Mark Zuckerberg Details Meta AI Strategy and $135B Capex Surge
Meta is earmarking between $115 billion and $135 billion in capital expenditure for 2026, nearly double the prior year's outlay, according to Crypto Briefing's read of the company's latest earnings call.
The figure, framed by CEO Mark Zuckerberg as the price of admission to "personal superintelligence," effectively triples down on AI while the metaverse narrative recedes into the rearview.
From open weights to closed stacks
The architectural shift matters more than the dollar volume. Meta spent years positioning itself as the open-source counterweight in the foundation model race, anchored by the Llama family. That thesis took a hit when Llama 4, released in 2025, landed with underwhelming reception. The April 2026 launch of Muse Spark — Meta's first major closed-source system, built specifically for agentic, multi-step workflows — signals the company's exit from that posture. Reinforcement followed: a $14.3 billion stake in Scale AI and the installation of its founder, Alexandr Wang, at the head of the newly formed Meta Superintelligence Labs.
For a Web3 audience, the optics are familiar. A platform that once championed open weights is now consolidating compute, talent, and model parameters behind a proprietary wall. The incentives align with vendor lock-in rather than ecosystem composability — the same gravitational pull that has historically dragged Web2 platforms toward centralized control.
Metaverse capex, redirected
The capital reallocation lands at the expense of the metaverse roadmap that defined Meta's identity since the 2021 rebrand. During the earnings call, Zuckerberg framed 2026 as "steady model releases and product integrations rather than one giant moonshot." Hardware momentum has migrated to Ray-Ban Meta smart glasses, whose sales tripled in 2025. Roughly four billion monthly active users across Meta's app family provide the distribution surface for AI-native features; immersive 3D worlds do not.
For Web3 gaming and open metaverse builders, the implication is structural. Meta is no longer competing for talent and compute in the same lane. That frees GPU supply and developer mindshare for decentralized alternatives — but only if those alternatives can demonstrate throughput and latency parity with closed stacks. So far, they have not.
What to verify next
The numbers will compress or expand the verdict. Three signals to track: the realized capex print across 2026 quarterly filings (the $115–135B range is guidance, not spend); whether Muse Spark benchmarks outperform peer agentic frameworks on multi-step task completion; and whether MSL ships a second closed-source model in the back half of the year. Zuckerberg's own framing — "I expect our first models will be good… but more importantly, we'll show the rapid trajectory that we're on" — concedes the first release is a baseline, not proof. The proof is the next one.