This Data Centre Investor Says Most Hyperscaler Builds Won't Get Finished

I spoke to a man who invests in Data Centres.

He says most hyperscaler builds won't get finished...

His reasoning:

These projects take years to build, but the chips inside them move much faster than that.

By the time a site goes live, the hardware it was designed around is already old news. And every new GPU release makes the last one worth less.

- B200/B300 GPUs launched near $8/hour
- Now they're trading closer to $4
- Next-gen Vera Rubin chips rumoured at $10/hour
- Each new release resets the clock on older models

That's what caught my attention.

Most people in this space talk about power and land.

He was the first to walk me through the chips themselves as the real variable. This makes DCs very hard to underwrite given the speed of technology.

Once we got past that, he laid out exactly what he looks for in a deal:

- Wants power secured and live first
- Won't discuss tech stack until power's confirmed
- Takes a 25-30% deposit on total contract value
- That deposit covers build cost upfront

Are chips, power or buildings the biggest risk?