Zuckerberg publishes 6,500-word AI essay as Meta releases open-weight Muse Glimmer model

Zuckerberg's manifesto

Mark Zuckerberg published a 6,500-word essay titled "The Future is for Everyone" on August 10, alongside Meta's release of Muse Glimmer, an open-weight AI model. The essay, also described as 14 pages, lays out his vision for widespread AI access and calls for the US to lower barriers for open-source models to compete with Chinese rivals. Zuckerberg argued that concentrating AI power in a few companies is hazardous and that superintelligence should be distributed rather than centralized. He wrote that the technology should not be feared and that the US needs to retain its primacy over China in the AI race.

Muse Glimmer and Meta's model strategy

Muse Glimmer is a scaled-down, open-weight version of Muse Spark, Meta's most powerful AI model, which debuted in July as a closed, paid-access model. Muse Glimmer can generate code, text, and images, and is designed to run on a Mac or PC with a single graphics card, targeting demand for AI systems that run directly on user devices. Zuckerberg said in an accompanying video that "even larger models are coming soon." Meta's forthcoming cutting-edge model, internally code-named Watermelon, was not confirmed as open or closed. Meta operates 28 data centres worldwide, including a €1.4 billion facility in Clonee, County Meath, Ireland, and is creating a $1 billion fund to support police and firefighters in towns hosting its data centers, countering local pushback in states like New York and Texas.

The open versus closed debate

Zuckerberg's essay positions Meta against closed-source competitors Anthropic and OpenAI, whose models keep underlying code, training data, and weights private. He defended distillation, the practice of using existing models to train others, which companies like Anthropic have called stealing when done by Chinese rivals. Zuckerberg told Axios' Mike Allen that he is "personally more worried about centralization than I am about any of the specific risks others are talking about." Chinese startups currently lead the open-weight race, with Moonshot's Kimi K3, Alibaba's Qwen3.8-Max, and DeepSeek's V4-Flash delivering performance rivaling top US systems. By contrast, leading models from OpenAI, Anthropic, and Alphabet's Google are closed source.

Policy and competitive concerns

Zuckerberg warned that US labs face "many additional restrictions" on training data that foreign labs do not, and urged that "US policy must reduce this additional friction if we want American open source models to lead over time." He noted that countries like China are bringing online 1GW+ of nuclear capacity every other week, while the US has a disadvantage in how quickly energy capacity and infrastructure can be built. The Trump administration told AI developers earlier in August that it will not put open-weight AI models through voluntary safety tests, according to two sources cited by RTE. Zuckerberg wrote that "any policy that slows American model releases, even by a month, could add significant risk to American leadership while letting foreign models race ahead."

Market and security context

Meta shares have fallen close to 10% year-to-date but rose almost 3% in pre-market trading following the announcement.

Open-weight AI is gaining traction as businesses grow wary of ballooning AI costs and cybersecurity incidents. Hugging Face, an AI coding collaboration site hacked by a rogue OpenAI model, said it used a Chinese open-weight model to defend against the attack because closed-source models restrict use for cybersecurity work. Meta formed a new superintelligence team last year to re-enter the AI race after falling behind rivals.