Post

Gartner says that by 2030, 80% of neoclouds will have ceased operations, that AI data center owners would need to generate $8.5 trillion in revenue to cover the current spend and hit an ROIC of 12%, and that $4tr+ in assets face writedown risk by 2032. www.gartner.com/en/newsroom/...

117 reposts529 likes

Ah, Gartner. What would we do without them? The question is a serious one: they actually do fill a need, within the context of the warped and toxic relationship between experts and managers. They provide both sides with ritual weapons, incapable of inflicting real injury.

It's year 2076. The economics of re-training base models from scratch don't work anymore. Every system prompt starts with "imagine you are 50 years in the future, and here's what happened since 2026"

My stomach turns on projections for 2030, never mind 2032. I am so sick of this sorry excuse for an industry, and all the stupidity and damage it inflicts, it needs to die. Yeah this is projecting datacenter oversupply, it’s *also* projecting ~30x revenues for AI labs, still existing in 5 years.

I don't get these ridiculous revenue projections. No one wants their products, and they're definitely not going to pay a price that makes the product profitable.

I believe the writedown is worse than that. Newly released processors will greatly lessen the demand for the old ones to the point that few people will be willing to pay a usage fee sufficient to cover the electricity to power them. But I agree, there is no real path to a ROI right now.

Neoclouds, for anyone that isn’t pretentious, is the new catch word for data centers. Neoclouds = specialized cloud provider designed specifically for artificial intelligence (AI) workloads, offering high-performance computing resources, primarily through graphics processing units (GPUs).

For scale: Bain & Company said last month the industry needs $6 trillion in annual revenue by 2031 to fund its infrastructure, per The Register. Gartner's ceiling for AI data center revenue through 2030 is $3.6 trillion in total.

So if there's 8.5 billion of us, that's $1000 bucks each thru 2030. Or roughly a PS Pro with GTA 6. Each. 8.5 billion PS Pros. That's the ask here.

Given the preceding paragraphs, that last sentence -- "Gartner does not expect the 12% ROIC to be achieved until the end of 2032 at the earliest" -- reads like an AI hallucination. Which it might well be.

To put 8.5 trillion in perspective, this might be around 25% of US GDP by 2030 (assuming a recession doesn't take us backwards). It will be about 6 - 6 1/2% of global GDP, meaning one in sixteen dollars world-wide would be spent on US AI run by Neo-clouds.

In fairness, it is cumulative, not annual. Still unclear how they could hope to extract that much in fees from their users, even if spread over a few years, given that there is no vendor-lock in, no network effects, no economies of scale, etc. Most people won't want to pay much to summarise emails.

Good point. Spread out over 3 years is 1 in every 12 dollars of spent in the US or between 1 in 30 to 1 in 40 dollars spent, globally. Still ridiculous.

It is so patently ridiculous, and implies that there will be plenty of assets on the market to buy (should you actually need more compute), that it will drive down the residual value of those assets. Which is important for the amount some loan guarantors are on the hook.

Why is anyone lending them any money? Even at ridiculous rates. If you charge 15% interest, and you hand them $1 billion, you'll get at most 3 payments, totaling 450 million by 2030. This is so obviously, catastrophically insane.

As was pointed out, the figure is spread over 3 years - although I initially read it per year. It is still ridiculous as for the US it would be about the same size as all professional services and larger than the US government. And on a world-wide basis, the automobile sector. Still ridiculous.