With Statsig, OpenAI buys control over the learning cycle
OpenAI has agreed to acquire product testing startup Statsig for $1.1 billion in stock—one of its largest deals to date. Founded in 2021, Statsig helps developers run experiments and feature flags, powering companies like Eventbrite and SoundCloud. Now, its tools will sit at the core of OpenAI’s strategy.
Statsig’s founder, Vijaye Raji, will become OpenAI’s CTO of Applications , reporting directly to Fidji Simo. That move signals something deeper: OpenAI sees experimentation not as a feature, but as critical infrastructure.
Why? In deterministic software, A/B tests help improve features. But in probabilistic AI systems, experimentation decides whether products even survive. Each interaction with users feeds a preference loop , showing what works and what doesn’t. That loop becomes the real moat—competitors can copy weights, but not user data at scale.
Feature flags, in this world, act like financial hedges: emergency shut-offs, guardrails against drift, and ways to measure what actually happened versus what might have.
The bigger picture: OpenAI isn’t just buying Statsig’s tools—it’s buying control over the learning cycle that defines what “good” AI looks like. Whoever owns that cycle sets the benchmark for the entire industry.
OpenAI’s recent spree of large acquisitions (including Jony Ive’s AI device startup) shows it’s consolidating the layers of AI infrastructure: from hardware and applications to now experimentation. If training data was the moat of the last AI wave, OpenAI is betting that experimentation infrastructure will define the next one.
