The biggest AI infrastructure opportunity may not be in building AI models. It could be in running them. DigitalOcean Holdings, Inc. (NYSE:DOCN) CEO Paddy Srinivasan told Benzinga in an exclusive email interview that the company’s inference services grew roughly 800% year over year, calling the pace “breathtaking.”
Inference is what happens after an AI model has been trained: It is the computing power used when that model actually answers a question, writes code, searches for information or completes a task for a user.
AI’s Next Growth Engine May Be Inference
Srinivasan said the growth is being driven by the rise of AI agents alongside applications in “personal productivity, Searches, Ad Technology, Generative Media, Coding” and other real-world use cases.
“We are starting to see mainstream adoption (via Inference) in many real world use cases,” he said.
He also pointed to DigitalOcean customer OpenCode, a coding agent, as an example of how the rise of agents is contributing to inference growth.
The economics are helping fuel that adoption. Srinivasan said AI is benefiting from rapidly improving “intelligence per dollar” — essentially, how much useful AI output customers get for what they spend.
That could create a powerful feedback loop: cheaper, more capable models make more AI applications economically viable, while greater usage creates more demand for the infrastructure running those models.
The distinction matters for investors. The first AI infrastructure boom was dominated by companies building enormous systems to train increasingly powerful models. The next could increasingly be driven by what happens after training — when those models are actually put to work.
Srinivasan’s 800% figure is DigitalOcean’s growth, not a measure of the broader inference market. But if AI agents and cheaper models continue driving usage higher, running AI could become one of the next major infrastructure businesses.
Photo: Shutterstock
