Video highlights

The hosts frame the central question as whether that spending will pay off, given that Meta makes most of its money from advertising and that model does not appear to be the AI plan.

He adds that Zuckerberg reportedly does not want to rent out computing power as a neocloud, as SpaceX has done, and instead wants to monetize intelligence built on Meta's own compute.

Operating income was $28 billion in 2019 and has risen to $106 billion over the past 12 months, roughly four times higher, on $222 billion of advertising revenue.

Travis Hoium says monetizing attention through advertising, a model TV and radio have used for decades, has made Meta extremely profitable and let it outrun rivals like Snap.

The rest of the company's decade has been far less successful, most visibly in Reality Labs, which posted about $19 billion in operating losses over the past year.

Jon Quast says AI has so far been a net positive for the advertising business, helping optimize and target existing inventory.

Ad pricing rose 12% in the latest quarter, with an even larger uptick in the US.

Longer term, Muse is not monetized through ads and may draw attention away from Facebook and Instagram. If the agentic world takes off, it could cut into the advertising thesis.

Travis Hoium says large tech companies such as Alphabet and Microsoft are long past the scale-building phase, and Meta now focuses on extracting more attention and more dollars per unit of attention.

Meta generated about $130 billion in operating cash flow over the past year.

The company is now spending more on AI buildout capex, and in the third quarter of 2026 it is expected to surpass operating expenditures on capex, excluding off-balance-sheet items.

Travis questions whether making the existing business a little better is worth burning cash flow in the meantime.

Jon Quast says he started using Muse, then stopped and disconnected all his connections because it began doing things he had not asked it to do.

Muse alerted him to an online dinner order he had not requested, and he says other users have reported similar overreach.

Travis Hoium says Meta has created a product far more compelling than he thought it could, partly by rethinking the product from the ground up and hiring talent that drove the model and product design.

He adds that Meta is collecting data that could help monetize the rest of the business and lay the groundwork for advertising.

Jon Quast says he is willing to give Meta the benefit of the doubt that this will work as a long-term investment for shareholders.

He argues execution risk is much higher than in the past, because capex is now an order of magnitude above what the company spent a decade ago.

During the metaverse push, Meta was able to pull back and start paying a dividend, which was well rewarded and still financially manageable. A failed AI bet could hurt shareholders far more.

Travis Hoium says he is not buying Meta today, mainly because the company is entering a valley of financial performance without a rental business to fall back on, unlike Alphabet and Microsoft.

He explains that Alphabet and Microsoft can hold back on building their own AI tools because they earn money renting compute, which he says has slowed Alphabet's product progress while probably helping its finances.

Jon Quast says that if forced to choose between buying and selling, he chooses to buy, judging the opportunity and history of surprises worth the bet within a diversified portfolio.