Enterprises struggle to pick trusted AI models

Arena CEO Anastasios Angelopoulos said enterprises are unsure where to build their AI future as they weigh reliance on frontier labs against Chinese open-weight alternatives. He said companies are worried that providers such as OpenAI and Anthropic can impose safeguards or cut off access if they decide a use case is unsafe, making it difficult for businesses to depend on a stable AI stack.

Arena has a close view of the market through its crowdsourced model comparison platform, where users pit generative AI systems against each other and vote on the stronger response. The company launched a commercial service to sell battle data to AI companies and enterprises, and in June said it reached $100 million in annualized run-rate revenue within 8 months of launching its enterprise service.

Chinese open model companies such as Moonshot and DeepSeek offer lower-cost models that can run locally, but security concerns and potential U.S. restrictions complicate adoption. Palantir CEO Alex Karp has also warned that frontier labs could use customers’ intellectual property and data to compete with them, while Nvidia CEO Jensen Huang has argued that businesses will use both closed and open models depending on cost, customization, and IP needs.