Databricks CEO: We're hosting open source models like Kimi and running out of GPUs

Summary

Databricks CEO Ali Ghodsi discusses the company's massive $188B funding round, driven by accelerating revenue and unprecedented demand for its AI gateway that supports open-source and proprietary models. He highlights that enterprises are embracing open-source models like Kimi to control soaring AI costs, leading to a global GPU shortage that required the fundraise. The interview underscores the rapid enterprise shift toward AI agents and the growing tension between cost-efficient open-source models and expensive frontier labs.

• Databricks is fundraising at a $188B valuation after 6-7 quarters of revenue acceleration across all regions and product lines.
• Enterprise AI agents are driving consumption-based revenue for Databricks as they query its data platform.
• The company's AI gateway, which provides capacity for OpenAI, Anthropic, Gemini, Grok, and open-source models, has seen explosive demand.
• Customers want to use cheap open-source models for mundane tasks and reserve expensive frontier models for complex queries, aiming to control exponentially rising AI costs.
• Databricks hosts Chinese open-source model Kimi and is running out of GPUs in Asia, Japan, Korea, the US, and India.
• The GPU shortage and need to scale capacity for open-source hosting triggered the fundraise.
• Ghodsi remains positive on frontier labs but sees a structural shift toward open-source for cost control.