When GPUs sing together: Solving the AI data centre power spike

AI data centres have a problem that no one talks about enough. When GPUs synchronise and fire together, power demand spikes exponentially, in milliseconds. Legacy backup systems were never built for this. Oversizing the grid is expensive. Throttling GPU utilisation wastes the very infrastructure operators are paying for.

Ramesh Narasimhan, EVP, CCO and CFO at Nyobolt, explains how the Cambridge deep-tech startup evolved from a university materials lab into a full solutions provider. And how its Dynamic Response System addresses the power spike problem at the rack level, in real time.

Ramesh discusses Nyobolt's journey from a vacuum cleaner prototype to a partnership with Symbotic's warehouse robots, the case for ultra-fast charging over high energy density, and why the company's core lesson applies as much to startups as to battery chemistry: don't confuse your product with your purpose.

#battery #fastcharging #AIDataCentre #robotics #IoT #WeTalkIoT

Summary of this week's episode:

From automotive to deep-tech: Ramesh's background

Nyobolt's origin: Cambridge materials lab to commercial company

What forced the shift from materials to solutions provider

The secret sauce: what Nyobolt's battery does differently

Charging a car in under five minutes, and why range obsession misses the point

The AI data centre power problem explained

The Dynamic Response System vs. conventional UPS

Off-grid in Rajasthan: what the MoU actually means

Symbotic and the warehouse robot use case

Humanoids, trade-offs, and scaling the supply chain