The current outlook for generative AI will evolve to deliver practical, valuable use cases and models with implications for future data center growth.

“Take nothing on its looks; take everything on evidence. There's no better rule,” Great Expectations, Charles Dickens.

Charles Dickens was a great documenter, and campaigner, for societal change. Given the societal shifts promised (and expected) from the AI era, he’s arguably just as valid a guide for our times as any of the more recent crop of science-fiction writers.

Dickens’ seminal novel Great Expectations is particularly relevant. It explores a range of themes but one of the central conceits is the importance of telling illusions from reality.

The central character Pip believes a wealthy aristocrat is his benefactor but finds out his real sponsor is a criminal. The illusion of proper lineage and wealth is undone by a harsher reality. Spoiler: it all works out in the end.

Great expectations for AI

Recent research from analyst firm Gartner has some similar lessons to impart about the ‘great expectations’ around generative AI – albeit with a more prosperous outcome all around.

The positive headline news from Gartner is that worldwide IT spending is set to grow 9.8% in 2026 and exceed $6 trillion for the first time. This growth will include more than $489 billion spent on data center systems in 2025, increasing to a projected $582 billion in 2026.

However, despite this growth, the analyst has expressed some caution around AI - specifically the expectations versus the reality. Referencing Gartners’ well-known Hype Cycle framework, distinguished VP analyst at Gartner John-David Lovelock recently described how generative AI is sliding from the ‘peak of inflated expectations’ in 2024 and into the ‘trough of disillusionment’ in 2025 and 2026.

“This year we are on the slide towards the trough of disillusionment. We won’t make it into the trough until 2026 but all of that proof of concept work we did and moonshot projects of 2024 have been abandoned this year,” said Lovelock. “As expectations for generative AI decline, the size and scope of projects associated with generative AI also decline.”

The trough of disillusionment

However, while expectations may be declining, more value is beginning to be extracted from generative AI according to Lovelock with enterprises increasingly “seeing what can be done rather than what might be done”.

Counterintuitively, a lowering of expectations around generative AI will result in continued spending and greater value, according to Gartner. “The interesting thing for me about this market is as we are declinine in expectations and heading towards the trough; as people are expecting less of AI, the people that are making the products and services keep making them better every day,” said Lovelock. “The ability of AI is going up and the expectations are coming down.”

“The ability of AI is going up and the expectations are coming down ”

Spending on AI will continue to be strong despite the recalibration in expectations according to Gartner. “We are going from a market that was $649 billion in 2023 to one that will be $3.2 trillion in 2029. It is not slowing down because expectations are slowing down,” said Lovelock. “This year’s growth is very much about the foundations. It is about service providers and software companies building out the functionality that CIOs will use. It is about building out the massive data centers that these AI optimized servers sit in to provide both training and inferencing capacity.”

Looking more specifically into how the maturation of generative AI will impact data center investment over the coming years, Gartner points towards the increased segmentation of different model types. There will be a shift away from specialized models, created by service providers, to in-house domain-specific language models (DSLMs) created and owned by software providers but also enterprise companies.

“We are still looking at 100% growth rates next year. But the difference between large language models and domain specific models starts to become very evident in 2027 and 2028,” said Lovelock. “The ability to create DSLMs will be fairly common and ubiquitous and will become the preferred approach for software companies and enterprises.”

Domain-specific language models

Gartner did not spell out how this AI model segmentation will translate into specific data center build-outs, but it seems likely that a proportion of the DSLMs will require dedicated, owned AI infrastructure by enterprises including on-premise facilities and colocation. This shift from mostly hyperscale spending on AI factories to enterprise-owned AI infrastructure will mark a maturation in the AI adoption cycle. Spending, risk, and value will be spread across a much wider group of organizations.

Guidance for enterprises as they move along their AI adoption journey is similarly maturing. From practical resources to reference designs.

Overall, it seems that generative AI is on a similar journey to Great Expectations’ Pip: from relative obscurity to great expectations to eventual stability and sustainable prosperity. Indeed, Gartner predicts that gen-AI will reach the ‘plateau of productivity’ by 2027/28. However, just like in Dickens’ novel, it seems there will be some adventures along the way before those great expectations are realized one way or the other.