Meta: Nobody Uses AI Better

The thesis in one paragraph

Meta is quietly turning the best attention-capture engine ever built into a machine that does not just show you ads, but works out what you will buy next. If that lands, the business stops looking like a media company and starts looking like a toll on commerce itself.

"Nobody uses AI better than Meta"

That is Jensen Huang, talking to CNBC. His point was not a compliment about chatbots. It was about plumbing. Meta's recommendation engine has moved from a classical, CPU-bound recommender to a generative, agentic system that decides what you see, in what order, and what gets sold against it.

When the person selling the picks and shovels tells you who his best customer is, it's worth listening.

Just notice your own behaviour.

Think about the last time you picked up your phone to check one thing. You opened Instagram. Ten minutes later you surfaced, with no memory of why you reached for the phone in the first place.

That is not an accident. It is the product working exactly as designed.

The habit is old. Families used to gather around one TV from dinner until bedtime. The habit hasn't changed. The vector has. Instead of one screen beaming the same toilet-paper ad at a family of five, where the teenagers tune out and the toddler can't buy anything, there are now four screens (toddler excluded), each fed a separate, curated stream tuned to one person.

In Q1 2026, Meta's apps reached 3.56 billion daily active people. Time spent on Instagram Reels rose another 10% year on year in the same quarter. The funnel is still widening.

What that does to an ad budget

Picture Quilton, the Australian-owned toilet paper brand. Why keep throwing $100,000 at a TV campaign when you can't tell whether it reached a parent, a teenager, or a baby in nappies? A big chunk of that spend is spray-and-pray. The baby was never going to buy.

Now move the same budget to Meta, where the targeting is built around an actual profile of intent. Maybe $50,000 buys the same sales. Here is the part advertisers don't talk about... they almost never pocket the savings.

This is Jevons paradox at work. When something gets cheaper and more efficient, we don't use less of it. We use more. The advertiser sees the return, keeps the full $100,000 in the channel, and books double the sales. Efficiency doesn't shrink the market. It expands it.

Source: Meta Platforms Earnings Presentation - Q1 2026

The numbers already show the pull. In Q1 2026, ad impressions across the family of apps rose 19% and the average price per ad rose 12%, both year on year. More inventory and higher prices at the same time is the signature of demand outrunning supply.

The brain behind it is getting an upgrade

Now imagine that targeting engine running on a supercluster instead of a server rack. Bigger context, sharper inference, a system that doesn't just know roughly who you are but can call what you want next, sometimes before you can.

This is not speculation about the far future. Meta has already shown the conversion math. Improvements to its AI recommendation systems lifted ad conversions on Instagram by roughly 5% and on Facebook by around 3%, with online commerce the single largest contributor to ad revenue growth.

On the Q4 2025 call, Mark Zuckerberg put it plainly: "We're starting to see agents really work." He has described the edge as AI that understands context drawn from a person's history, interests, content and relationships. And he has pointed directly at agentic shopping tools that help people find the right product from a business inside Meta's catalogue.

That last point is the whole thesis. Read it again.

This is what the capex is for.

The spending is not subtle. Meta's capital expenditure ran to $72.2 billion in 2025. Guidance for 2026 has since been lifted to a range of $125 billion to $145 billion. The company is roughly doubling its infrastructure bill in a single year, and when it raised the range in April, the stock fell on the news.

For context, Meta did $200.97 billion of revenue in 2025, up 22%, with advertising still around 97% of the total. The ad machine is funding the compute machine. The bet is that the compute machine turns the ad machine into something bigger.

Where this ends up

Here is the way I think about it.

If I ran a cookie shop and you told me, "I can send you 100 people a day who I know, with near certainty, will buy from you, and in return I want 30% of the sale," I take that deal every time. At 50% I'm probably still in, depending on how long you can keep it running.

That is the position Meta is building toward. Not a billboard. Not a media channel. An operator that sits between intent and purchase and clips a ticket on the transaction.

If it gets there, the better comparison isn't a media company at all. It's a payments network. Something closer to Mastercard or Visa: a toll on activity that happens whether or not you think about the toll. The difference is that Meta would also be the thing that created the intent in the first place.

The other side of the trade

None of this is a sure thing, and the risks are not small.

The payback on $125 billion to $145 billion of annual capex is unproven. Depreciation from that build flows straight through the income statement for years, and Meta's own operating margin has already slipped from a 2024 peak near 48% toward 41% in Q1 2026. Reality Labs is still losing roughly $4 billion a quarter.

Regulation in the EU and US remains live, and the entire targeting advantage rests on a data and privacy regime that politicians can change. Agentic commerce is a roadmap item, not a revenue line yet. And a large, far-off opportunity is exactly the kind of catalyst that is easy to narrate and hard to underwrite.

In my eyes, Meta is an opportunity where the predictive nature of AI will be powerful, and a captured base of over 3 billion people (with strong network effects to protect it) makes it an appealing proposition at a time when other pockets of the 'AI trade' are fetching multiples of 60, 90, even 400 times earnings.