Analysis
Mark Zuckerberg told Alexandr Wang he was writing a check for 'like $100 billion or $200 billion' when he recruited the 29-year-old Scale AI cofounder to lead Meta's AI effort in 2025, according to Fortune. More than a year later, that off-the-cuff number has turned into real guidance:
- 2026 AI capex -- $130B-$145B.
- 2025 capex -- $72.2B.
That's nearly double year over year, and the first flagship model from the lab Wang runs, Muse Wonder, shipped this week.
“on September 8 and topped Apple's App Store free chart, pulling in more than 2.5 million downloads in about two weeks, per Sensor Tower data cited by CNN.”
How Wang got the job
Wang is not a computer scientist and had never trained a frontier model before Meta brought him in. What he had was Scale AI, the data-labeling company he cofounded, and Zuckerberg paid $14.3 billion for a 49% stake in it specifically to bring Wang aboard as Meta's chief AI officer. Zuckerberg has said he was 'not happy with the trajectory we were on with Llama 4,' whose lukewarm 2025 reception prompted the creation of Meta Superintelligence Labs (MSL) under Wang. Pulse previously covered Muse Wonder's predecessor model, back when MSL still lacked a model that could credibly compete with OpenAI or Anthropic.
The competitive gap Muse Wonder is meant to close
On Artificial Analysis's Intelligence Index, Muse Wonder Full now sits roughly level with top OpenAI and Anthropic models, a couple of points behind the leader, per AI Weekly's measurements cited by Fortune. MSL's core research team remains small by industry standards -- Wang told Fortune it's 'closer to 100' people, even though he's publicly cited a figure closer to 200 -- backed by larger infrastructure and product teams inside a company of roughly 75,000 employees. Muse, the consumer-facing personal AI agent built on top of this work, launched in the U.S. on September 8 and topped Apple's App Store free chart, pulling in more than 2.5 million downloads in about two weeks, per Sensor Tower data cited by CNN.
Meta's capex jump isn't happening in isolation:
- Alphabet -- guided to $195B-$205B in 2026 AI capex this same week, more than double the $91.4B it spent in 2025, according to Fortune.
- Big Tech combined -- Amazon, Alphabet, Microsoft and Meta are on pace for more than $720B in 2026 capex, per Bloomberg-compiled analyst estimates.
- Private credit -- Blue Owl, Apollo and KKR have extended more than $200B in loans to AI-related companies on top of that.
Meta's bet is enormous in absolute terms, but it's now the norm rather than the outlier among its direct AI-model competitors.
Not everyone is convinced the spending matches the output. Wang himself has acknowledged the downside, relaying CFO Susan Li's warning that the bet is binary: good outcomes are 'all good,' but a bad outcome is 'really, really, really bad' -- a rare on-record admission from inside a company usually allergic to public doubt about its own capex. The risk is structural as much as financial: a sub-200-person core research team is being asked to keep pace with labs several times its size, and Muse Wonder's benchmark parity doesn't yet tell founders or investors whether Meta's AI products convert into the kind of recurring revenue that would make the capex self-funding rather than a bet on faith.
What happens next matters more than the launch itself. Muse Wonder's retention numbers over the next two quarters, not its Intelligence Index score, will show whether Wang's small team can out-execute labs with far more headcount -- and whether Zuckerberg's 'like $100 billion or $200 billion' aside turns out to be the cheapest line in Meta's 2026 shareholder letter or the one that gets revisited hardest.
