Meta is opening up access to some of its most powerful artificial intelligence models, as CEO Mark Zuckerberg made a case for a more open approach to AI amid growing investor scrutiny of the industry’s massive spending.
In an Instagram video posted on Aug. 10, Zuckerberg said Meta would release the weights for its latest AI model, Muse Spark 1.2.
Weights are learned settings that determine how an AI model processes information and generates responses. An open-weights model can be downloaded by developers and run on their private infrastructure.
Zuckerberg also said Meta would release a new family of open-weight models called Muse Glimmer, designed to run locally on laptops.
Much of today’s most advanced AI relies on cloud computing and expensive data centers.
Zuckerberg Calls for AI Policy Reform
In a lengthy blog post on Aug. 10, Zuckerberg argued that U.S. policies are putting domestic AI developers at a disadvantage to foreign competitors, particularly in the race to develop open models.
“Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data,” Zuckerberg wrote.
Specifically, he called for the United States and its allies to accelerate the construction of energy infrastructure and data centers to remain competitive with countries such as China, which he said can bring new power-generation capacity online more quickly.
Zuckerberg also asked U.S. policymakers to “rethink” their approach to distillation, a widely used technique in which the outputs of a bigger, more capable AI model are used to help train a smaller, less capable one.
Zuckerberg warned that the authorities should be careful not to restrict legitimate forms of distillation.
“Some have tried to frame distillation as harmful, but I think it is important to protect the principle that you can learn from anything you can observe,” he wrote. “This is how the world works, and the US will not be able to lead if we restrict ourselves on this front.”
Zuckerberg Pushes Back on AI Doom
Zuckerberg’s statement comes as some leading AI developers have struck a more cautious tone about the risks posed by increasingly powerful systems.
OpenAI recently said it suspended work on some aspects of its upcoming Astra after an internal review found that this model had reached a “critical cybersecurity threshold.”
Anthropic has also raised concerns about the potential misuse of its most sophisticated Claude Mythos model.
In the Aug. 10 post, Zuckerberg did not mention either company by name but pushed back against what he characterized as an overly pessimistic view of AI development.
“[I]t is surprising that the discourse from many developing AI is so filled with doom,” Zuckerberg wrote.
“While there are risks to releasing capable models, the most dangerous scenario from this perspective would be leading AI labs training powerful models and keeping them for themselves.”
Investors Question AI Spending
The comments also come as investors demand clearer answers about whether the hundreds of billions of dollars being poured into data centers, computing infrastructure, and AI model development will generate sufficient returns.
On July 29, Meta shares plunged hours after investors listened to the company’s second-quarter earnings call but received few new details from Zuckerberg about how Meta plans to generate revenue from enterprise AI tools or computing capacity.
According to Meta’s second-quarter results, free cash flow plunged by 91 percent to $784 million, down from $8.55 billion in the same period a year earlier.
The decline came as Meta spent $31.1 billion on capital expenditures, much of it tied to AI infrastructure. That spending consumed nearly all of the company’s $31.86 billion in cash generated during the quarter.
