Reflection’s Beam model signals deepening split in global AI market

Division in the AI market

“We’re seeing a more bifurcated world, where if it’s Chinese, then we always have to have a counter or alternative toward the Chinese option,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.

This division is evident not only in new U.S. models but also in models from other parts of the West. For example, on Oct. 6, French AI lab Mistral previewed Mistral Large 4, a new 1-trillion-parameter multimodal model. Mistral said it will release the model weights by the end of the month but is still testing the model with cybersecurity leaders and vetted partners. The French vendor claimed on Tuesday that ML4 ranks among the top five models globally and leads open-weight models developed outside of China.

“What all Western model makers are currently endeavoring to show is that ‘we’re actually able to create open weight models that are useful,’” said Bradley Shimmin, an analyst at Futurum Group.

The momentum behind open-weight and fully open source AI technology comes as enterprises look to gain insight into what their models are doing, how they’re working and be able to retrain models for their business -- all at lower cost than and with comparable performance to proprietary closed models from U.S. AI vendors such as Anthropic, Google and OpenAI.

“What companies are really pushing for is a more adaptive landscape of models that can go to the frontier when they need to,” Shimmin said. “They can turn to localized models that they have finetuned to their exact needs.

Support from Nvidia

However, Reflection’s value is not just as an alternative to China. What is more interesting is who is backing the vendor, Su said. AI hardware giant Nvidia -- the most prominent open source AI advocate in the West -- has supported the vendor with an $800 million equity investment and access to its GPUs to power its AI development and infrastructure.

The relationship with Nvidia is important as many Chinese vendors are trying to depend less on Nvidia and more on Chinese chipmakers such as Huawei. However, Chinese AI chips still have a long way to go, Su said.

“A lot of the Chinese vendors, eventually, will have to accept the fact that … Chinese chipsets are a bit slower, so they have to adjust them,” he said. “Western vendors don't really have that concern.”

The backing also provides support, while also providing the Nvidia platform with another model aside from Nemotron, Nvidia’s open model family. Shimmin said.

In addition to backing from Nvidia, Reflection has a partnership with SpaceXAI. The Elon Musk-owned company supports the vendor with massive compute infrastructure at its Colossus data center and access to Nvidia GB300 AI chips. The deal costs Reflection $150 million per month through 2029.

Training data and options for enterprises

Another notable aspect of Beam is the trillion tokens it was trained on.

“It is amazing just how quickly we as an industry have been able to scale some of these training methodologies and to show that what matters most is the quality of portraying the data, not the uniqueness of it, the quality of it,” Shimmin said.

Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.