Companies don’t know which side to choose in AI, the leader of one of the largest model evaluators said.
“It’s not only true that they’re terrified of working with the frontier labs, but they’re also terrified of working with the Chinese open source,” Arena CEO Anastasios Angelopoulos said during a recent episode of the “20VC” podcast.
Angelopoulos said the companies he talks to are struggling to figure out where to build their AI future.
“I think we’re in a really tricky situation for enterprises because just the way that the cards have landed,” Angelopoulos told Business Insider during an interview where he expanded on his comments.
The current situation is, by all accounts, a bit messy. Arena has an up-close view. The company offers a crowdsourced AI model battle royale that lets users pit generative AI models against each other and vote on which one best completes the task. In September, it announced a commercial business where it would sell AI companies and businesses data gathered from the battles. In June, Arena announced it had reached $100 million in annualized run-rate revenue within 8 months of launching its enterprise service.
Frontier AI companies, including OpenAI and Anthropic, offer cutting-edge models, but that technology can come with a steep price tag. The Trump administration’s recent actions also illustrate that the model providers can be forced to shut off access without warning. Those companies, as Anthropic has demonstrated, can also impose safeguards on the models, which may mean that many companies cannot access the full extent of the model’s capabilities.
“Anthropic or OpenAI could decide to turn off their AI at any moment if they feel that it is, quote unquote, ‘unsafe,’” Angelopoulos told Business Insider. “And so if you’re a business, what are you supposed to do? You want to be able to build on a stable and reliable AI stack.”
In contrast, China-based open model companies, including Moonshot and DeepSeek, offer open-weight models that can be run locally, often at a fraction of the cost of frontier models. Some are worried about the security risks of relying on Chinese companies. Last month, Axios reported that the Trump administration has considered imposing restrictions or even issuing de facto bans on Chinese models.
Angelopoulos isn’t the only one to hit on these themes. Palantir CEO Alex Karp has repeatedly warned that frontier AI companies, with which Palantir works, want to gather valuable intellectual property and data that they could then use to compete with their current customers.
“They deserve to colonize your enterprise,” Karp said during Palantir’s second-quarter earnings call on Monday, characterizing what he believed was the labs’ attitude.
Palantir and Karp also have some incentive to stoke those fears, as the company offers its customers an application layer that allows businesses to control their data and the AI models they use.
Angelopoulos said Arena has experienced its own struggles with AI companies. He said that Arena has seen model providers use Arena’s internal data “to serve synthetic battles and arenas to our employees in an effort to essentially get them to label data.” Angelopoulos declined to name the offending companies, but said the practice “is not uncommon.”
Nvidia CEO Jensen Huang, whose company does business with frontier model makers and offers open models, has tried to strike a balance in the debate. He has said that companies will deploy both open models and proprietary ones, routing tasks based on both cost effectiveness and maintaining their IP.
“The world is going to have closed and open models,” Huang recently told Axios. “We should use closed models as often as we can, and we should customize and use open models to build our own personal, our own customized AI if we must.”