Zuckerberg unveils $145B AI plan with open-weight models, $1B fund

Meta is betting $145 billion this year that open-weight AI models and a $1 billion community fund can win the trust regulators and investors have withheld.

Meta CEO Mark Zuckerberg published a 6,500-word essay arguing the biggest risk in AI is centralized control, as the company plans $145 billion in capital spending this year and $600 billion by 2028 to build out data centers and superintelligence infrastructure. The manifesto, released Monday, pairs a utopian vision of AI distributed to every individual with concrete policy proposals and a fresh wave of open-weight model releases.

"One of the ways to achieve both individual empowerment and checks and balances is to make sure you put the technology in everybody's hands," Zuckerberg told Axios' Mike Allen ahead of the release. "It's definitely a different view than what you're hearing from a lot of other people in the tech industry right now."

Meta is launching Muse Glimmer, an open-weight model the company bills as one of the highest-performing of its size, and will open the weights for a version of its foundational Muse Spark 1.2 model in coming weeks. The company also seeded a $1 billion Future is for Everyone Fund to support U.S. communities where it owns and operates data centers, with money earmarked for teachers, first responders, and energy and water infrastructure. Meta's independent board will gain power to approve model-release safety criteria, and the company pledged a "community compact" covering local jobs, school investment, and commitments on energy prices and water use.

The push comes as Meta trails Anthropic and OpenAI on frontier models while investors grow impatient with the returns on its AI buildout. Shares fell from $700 to $583 after the July earnings call, when Meta raised its 2026 capex ceiling to $145 billion from an earlier $60 billion to $65 billion range without detailing how the infrastructure would be monetized.

Open weights as a counterweight to OpenAI and Anthropic

Zuckerberg's essay frames open-weight distribution as the antidote to the concentration of power he says would follow if rival labs — which build AI for companies, governments, and institutions — set the industry's direction. "If those labs lead, then the balance of power will favor larger institutions over individuals," he wrote. "Meta's mission since our founding has focused on putting power in people's hands."

The open-weight commitment marks a significant shift in the debate over whether advanced AI is too dangerous to release publicly. Meta paused open-source releases in recent quarters while building out Meta Superintelligence Labs, which unveiled its first model in the first quarter. The company says it will resume releasing open-weight models now that the lab is running, with the independent board reviewing whether each release meets safety criteria.

Meta's advertising business, meanwhile, is already showing measurable AI returns. First-quarter revenue rose 33 percent year over year to $56.31 billion, with ad impressions up 19 percent and price per ad up 12 percent — a dual increase the company attributes to a 6 percent-plus improvement in ad conversion rates from its Lattice and GEM AI models. Operating income reached $22.87 billion at a 40.6 percent margin, and second-quarter revenue guidance came in at $58 billion to $61 billion.

The $145 billion question: monetizing the buildout

The capex plan — which includes a $27 billion five-year GPU compute deal with Nebius and a $14.3 billion stake in Scale AI — has drawn skepticism from investors who want to see the infrastructure generate returns beyond advertising. Zuckerberg has signaled a cloud computing business is "definitely on the table," and Bloomberg reported Meta is forming a unit to sell excess AI compute and hosted models to enterprise customers, putting it in direct competition with Amazon Web Services, Microsoft Azure, and Google Cloud.

Meta spent $31.1 billion on capital outlays in the second quarter alone and told investors it is "supply-constrained," prioritizing near-term compute capacity for model training and AI agents. The company issued $25 billion in bonds to help fund the buildout.

The bet is that AI-driven ad conversion gains compound for years, making this year's outlay look cheap in hindsight. Meta trades at roughly 25 times forward earnings, and the stock's slide from its $700 high suggests the market is pricing in a higher required return than the company's "prove-it year" rhetoric supports. Zuckerberg's essay offers a broader answer: distribute the technology widely, win over the communities hosting the data centers, and let open models carry the company's credibility into the next phase of the AI race.

This article is for informational purposes only and does not constitute investment advice.