Arena AI: Companies Confused by AI Model Choices

Companies Facing Uncertainty in AI Models

Anastasios Angelopoulos, head of Arena AI, recently expressed companies' concerns regarding the selection of artificial intelligence models. According to him, these companies find themselves in a complex situation, unsure of whom to trust among the numerous available models. Angelopoulos emphasized that this uncertainty is fueled by distrust towards cutting-edge laboratories as well as Chinese open-source models.

During a discussion, Angelopoulos described the situation as "really delicate." He explained that companies are hesitating between different options, not knowing which path to take in the field of AI. This dilemma is particularly acute for companies looking to build their technological future on solid foundations.

"It is not only true that they are afraid to work with cutting-edge labs, but they are also afraid to work with Chinese open-source," Angelopoulos stated during an episode of the "20VC" podcast.

Arena AI and the AI Model Market

Arena AI, a company specializing in evaluating AI models, has a unique perspective on this dynamic. It organizes competitions where different generative AI models compete, allowing users to vote for the best one. In September, Arena launched a business initiative aimed at selling the data collected during these competitions to companies. In just eight months, Arena achieved an impressive annualized revenue of $100 million.

Leading AI companies, such as OpenAI and Anthropic, offer highly advanced models, but often at high costs. Moreover, these companies may be forced to restrict access to their models without notice, as demonstrated by certain actions taken by the Trump administration. Additionally, security measures imposed by these companies can limit full access to the capabilities of the models.

"Anthropic or OpenAI could decide to disable their AI at any moment if they believe it is, I quote, 'dangerous'," explained Angelopoulos. "So, if you are a company, what should you do? You want to be able to build on a stable and reliable AI stack."

Chinese Open-Source Models: A Risky Alternative?

Meanwhile, Chinese companies like Moonshot and DeepSeek offer open-source models that can be run locally, often at a lower cost. However, reliance on these models raises security concerns. Last month, it was reported that the Trump administration was considering restricting or even banning the use of these Chinese models.

Angelopoulos is not alone in expressing these concerns. Alex Karp, CEO of Palantir, also warned against leading AI companies that might exploit their clients' data and intellectual property to compete against them.

"They deserve to colonize your company," Karp stated during a Palantir earnings call, describing the perceived attitude of the labs.

The Challenges Faced by Arena AI Against Supplier Practices

Arena AI itself has encountered difficulties with certain AI model suppliers. Angelopoulos revealed that some suppliers had used Arena's internal data to create simulations aimed at encouraging their employees to label data. Although he did not name these companies, he clarified that this practice is not uncommon.

Jensen Huang, CEO of Nvidia, who collaborates with creators of cutting-edge models and offers open models, has attempted to find a balance in this debate. He suggested that companies should use closed and open models depending on cost-effectiveness and the protection of their intellectual property.

"The world is going to have closed and open models," Huang recently stated. "We should use closed models as often as possible, and we should customize and use open models to build our own customized AI if necessary."