| Meta and Nvidia |
Meta and Nvidia enter the open-weighted AI race
After major US technology companies met last month to urge policymakers not to impose early restrictions on open-weighted AI models , even when those models are Chinese, two of the signatories of that call have begun to take practical steps to compete with Chinese labs by launching their own open models.
Meta and Nvidia this week released AI models that developers can download for free via the open-source ecosystem, a move that differs from the approach taken by popular, closed models such as those offered by OpenAI and Anthropic.
Open-source artificial intelligence has become a controversial issue stretching from Silicon Valley to Washington, with critics warning of the potential risks that Chinese models may pose to national security, along with the drip-feed technology used to train and improve AI models, which some see as potentially amounting to a form of intellectual property theft.
In contrast, most major companies in the sector believe that imposing broad restrictions on the use of these models could harm American interests themselves, and give a limited number of companies too much influence over the artificial intelligence market, according to a report published by CNBC and reviewed by Al Arabiya Business.
In an open letter published on July 24, a group of technology companies said that the age of artificial intelligence could be an age of prosperity, and that open-weighted AI could expand opportunities, enhance competition, support American technological leadership, reduce risks, and ensure that the benefits of this technology reach different sectors of the economy.
As for the distillation technique, the signatories described it as "a widely used technique for improving, evaluating, and verifying the performance of models."
Meta returns to betting on open models
At the beginning of the week, Meta launched the Muse Glimmer model, as part of a new strategy aimed at making its most powerful models available to the open-source community.The company's CEO, Mark Zuckerberg, said that Meta will open up the weights of its latest model, the Muse Spark 1.2.
"Weights" refers to the calculations and rules that determine how an AI model works, responds, and behaves.
A day later, Nvidia unveiled the Nemotron 3.5 Lightning model, part of the Nemotron 3 family that the company launched last December.
Nvidia said its models are effectively open source because it publishes training datasets and methodologies used in developing the models, along with their weights, allowing developers to examine and study them.
The real competition comes from China.
But Meta and Nvidia still need to prove there is real demand for their models in a market dominated by strong models from Chinese AI labs, such as Moonshot and DeepSec, along with Alibaba’s Qwen model.However, Aaron Levy, CEO of Box and one of the signatories of last month's letter, is clearly optimistic.
Levi believes that Zuckerberg's plan to release the Muse Spark 1.2 weights is of paramount importance, given that the model is robust enough to compete with leading foundational models from Anthropic and OpenAI.
The models launched this week by Meta and Nvidia are smaller and designed to work on laptops to perform tasks such as running artificial intelligence agents that work directly on the device.
Levi said the move represents "a very clear signal that the United States will have open-source models approaching the frontiers of advanced capabilities."
"Meta" recreates the Llama experience
This is not the first time that Meta has bet on this path.
The Llama family was Zuckerberg’s first entry point into the basic AI model market, but the launch of Llama 4 in April 2025 did not receive the expected reception from developers.Meta responded by investing billions of dollars to restructure its artificial intelligence unit, and appointed Alexander Wang, CEO of Scale AI, to lead the division.
Recently, the Wang Group has begun introducing closed-loop models under the Muse brand, in an attempt to create new revenue streams for the company.
A huge opportunity for American companies
Levi believes that companies hesitant to use open-weighted AI models coming from China will be more willing to try models developed by Meta.He explained that large government institutions, for example, may not be able to use an open-source model from outside the United States, and that large banks may face similar restrictions.
But with American models like Muse, the opportunities for use can expand significantly, especially for organizations that need to control technology and data.
Meta faces a trust problem
However, Meta faces a significant challenge in regaining the trust of developers.Umesh Sachdev, CEO of AI for business company Uniphore, said that Meta burned bridges with third-party developers when it moved from an open-weights strategy to closed models.
He added that it would take more than a long article written by Zuckerberg to convince developers to return, noting that some developers in his company feel that Meta has betrayed their trust.
0 Comments