The hard part was never the chips

Most of the AI-infrastructure conversation is about compute, as if the chips were the easy part.

They're not.

First, you don't just buy the GPUs.

Allocation runs on relationships and trust as much as money.

NVIDIA backs the partners it believes can deliver, and some of them it backs with its own capital.

Getting to the front of that queue is a relationship you build long before the purchase order.

Then, once you have them, the hard part is everything around them:

power, cooling, cabling, network, commissioning, and coordinating procurement, logistics, data-centre operations and the customer so it all lands together, on more than one site, at once.

That's the last mile, and it's where builds slip.

Not because the kit didn't turn up, but because the seams between the parties had no owner.

A career of delivering data centres and networks taught me the same thing, every time:

the technology is rarely the risk.

The relationships and the coordination are.