(Axios) – Reflection, an AI startup backed by Nvidia, is preparing to release an open-weight model that could add a new competitor to an industry dominated by major U.S. AI companies and increasingly capable Chinese developers.

The company’s approach could give businesses another option for obtaining advanced AI technology without relying entirely on expensive closed systems operated by companies such as Anthropic, OpenAI and Google.

Open-weight technology is attracting significant attention because users can generally download, adapt and deploy these models themselves. That flexibility can make them more difficult to control, but advocates argue that openness and transparency can also provide security and innovation benefits.

People familiar with Reflection’s plans say its model will initially trail the most sophisticated American systems but should be competitive with leading Chinese open-weight models. The company believes the technology will nevertheless be powerful enough for organizations to develop customized AI applications using their own data.

For certain tasks, combining an open model with specialized company information can produce performance comparable to much more expensive AI systems, particularly when the model is customized for a specific purpose.

Reflection did not comment on the planned release.

The company is pursuing a broader strategy beyond simply releasing an AI model. Its long-term objective includes an “AI factory,” in which organizations could combine their private data, AI software and computing resources to create systems designed specifically for their own operations.

That approach could appeal to businesses that do not want sensitive information processed through outside AI services. Financial firms and other organizations that closely protect proprietary data are among the potential users.

Reflection has already started exploring the concept through a sovereign AI factory arrangement with South Korea’s Shinsegae Group.

The startup has also been working to secure enough computing power for its expansion. Sources say Reflection recently reached agreements with Nebius and SpaceX involving access to Nvidia AI servers.

Before the launch, company representatives have reportedly been speaking with interested groups in Washington and elsewhere about the upcoming model and the AI factory concept.

Reflection CEO Misha Laskin has acknowledged that developing the most powerful AI systems requires considerable time.

“They’re kind of like rocket ships,” Laskin, who previously worked at Google, told CNBC. “To build a big rocket ship, it takes time.”

Nvidia CEO Jensen Huang has promoted a similar vision in which companies combine powerful Nvidia hardware with open AI technology while retaining control over their own information.

That strategy could be particularly important for Western organizations that are uncomfortable relying on sophisticated Chinese AI systems because of potential security concerns. Reflection is among the companies attempting to offer a Western alternative.

Nvidia has been supportive of Reflection’s effort to develop a major open AI system, according to reports cited in connection with the startup.

Reflection’s planned release could therefore mark an important development in the competition between open and closed AI. With other Western companies also preparing open-weight offerings, the next phase of the AI race could become considerably more competitive.