A little-known American AI startup is looking to steal frontier AI companies’ bread and butter by letting enterprise customers build their own custom AI models.
OpenAI’s and Anthropic’s business models largely revolve around selling AI to business customers, via APIs and enterprise subscriptions. OpenAI CFO Sarah Friar reportedly told investors in an August closed-door meeting that while the startup’s revenue had historically been approximately split 60/40 in favor of individual customers, that ratio had recently reversed: “The majority of our revenue is now enterprise,” Friar said, according to CNBC. In February, Anthropic said it had more than 500 business customers who spent more than $1 million annually on API usage or Claude subscriptions. A leaked IPO prospectus revealed last month that one-quarter of Anthropic’s revenue last year came from just two customers (both unnamed, but almost certainly businesses).
Several powerful open-source and open-weight models developed by Chinese AI labs have been a tempting alternative, since they can be modified to suit each customer’s needs and, more importantly, are cheaper than American frontier models. ChatGPT and Claude, in contrast, come with a confusing jumble of pricing plans that can be difficult to track. The era of so-called “token-maxxing,” when businesses encouraged employees to use AI tools as much as possible in their day-to-day work, has yielded to sober frugality. But Chinese models bring problems of their own: most major U.S. companies are understandably hesitant to hand over sensitive customer information to an AI developer with direct ties to the Chinese Communist Party.
Reflection AI wants to fill that gap.
Launched in 2024 by a pair of former Google DeepMind researchers, the company is working to build what it calls an “AI factory”: software that lets businesses build their own AI models, negating the need to rely on subscriptions to closed systems like ChatGPT or risky Chinese alternatives. “Today the best open models are coming out of China,” the startup’s cofounder and CEO Misha Laskin told CNBC in July. “And as a result, global AI adoption is actually very much being built on top of Chinese models, including [by] American companies. And that’s great, but we wish to build a thriving AI ecosystem that has a lot of competition and a lot of players, and we’re building the counterbalance to that—building great open models here in America.”
According to a new report from Axios, Reflection is also now preparing to launch its own open-weight AI model to compete directly with the biggest frontier labs on both sides of the Pacific. While the report was scant on specifics, it said the model could be expected soon and that it will have capabilities on par with the most advanced Chinese open-weight models. Reflection’s debut model will “be joined by other open-weight models from other Western players this month,” Axios added in its report. Reflection didn’t immediately respond to Gizmodo’s request for comment about the new model.
Echoing Nvidia’s Jensen Huang and Meta’s Mark Zuckerberg, Reflection believes that a completely open ecosystem is the safest and most equitable pathway into the future: “We need to build open models so capable that they become the obvious choice for users and developers worldwide, ensuring the foundation of intelligence remains open and accessible rather than controlled by a few,” the company wrote in a blog post last year.
Nvidia had invested about $800 million in Reflection as of March of this year, according to the Wall Street Journal. Reflection also signed a multiyear partnership with SpaceX in June, through which it agreed to pay Elon Musk’s aerospace company $150 million per month for compute generated by the Colossus data center facility in Tennessee.
Given how little is known at this point about Reflection’s forthcoming model, it’s too soon to say if it will be capable enough to convince businesses en masse to ditch ChatGPT or Claude. No matter how it pans out, the growing excitement around open models is likely to force both OpenAI and Anthropic to focus less on building new and more powerful AI systems and more on making their existing products more affordable and accessible—especially as they both inch closer to their stock market debuts.
