'GPU supply chain under stress, but not yet disrupted'

Neysa, a neocloud that builds purpose-built cloud infrastructure for AI workloads, raised $1.2 billion earlier this year in a Blackstone-led round with plans to scale up its GPU capacity to over 20,000 units across India. Karan Kirpalani, chief product officer, Neysa, tells Anees Hussain that the West Asia crisis and surging dynamic random access memory (DRAM) costs have already tightened delivery timelines and made OEM pricing volatile. However, the disruption has not yet reached a point where it is derailing the company’s build-out. Excerpts:

- 

You recently raised $1.2 billion structured as $600 million in equity and $600 million in debt. Where does that capital go?

It’s going to fuel three things — service development and R&D, capacity and data centre expansion, and go-to-market. A significant portion goes into capacity and inventory because this is a very capital-intensive business. But a significant amount is also going into R&D. We intend to be more than simple GPU rentals. We’re building aggressively on the platform and services side so we can offer a single-pane-of-glass service across training, fine-tuning and inference. You should see something significant from us in the April-June quarter in terms of breakthrough product capabilities.

What is the revenue split between bare-metal GPU rentals and value-added platform services?