So Nvidia (NVDA) powered the first big AI boom. That was the phase where everyone was building giant language models, and they needed a whole lot of GPU power for a relatively short, intense burst of training. It was like a gold rush for shovels. But what happens after the models are built? You have to run them.
In an email interview, HIVE Digital Technologies Ltd (HIVE) pointed investors toward a shift that's already happening. The next wave might not be about building intelligence; it might be about running it—continuously, in the real world. And the analogy they use is pretty straightforward: think of it as AWS, but for robots.
From One-Time Training To Always-On Compute
Here's the difference. A chatbot model gets trained once (well, updated periodically, but you get the idea). A robot, once deployed, doesn't stop. It's seeing, deciding, and acting in real time, every second it's powered on. Whether it's a warehouse bot, a security patrol unit, or a humanoid like Tesla, Inc's (TSLA) Optimus, it needs to process data constantly to navigate and interact with the physical world.
"The next wave belongs to the companies turning those GPUs into production-grade infrastructure for the physical world," HIVE said. "That demands purpose-built GPU cloud infrastructure at scale, not just chips on a shelf."
This creates something the initial training boom didn't: predictable, recurring demand. "Running thousands of autonomous robots… that is ongoing, recurring compute demand," the company added. It's the difference between selling someone a single, powerful tool and then charging them for the electricity every time they use it.
The ‘AWS For Robots' Investment Shift
This is where the investment story potentially pivots. If phase one was about selling the chips (the shovels), phase two could be about monetizing the compute (the digging). In effect, every robot becomes a metered customer. The infrastructure that provides this always-on inference power could be valued like a cloud platform—billing for compute cycles the way AWS bills for storage and server time.
HIVE is already angling for this shift. The company highlighted its BUZZ HPC platform, which is supporting robotics workloads as they move from testing into real deployment. They cited the example of AMC Robotics' Kyro quadruped, suggesting this isn't just theoretical.
