Meta shares gain as Muse Glimmer launch highlights broader AI ambitions

Meta (NASDAQ:META) shares rose 2.4% on Monday following the launch of Muse Glimmer, a compact artificial intelligence model with 30 billion parameters that is designed to operate on a single GPU.

The model is positioned as a smaller, distilled version of the more powerful Muse Spark 1.2 and has been developed primarily for “agent-like” tasks such as managing schedules, organising files and rapidly creating prototypes.

Meta has made the model weights freely available through Hugging Face, strengthening its commitment to open AI development. The strategy places the company in more direct competition with Chinese developers including DeepSeek and Alibaba, while differentiating its approach from major U.S. technology groups such as Amazon, Alphabet and Microsoft, which have concentrated heavily on enterprise and government AI applications.

The release coincided with a broader vision outlined by Chief Executive Mark Zuckerberg, who argued for a transition away from highly centralised artificial intelligence towards what he described as “Personal Superintelligence.”

Meta proposes changes to AI safety governance

One of the most significant elements of Zuckerberg’s strategy involves changing how Meta oversees the safety of its artificial intelligence models.

Rather than leaving final authority solely with the company’s founder and chief executive, Meta intends to give independent directors responsibility for approving safety standards governing model releases and reviewing whether those requirements have been followed.

Zuckerberg also advocated similar governance structures across other frontier AI developers, arguing that decisions with potentially broad societal consequences should not rest exclusively with individual chief executives.

Meta outlines new approach to government AI cooperation

Meta is also proposing a different framework for working with the U.S. government on national security concerns surrounding advanced AI.

Instead of delaying public model launches while government agencies conduct reviews, the company would provide authorities with access to intermediate training checkpoints and relevant technical personnel before development is complete.

The proposed system is intended to give government agencies additional time to identify vulnerabilities and strengthen critical infrastructure without substantially delaying public access to new models.

Meta Superintelligence Labs to restart open-source releases

Zuckerberg said the recently established “Meta Superintelligence Labs” is now operational and confirmed that Meta will “resume releasing some open source models soon.”

The company also continues to support model distillation, a technique in which knowledge from a more powerful AI system is used to train a smaller and more efficient model.

Zuckerberg views the practice as important for maintaining U.S. competitiveness, particularly as Chinese developers expand their presence in open-weight artificial intelligence.

Meta plans cloud computing business with dynamic pricing

Alongside its commitment to making AI widely accessible, Meta is preparing a cloud infrastructure offering aimed at users requiring substantially greater computing capacity.

The company plans to introduce a dynamic auction system for allocating computing resources. Pricing would respond to available capacity and aggregate demand, with the goal of providing users with competitively priced access to AI infrastructure.

The initiative could give Meta another route to monetise its enormous investment in computing capacity while continuing to provide basic AI services freely to a much broader audience.

Private AI agents could adopt WhatsApp-style protections

Privacy is another major component of Meta’s emerging personal AI strategy.

The company is developing a “fully private mode” for AI agents that would apply privacy protections similar in principle to those associated with WhatsApp’s end-to-end encryption.

Under the proposed approach, interactions, tasks and information processed by a personal AI agent would be protected so that Meta itself could not access the underlying content.

The feature could become an important differentiator as consumers increasingly use AI systems to handle sensitive personal and professional activities.

Meta targets local support for massive data centre expansion

Meta is also introducing community programmes designed to address the infrastructure and local economic challenges associated with constructing increasingly large AI data centres across the United States.

Its Future Is For Everyone Fund would provide direct financial support to communities hosting infrastructure projects, including initiatives such as $50,000 teacher bonuses in Richland Parish, Louisiana, where Meta is developing a data centre.

The America’s Workforce Academy is intended to provide free skills training and pathways into well-paid jobs for electricians, carpenters and other trades needed for AI infrastructure construction.

Meta has also outlined ambitious water and energy commitments, including a target of becoming 200% water-positive in highly stressed regions by 2030. The company is investing in its own energy infrastructure and says some projects could also provide surplus lower-cost electricity to surrounding grids.

Zuckerberg argues wider AI access could provide checks and balances

At the centre of Zuckerberg’s argument is a broader theory about how advanced artificial intelligence should be distributed.

Rather than treating restricted access as the primary route to AI safety, he argues that widespread availability of advanced systems could create its own balance of power.

Under this vision, individuals and smaller organisations could eventually have access to highly capable lawyers, cybersecurity tools, business advisers and other specialised AI agents.

Zuckerberg believes distributing those capabilities broadly would reduce the risk of artificial intelligence becoming concentrated in the hands of a small number of governments or corporations.

For investors, the Muse Glimmer release therefore represents more than another model launch. It forms part of a much broader Meta strategy spanning open-source AI, personal agents, cloud infrastructure, privacy, data centre investment and new governance structures as the company attempts to establish a distinctive position in the next stage of the AI market.

Meta stock price