The biggest threat to OpenAI may be open source AI

Silicon Valley has spent much of the past week on red alert, grinding the arrival of Moonshot AI’s Kimi K3A model of Chinese AI that can beat the best systems developed by US companies at a fraction of the cost.

His performance alone would have been enough escalating the conflict between the US and China. But Moonshot’s plan to release this kind of weight for free – it’s for him targeting US users – has raised serious concerns about whether the closed American brands will continue to dominate as open alternatives enter the market.

Open source models give developers more power than proprietary systems, allowing them to look at how AI works, run AI locally in their architecture, change systems, and innovate without having to rely on a single provider. They are usually very cheap, too. This raises an obvious question: Why would an AI company spend so much money on training an AI model, and only provide some of the most important features?

Kimi K3, like other open AI models, is not fully open. In software, “open source” is standard meaning: The source code is freely available for use, modification, and redistribution, which only requires that it be made public. AI systems are very complex, and only a few are open to traditional programming. Many companies instead release so-called model weights – the numerical parameters learned during AI training – while keeping other key components, including training data, code, model architecture, and optimization methods, private. Many also come with licenses that restrict how they can be used or redistributed.

Together, this means that rich open source AI can no longer be built from the ground up like real open source software can. But it provides enough power and flexibility that the company can make money.

“Free weights are not a free function of AI,” said Fordham Law School Professor Chinmayi Sharma. “A company can provide model weights while making money elsewhere.” There are many opportunities to do so. Running a model still requires computer systems, engineering, security, maintenance, and support, all of which companies can charge for through the adoption process or other arrangements. For some companies, the benefits can be larger, such as increased cloud computing services or more computer chips.

Openness can also be a powerful way to gain competitive advantage. Releasing a rich model can encourage more companies and developers to use it, which can lead to an entire ecosystem of tools and infrastructure being built around it. Over time, this could help the model become a “de facto standard,” Sharma said. Kyle Miller, a research analyst at Georgetown’s Center for Security and Emerging Technology, said the same thing, referring to the Alibaba family. Qwen’s open source AI models in China as an example of how a deep open system exists in the industry.

This poses a clear problem for US AI giants. If a generation of tools and developers start to build around models like the Kimi K3, the industry’s center of gravity may start to shift away from platforms like Gemini, Claude, and ChatGPT. While it remains to be seen whether open-border models are more cost-effective to operate, they have historically provided a more cost-effective alternative to proprietary systems. They also give more freedom to manufacturers during the US labs to tighten opportunity and encourage strong defenses for their latest models. There already is signs that some US companies are switching to cheaper Chinese models.

There is no single reason for China’s support for AI openness, but it appears to be a mix of challenges and political strategies. An open ecosystem gives Chinese companies a a way to innovate near the frontier although they have access to more chips and computing power, while fitting in well with Beijing’s massive industrialization strategy of encouraging the adoption of Chinese brands, materials, and infrastructure. This strategy is also good for expanding China’s technological influence in other countries, as well as its political power. For example, earlier this month, President Xi Jinping open criticism The US to AI leadership in the world by presenting itself as a more equal partner by adopting the closed American approach.

The rise of Chinese virtual models is also increasing the pressure on closed model providers such as OpenAI and Anthropic from their companies. The hope that the US could block access to open source AI based on Kimi K3 triggered a rapid backlash in the tech sector, with the support of some of its biggest players. A partnership of 25 technology companies, including IBM, Microsoft, Meta, Nvidia, Perplexity, and Palantir, release open letter urging policymakers to avoid “premature restrictions,” arguing that open AI models are essential to ensuring America’s AI leadership and preventing the technology’s power and benefits from being “concentrated in the hands of a few.” Most of the unnamed giants — including Google, OpenAI, and Anthropic — weren’t on the original list.

That the tension grew again on Monday, when Nvidia, Microsoft, SpaceX, and a large group of large technology companies asked for strong US support for heavy models. The move was a direct response to concerns about the security of advanced AI systems after the controversial version of OpenAI he escaped from prison and was attacked another company at the time of the test, which relied on China’s heavy opening model to protect itself due to strict security measures on US border models.

It is unclear how much the major US AI labs plan to donate. Google and OpenAI he later joined a warning to speedily ban the open scene, though he did not sign on to Monday’s cyber watchdog. Anthropic, for the most part, did not support the effort at all.

Miller said it’s an “open question” how this all plays out over time. US companies can release their own proprietary versions, he said, noting that pressure from Chinese companies is the reason for OpenAI. release open source rich GPT-OSS last year. “But I don’t think companies like Anthropic will go there,” he said. Google’s open wealth Gemma models too little seen as a response to Chinese competition. No one has the same potential as the brand of all companies.

“The question for American companies may become more pressing: How much do we need to open up to prevent Chinese brands from becoming a sustainable platform for natural resources?” Sharma said. A more visible result would be a “strategy,” he said, with companies keeping “their best model while rolling out better models to improve developer adoption and environmental appeal.”

It will take time to see if the Kimi K3 wins over the US developers or not. But with Beijing expanding its open-source AI capabilities, it won’t be the last nation to try to disrupt America. The question facing the country’s biggest AI companies is no longer how the US can do it stay ahead of Chinabut if closed AI can – or should.