America Makes a 180-Degree Turn, Finally Yields to China's Approach
Meta and Nvidia from the United States have officially launched artificial intelligence (AI) models that can be downloaded free of charge by developers through an open-source ecosystem. This strategy is similar to what China has been doing all along.
This move is also an effort by US technology companies to compete with open-source AI models from China such as DeepSeek, Moonshot AI, and Alibaba’s Qwen.
The launch took place only a few weeks after a number of US tech giants asked the government not to impose early restrictions on open-weight AI models, including those originating from China.
China-style open-source AI models stand in contrast to the proprietary AI models owned by OpenAI and Anthropic from the US that currently dominate the market.
In an open letter released on 24 July, a consortium of technology companies stated that open-weight AI can actually strengthen the competitiveness of the United States.
“The AI era could be an era of prosperity. With the right policy choices, open-weight AI can expand opportunity, strengthen competition, extend American technological leadership, reduce risk, and ensure the benefits of this remarkable technology are shared broadly across the economy,” they wrote.
Regarding the AI training technique known as distillation, they described the method as a widely used technique for improving, evaluating, and validating models.
The issue of open-source AI has indeed become a fierce debate in Silicon Valley and Washington. Critics argue that Chinese AI models could pose national security risks, while the practice of distillation is seen as potentially resembling intellectual property theft.
However, many industry players argue that restricting the use of AI models would actually harm America and give too much power to a handful of companies.
Meta on Monday introduced Muse Glimmer as part of its strategy to open its most powerful AI models to the open-source community. Meta CEO Mark Zuckerberg said the company will open the weights of their latest AI model, Muse Spark 1.2.
Weights are parameters that determine how an AI model works and behaves.
A day later, Nvidia launched Nemotron 3.5 Lightning, which is part of the Nemotron 3 model family first introduced in December last year.
Nvidia emphasised that their model is truly open-source because the company also released the training dataset, training techniques, and model weights for developers to examine.
Nevertheless, Meta and Nvidia still have to prove that their models can attract users amid the popularity of Chinese AI models such as DeepSeek, Moonshot AI, and Qwen.
Box CEO Aaron Levie is optimistic that Meta’s move will be well received. According to him, Muse Spark 1.2 is a very powerful model capable of rivalling top-tier AI models from Anthropic and OpenAI.
The models released by Meta and Nvidia this week are smaller in size and designed to run on laptops, including to support digital agents that work directly on devices.
“There is a very strong commitment that America will have open-source models that approach the leading edge,” Levie said.