News and Views by the Fintech Times

There are several genuinely non-obvious, specific claims - particularly around rack density thresholds, the shift from training to inference, and the case against sub-10MW edge data centers - but the episode is badly undermined by the guest repeating the rack-density/liquid-cooling argument almost verbatim twice, and the early section on cloud basics is entirely elementary.

“NVIDIA's GB300s, which are becoming commonplace now, require 150 kilowatts a rack”

“some major data center operators are telling me that in that pipeline 70 favor of infancy, some cases 80-20 in terms of the requirements and RFIs, RFPs that they're now seeing”

A few counterintuitive positions break through - pushing back on the 'AI inferencing isn't latency sensitive' consensus and dismissing the edge-datacenter narrative - but much of the Middle East hub discussion and data sovereignty framing is well-circulated, and the 'AI minus BS equals software' borrowed quip is doing a lot of originality work it doesn't fully earn.

“AI minus BS equals software”

“I used to have this debate with some of the investors and they would say, well, no, actually, that kind of AI thing is not latency sensitive. People are willing to wait. Well, not really, actually”

Milan Radia is a credible practitioner - an investment banker with a real track record in data center deals and clearly current on infrastructure specifics like GPU rack specs and hyperscaler procurement patterns - but he is an advisor/financier rather than an operator who has actually built or run these assets at scale.

“I had a battle on my hands as an investment banker trying to persuade investors to look at data centers”

“I remember working for UBS, one of the last financial trading companies or banks to move to the third party”

The episode is meaningfully anchored in concrete numbers - GB300 rack power draw, 1MW rack targets, 98% of IP traffic over subsea cables, 70-80% inferencing share in RFP pipelines - and named entities (NVIDIA, Meta, IHC, UBS), which lifts it well above average, though several claims about regional demand and obsolescence risk are asserted without supporting data.

“NVIDIA's GB300s, which are becoming commonplace now, require 150 kilowatts a rack”

“NVIDIA is already now talking about one megawatt racks and high voltage DC”

The host asks a few substantive questions - particularly on obsolescence risk - but effectively asks the same question twice without noticing, allows the guest to repeat his rack-density argument word-for-word without flagging the repetition, and never meaningfully pushes back on any claim; the format is closer to a friendly briefing than an interview.

“Do we know we're building the right things? And is there a chance that in like a year's time, we're going to have to rip some of these things down and build them again?”

“What's to stop people building data centers now that you might have to rip them down in years time if things take”