News: 1743431531

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

LLM providers on the cusp of an 'extinction' phase as capex realities bite

(2025/03/31)


Gartner says the market for large language model (LLM) providers is on the cusp of an extinction phase as it grapples with the capital-intensive costs of building products in a competitive market.

Adoption right now is the most important thing: speed, adoption, being in the market – there is extinction coming. The market will not be able to support this number of model providers that we currently have

As the global tech research company forecasts worldwide generative AI (GenAI) spending will reach $644 billion in 2025, up around 76 percent from 2024, John-David Lovelock, Distinguished VP Analyst at Gartner, said the market for model providers would see a similar consolidation to that of the cloud market, which is currently dominated by AWS, Microsoft Azure, and Google Cloud.

"Cloud was more transformational in one sense, [it] changed how we did and what we did and where we did it," Lovelock said. "But it was a replacement. I had to change from using my on-premises CRM system to move to a cloud-based system like Salesforce. GenAI comes in and adds to whatever it is you're running, and it will be part of every mobile phone, every PC, every laptop, every server, every piece of software, it will be in your car, your TV, and your watch."

Meanwhile, CIOs are set to opt for commercial-off-the-shelf solutions for more predictable implementation and business value, rather than building their own software around AI, he said.

"Despite model improvements, CIOs will reduce [proof of concept] and self-development efforts, focusing instead on GenAI features from existing software providers," said Lovelock.

[1]

As a result, LLM developers are currently chasing adoption among users, businesses, and software vendors as a higher priority than revenue or costs.

[2]You know that generative AI browser assistant extension is probably beaming everything to the cloud, right?

[3]AI agents swarm Microsoft Security Copilot

[4]A closer look at Dynamo, Nvidia's 'operating system' for AI inference

[5]Cloudflare builds an AI to lead AI scraper bots into a horrible maze of junk content

"Adoption right now is the most important thing: speed, adoption, being in the market – there is extinction coming. The market will not be able to support this number of model providers that we currently have. From the capital expenditure requirements through to revenue, the cloud market only bears three and you expect the same with GenAI model developers," he said.

However, the market was unlikely to see a rapid collapse along the lines of the dotcom crash, partly because that also came on the back of Y2K spending. "We're not going to wake up one morning and find 20 companies out of business. It will be a slower pruning."

[6]

AI services are set for the most rapid growth in the market, Gartner said, climbing from $10.6 billion in 2024 to $27.8 billion in 2025, a 163 percent increase. While devices and software-based GenAI were both set to nearly double, the server-based market for GenAI would see the slowest growth at 330 percent to reach $180 billion. ®

Get our [7]Tech Resources



[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z-q8LEBn7zjH6q00VzHObwAAA4w&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[2] https://www.theregister.com/2025/03/25/generative_ai_browser_extensions_privacy/

[3] https://www.theregister.com/2025/03/24/microsoft_security_copilot_agents/

[4] https://www.theregister.com/2025/03/23/nvidia_dynamo/

[5] https://www.theregister.com/2025/03/21/cloudflare_ai_labyrinth/

[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z-q8LEBn7zjH6q00VzHObwAAA4w&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[7] https://whitepapers.theregister.com/



Lee D

Told ya.

The big AI companies aren't profitable - OpenAI said even their top tier isn't profitable.

We're coasting on that "stupid investors just throw money at a company doing nothing in the hope it'll become the world leader" phase and it's about to tank. It's the time when they realise that they've given this company enough money to operate without profit for over 100 years, and now say they want something back for their money... NOW.

And when you consider the equipment, power, time and expertise to train even a basic model, it all becomes quite worthless when it can barely replace a basic Google search, but with a much higher margin of error (I'm still wondering why Google ever bothered to put AI search results at the top of its pages, because more often than not it's just nonsense and wrong).

Reality is about to hit, and these things will become like a speech recognition model... niche little pieces of software that only just work well enough to be vaguely useful, but people realise that you don't need them in EVERYTHING and that they just aren't accurate.

Anonymous Coward

This may all be true, but the likelihood is you'll be paying for the sunk costs anyway through price rises on products and services that are in demand.

Yup

codejunky

The market working as the market should.

Excellent

Rich 2

“…grapples with the capital-intensive costs of building products in a competitive market.”

Don’t you mean…

“… grapples with the capital-intensive costs of building products with no discernible useful value.”

Either way, it will be nice to have less of this crap constantly in the news

Cloud was more transformational

that one in the corner

Because all it did was rename a way of working that we'd long since demonstrated works (buying time on a sever), sprayed on go-faster stripes and upsold it like profits were going out of fashion.

The LLMs: not following any proven path, in fact selling THE Neural Net application that we hadn't demonstrated is actually useful or could be put into service with understood cost/benefits for both true & false responses.

Our POP server was kidnapped by a weasel.