it reveals about the real bet
Meta closed the quarter with $19.84 billion in capital expenditure, and raised its annual estimate for 2026 to a range of between $125 billion and $145 billion. That figure is where all the analytical tension around the case is concentrated.
An operating income of $22.9 billion with a 41% margin and a net profit of $26.8 billion in a single quarter provides enough room to absorb aggressive capex without the operating cash flow suffering any immediate strain. But the pace of that investment says something more than financial comfort: it says that Meta is betting that the competitive advantage in digital advertising will be decided at the model infrastructure layer before the visible product layer. Servers, data centres, network capacity. The generative AI that the advertiser sees is merely the interface of something that requires a computational foundation of a magnitude that most competitors cannot match or even approach.
Chief Financial Officer Susan Li was specific about the direction of that investment: greater depth of historical interaction data for training the models, recommendation architectures that can operate with more granularity over user interests, and capacity for the ad delivery systems to improve their prediction in real time. In that framework, the capex is not expenditure in pursuit of uncertain future growth. It is the maintenance cost of a competitive advantage that is already producing measurable results and that depreciates if investment is stopped.
There is a frequent argument in the analysis of advertising platforms that holds that Meta is excessively dependent on the ads business compared to competitors that have more diversified revenue streams, such as Amazon Web Services or the cloud businesses of Microsoft and Alphabet. The argument has substance, but it ignores an important asymmetry: Meta holds an asset of human behavioural data with a depth and scale that no cloud infrastructure platform can replicate. That asset, correctly exploited by higher-quality models, is not a structural vulnerability. It is the reason the advertiser keeps paying more every quarter.
The 41% margin is not the most important number of this quarter
The indicator that best summarises Meta's competitive position at this moment is not the net profit or the revenue growth. It is the simultaneous combination of growth in impression volume with growth in price per impression, while the advertiser's conversion rate is also improving.
When those three variables move together, it means the platform is delivering more value to the buyer, that the buyer is perceiving it, and that they are paying more to receive it. That is the structure of an advertising business with genuine pricing power — not pricing power derived from a monopoly position or from the absence of alternatives. The advertiser who uses Meta's generative AI tools and records 3% more conversion is not paying more because they have nowhere else to go. They are paying more because the product they are buying gives them more back than it cost before.
The commercial architecture revealed by the first-quarter results is, in its most technical terms, that of a business that has found the mechanism for the improvement of its technological infrastructure to translate directly into improved outcomes for its client and, by that route, into an increase in the selling price. That cycle, when it functions with the consistency that this quarter's numbers show, is difficult to interrupt from the outside and even more difficult to imitate without the same years of data and the same scale of user base. The $145 billion in annual capex is not a bet on the future. It is the cost of keeping that cycle turning.
