CPU demand is going to go bananas
And it's not just about AI agents
It's about physical AI
Welcome to the orchestration layer
Up until now, humanoid robots have depended on their onboard compute, or what you could call individualized intelligence
That's akin to a single chatbot
Just as we've moved from individual chatbots to swarms of AI agents, the next phase will be about large-scale robotic coordination
Imagine a factory where robots independently negotiate tasks, coordinate movements, share information and adapt production schedules
That creates demand for computing infrastructure beyond the processors installed inside individual robots
Now, you might think: we already have millions of warehouse robots, and they've barely made a dent in CPU demand
That's true, but that's because today's robots perform a very limited set of predefined tasks in structured environments, with no capacity to adapt
That's conventional fleet orchestration
The next wave of technology is AI-agent orchestration at the physical layer
In these next-generation systems, an AI receives a complex objective, develops a plan, delegates tasks to other robots, evaluates their results and modifies the plan as circumstances change, while each robot must individually adapt to its environment
That's millions of times more complex than what we have today, requiring inference capacity orders of magnitude higher
The result?
It wouldn't change the compute per robot, but it would require a CPU-heavy centralized brain to coordinate these deployments
顯示更多