Only 28% of AI infrastructure projects fully pay off, survey finds

Tech leaders hoping AI might help save money and improve efficiency in IT infrastructure should know that only 28 percent of use cases fully succeed and offer return on investment (ROI). According to new figures from Gartner, one in five AI projects in IT infrastructure and operations (I&O) fail outright. Its survey of 782 I& …

[cyberdemon]

And that's BEFORE slop-mongers have hiked their prices to break-even or above.

[EricM]

While there definitely were some price-hikes in the last weeks, they are still far from break-even, let alone being able to pay interest on their debt to their investors.

So please stand by for more (and more egregious) price hikes in 3 - 2 - 1 ....

[Michael Strorm]

My understanding is that, in some cases, even some not-inexpensive paid services (e.g. coding tools) have running costs many times that actually charged.

The whole thing is a house of cards based on everyone undercharging everyone else and obfuscating the cost, hoping to make back massive profits later once the dust has settled, their rivals have gone under and everyone is used to them or locked in.

Problem is that the industry has already set the expectations of the public and businesses that such services are free or very cheap. People barely want many of them even when they're having them forced in their faces. And of the ones they do use- often for novelty fluff purposes- it's questionable whether they'd be willing to actually *pay* anything like the real price.

And even with lock-in, it's not given that businesses which pivoted to AI on the basis of current, unrealistically cheap prices will be able to pay several times that regardless.

I recently saw an article which compared the current AI bubble to the subprime mortgage crisis. The latter was built on unsustainably low lending rates over the short term that made people think they could afford something they couldn't, and the shit hit the fan when rates went up.

Note the similarity.

[David Hicklin]

It almost feels like the AI slop-mongers are desperately holding on for as long as possible by using whatever cross-investment deals they can until businesses have sacked a critical mass of workers that will allow them to jack up the AI rates - the businesses will be stuffed as without "AI" they will not have enough workers to continue.

So a bit like drug dealers where the first hits are cheap until you are hooked and totally dependant on them.

[Michael Strorm]

That may well be the calculation they're making.

But even if that works in the short term, it's open to question whether- if they jack up the prices as much as they'll have to to make a profit, let alone want to- it won't be a sufficient impetus to drive many customers to overcome that "lock in", however painful that might be, and move away from it in the long term. Especially when the much-hyped cost savings don't turn out to be as great as was expected in the first place (even at the original, subsidised prices, let alone the new ones).

Or whether a sufficient percentage of those that *can't* move away will be able to afford to continue paying the inflated rates. Even if the AI companies have them over a barrel, that's not much use if they go bankrupt.

This is all a problem because- given the ludicrous amounts of money that have been invested in AI- I'm assuming that short-term profit won't be enough, and that it will *have* to be a sustained, long-term success before it comes close to breaking even, let alone significantly profitable. And it's not clear that there are (or ever would have been) a sufficient number of people and businesses willing and able to pay the prices necessary for that to be the case.

[Yet Another Anonymous coward]

28%?

So that's way higher than non-AI IT infrastructure projects paying off

[Clausewitz4.1]

Re: 28%?

” So that's way higher than non-AI IT infrastructure projects paying off”

Working in telecom and ISP for quite a few decades, allow me to disagree. The “cloud” indeed took a chunk of customers, but then a lot realized it was cheaper to run things locally, and started to come back.

[Yet Another Anonymous coward]

Re: 28%?

Unencumbered by vibe coding, a system given the herculean task of paying 1M federal workers in a single federal government in a single country is now nearly $6Bn over budget and so far behind schedule that it is now paying the descendent of people it was supposed to pay.

[cyberdemon]

Re: 28%?

Oh so why don't you vibe code it 4 times over with a better than 1 in 4 chance of success?

The figure of 28% success rate, while extremely dubious in itself, obviously is not going to apply uniformly across all levels of project complexity. It can easily be skewed if a large number of "projects" in the sample are "Arduino" or "Static Website" projects, for example. So it's a pretty useless figure at face value.

Just like human projects, the more complex the project, the less likely that it can be solved in one step by bodging something together.

Large important projects require a high ratio of software engineering vs coding, and Engineering is something that AI sucks at miserably.

[JohnSheeran]

Execs will just tout this like it's the "1 in a million" line from Dumb and Dumber. "So you're saying there's still a chance?"

[DJV]

Yeah, but (as a great man once said) magicians have calculated that million-to-one chances crop up nine times out of ten.

[UCAP]

Have an upvoter for the PTerry reference.

[Helcat]

However, a 1 in 999,999 or 1 in 1,000,001 are guaranteed to fail.

What was the odds of Sgt Colon hitting the Dragon in the vulnerables while standing on one leg singing the Ank-Morpork anthem. With one eye shut as I recall... (and now need to go read the book again!)

[Richard 12]

Fortunately

The chance of them surviving what happened moments later were exactly one in a million.

[Pete 2]

Compared with what?

> only 28 percent of use cases fully succeed and offer return on investment (ROI)

On its own, that tells us nothing useful. For example, is that better or worse than non-AI use cases?

I am not sure I've seen any IT project that could claim to have "fully succeeded". Unless you use the barn door method of determining success

[Clausewitz4.1]

Re: Compared with what?

” For example, is that better or worse than non-AI use cases?”

Compared to “normal” datacenters running baremetal servers, BGP, MPLS, VPS, CDN, etc… I am used to see a 100% success rate every time a company opened a new datacenter.

[retiredFool]

Will bonuses be cut?

"98 percent said there was increasing pressure from the board to demonstrate ROI"

Will the board actually take some money away from CEO's for this failure? Me thinks not. CEO's get rewarded for success and rewarded for failure. Why, well because the company learned something due to the CEO leadership. Mere mortals get fired.

[sarusa]

Ahaha all lies

That number is far too large. They can't possibly have recovered all the money they've dumped down the AI toilet. They're lying out their ass, like Microslop does when they count every sale of Office and Windows 11 as 'AI revenue' because they've shoe-horned copilot into those, as if people only bought them for Microslop's worst in class AI.

I do know that in some businesses it actually works, like for telemarketing, ransomware, customer support, and consultancies it doesn't matter how much the AI is wrong and lies - because you don't give a fig about your customers, they're just fools to be milked. So I do believe some predatory businesses are actually in the black from going AI. But certainly not 28%.

[Anonymous Coward]

Re: Ahaha all lies ... [Sorry, don't you mean hallucinations !!! :=) ]

"That number is far too large."

Exactly my thought ... not to mention that you cannot compare all projects to each other as they will vary by scale & complexity, at least.

More useless stats that prove nothing other than the 'AI' Bros will grasp at any straw to try to push 'AI' as a solution to ANY knowledge area.

'AI' is a scam and NO-ONE has demonstrated it working at scale such that the investment has been worth the cost.

Not interested in pilots or scaled down tests, show me the sucessful projects that have not been more costly than getting conventional skills to implement & deliver the end result.

:)

[David Hicklin]

Re: Ahaha all lies

I am very surprised the article did not include the line " ..because they are not fully committing enough to it and should spend more money on AI"

[ecofeco]

Pressure from the board?

The very board who forced everyone to adopt this shite?

"Here, make this cow pie taste like fillet mignon or you're fired!"

Also:

"Sorry we have to fired people and no raises this year because, reasons! And raises only lead to inflation anyway, not our burning several billion on cow pies!"

[Blackjack]

I am surprised any of them pay off at all.

[Bebu sa Ware]

surprised any of them pay off at all.

"57 percent have suffered at least one failure in applying AI to their area."

The other 43 either lied or were too clueless to know, which leaves the fabulous outlier of clueless mendacity obscuring a success.

[Groo The Wanderer - A Canuck]

Of course they had unrealistic expectations; that is exactly what the "big vendors" like OpenAI and Anthropic keep on doing - misleading the public and the community about what Artificial Ignorance LLMs can actually do rather than buzzword bingo fantasies.

[Arkitekt]

That number seems astronomically high.

[B33Dub]

28% of bullshit

28% of bullshit is still 100% bullshit.

It's like the bullshit "white papers" they keep releasing that claim X% gain in some fantasy metric for the latest AI models.

Like these idiots seriously think they invented a ruler for "cognition" and "intelligence". They have numbers for intangible and metaphysical things. Like we haven't been trying and failing at that with human minds since antiquity. Then they CONFIDENTLY post this crap on socials like they're doing actual science. F off with that bullshit.