Enterprises Stuck Between pricey frontier AI and risky Chinese open-source models, says Arena CEO

In a recent interview, Arena chief Anastasios Angelopoulos warned that businesses are caught between two uneasy options: expensive frontier AI services that can be abruptly disabled, and affordable Chinese open-source models that raise security doubts. He cited examples like OpenAI, Anthropic, Moonshot and DeepSeek, noting that the former may impose safeguards or shut down access without notice, while the latter can be run locally at lower cost but may be subject to governmental restrictions.

Arena, which hosts a “battle royale” platform for comparing generative models, announced a commercial data-selling service and reported a $100 million annualized run-rate within eight months. Angelopoulos also revealed that some model vendors have leveraged Arena’s internal data to generate synthetic labeling tasks, a practice he said is common. Parallel concerns were voiced by Palantir CEO Alex Karp, who warned that frontier labs could appropriate enterprise data, and Nvidia CEO Jensen Huang, who advocated a mixed approach using both closed and open models. The discussion highlights the broader industry dilemma of building a stable, trustworthy AI stack.