Meta’s AI Bet: Put Powerful Models in People’s Hands
Zuckerberg is betting on open models, personal agents and AI beyond the cloud as Meta’s competitive strategy against rivals in the U.S. — and China
Photo: Jonathan Raa / Sipa USA via Reuters Connect
Meta has released a scaled-down version of its most powerful open AI model, Muse Glimmer, designed to run on consumer hardware rather than exclusively in the cloud. The move puts increasingly capable AI closer to the individual user, while advancing Meta’s broader push for personal and open AI.
The 30-billion-parameter model is derived from the much larger Muse Spark and is designed to run locally on consumer hardware rather than exclusively in the cloud. Muse Glimmer is aimed at agentic tasks including coding and tool use, and can run on a single consumer GPU.
Mark Zuckerberg laid out that vision in a lengthy essay published Monday, The Future is for Everyone. He argues that advanced AI should be distributed as widely as possible rather than concentrated among a small number of companies or governments.
“We should distribute it widely and give every person the ability to direct it,” Zuckerberg wrote.
His vision goes beyond chatbots. Zuckerberg describes personal AI agents that could manage tasks, teach, help create businesses, assist scientific research and eventually act on behalf of their users around the clock. He argues that AI could give individuals capabilities that previously required large companies, teams or significant amounts of capital.
That vision also has an important economic dimension.
Today’s dominant AI model is centralized: enormous models run in data centers, and users pay companies to access them. Local models shift part of that equation. If increasingly capable AI can run on a laptop, workstation or other personal device, users and developers can deploy it without sending every request to a cloud provider.
That won't replace massive AI data centers. Frontier models will continue to require enormous amounts of compute. But it creates another layer of the market: AI that is software you have, rather than simply a service you access.
For Meta, that is also a competitive strategy. Open models can become infrastructure for developers, allowing Meta to build an ecosystem around its technology rather than competing only on access to a chatbot. The more widely its models are used, the harder it becomes for competitors to establish a closed ecosystem around their own AI.
China is an important part of that competition. Chinese companies have become increasingly influential in the open-model ecosystem, with models such as DeepSeek and Alibaba’s Qwen gaining global adoption. Zuckerberg explicitly identifies China as a strategic competitor, arguing that the U.S. has an advantage in semiconductor design but is slower to build the energy and physical infrastructure needed for AI.
For Zuckerberg, open AI is therefore also a geopolitical issue. He argues that the U.S. should lead the global open-source ecosystem rather than restrict access to foreign open models. The goal, he says, should be for American models to win by being better.
There is an obvious tension in that argument. Meta is hardly a neutral champion of decentralization. The company has enormous commercial incentives to prevent AI from being dominated by OpenAI, Anthropic or other centralized providers. Making Meta's models widely available can help ensure that Meta remains one of the companies defining the AI ecosystem.
That doesn't make the strategy any less significant. Meta is betting that the AI race will not be won solely by whoever builds the most powerful model. It may also be won by whoever gets powerful models onto the most devices, into the most applications and into the hands of the most developers.
Muse Glimmer is a concrete expression of that bet: instead of keeping increasingly capable AI behind a cloud interface, Meta is putting a version of it where users can actually run it themselves.