Generative AI enthusiasm continues to beat out business uncertainty

IT spending around the world is expected to grow even in the face of economic uncertainty, according to a new report by Gartner, which projects $5.43 trillion will be spent in the sector by the end of 2025.

The figure would represent a 7.9% increase on 2024 spending and is driven largely by growing investment in generative AI and associated infrastructure.

For example, data center systems growth is projected to hit 42.4% in 2025, as AI vendors and hyperscalers look to double down on the expansion of their AI-based products and services.

This compares to 10.5% growth across software and just 5.4% across devices.

John-David Lovelock, research VP and distinguished analyst at Gartner, told ITPro that the spending forecast shows investment is outpacing any slowdown caused by US tariffs, particularly as AI spending increases.

He acknowledged the “pause” linked to uncertainty, noting that the US is restructuring global trade, meaning some companies are pushing spending into Q3 and Q4 or avoiding expanding into new territories.

But Lovelock refuted claims that this spending pause could continue until uncertainty around the tariffs comes to an end, using the analogy of jumping into cold water. While initially one freezes up and floats, he said, once you get used to it, you start swimming.

“Same thing with the uncertainty: we have a new level of uncertainty, and people are freezing to watch it,” he said.

“It doesn't matter if it goes away or not. People will start moving. They'll either start moving because things start getting more certain and it's like ‘it's coming down, great, we're fine with where we are’, or you're going to say, ‘okay, fine, we have to operate now in this environment’.

"Businesses didn't get into business because they expected the world to be easy for them. They will start moving again.”

Generative AI is past peak hype, despite spending bump

When companies return to regular spending patterns, there are indications that businesses will chase more pointed returns on AI investment.

“AI projects are one of the big problems in 2025,” Lovelock told ITPro.

“2024 was not a great year for projects. Lots of them getting done, many of them failing, many of them getting canceled before they could fail. So, depending on how you count it, it wasn't a great year.

Lovelock added that the industry is fast sliding into what Gartner terms the ‘trough of disillusionment with generative AI, meaning it has passed peak excitement on the hype cycle and is failing to meet expectations for some customers.

This is largely down to enterprises lacking hands-on experience with generative AI, which meant they had been allowed to imagine use cases the technology is not actually capable of fulfilling.

Gartner predicts that the technology will move into the next two stages, the ‘slope of enlightenment’ and ‘plateau of productivity’, by which time it will be better understood, more commonplace, and delivering stable revenue by 2029.

Vendors are leading the charge both into and out of the trough, with Lovelock predicting AI becoming a standard feature in software over the next few years.

“Next year, more money will be spent on software with generative AI features than software without generative AI features. For something that came into reality three years ago, that is an amazing thing.”

AI is upending hyperscaler business models

Lovelock noted that until the cost of creating domain-specific models decreases, large language models (LLMs) remain the best path to revenue generation via AI.

This will continue to drive hyperscaler revenue streams, who Lovelock explained are spending an “astounding amount” on AI infrastructure.

The ratio of AI infrastructure spending to ‘traditional’ server spending by hyperscalers is expected to be three-to-one in 2026. This is a departure from the established model of hyperscalers, Lovelock explained.

“It's no longer their business model to buy these servers to rent them out. Yes, they're doing it, but it's the side hustle in the same way that Amazon.com runs on AWS but that's the side hustle.”

In the immediate future, hyperscalers will build out products and services based on the models they’ve made or that they host, Lovelock said, such as Google’s massive expansion of Gemini throughout its products, or Microsoft’s push for Copilot throughout its ecosystem.

“We’re going to see the buyers at enterprises getting generative AI that they didn’t actually buy,” he added.

“They will be sold this generative AI, it’s going to start coming in the software that they already own and operate.”

Vendors are driving the AI adoption curve, as companies turn to third-party model providers to get their AI projects off the ground.

Lovelock gave the example of Salesforce, which rolled out its Einstein AI tool in 2016 based on internal machine learning and deep learning models, but later debuted Einstein GPT which is heavily rooted in OpenAI’s proprietary models.

Organizations around the world are matching this trend in turning to LLM and small language model providers to rapidly roll out AI functionality in their products.

“And when they do that, they’re paying what we call the ‘generative AI tax’,” he said. “Whatever they charge for their product, as a tax to offering that functionality they need to pay a third party. And to a certain extent, they’ve now lost control of their goods sold.”

In the long run, therefore, Lovelock predicted that the price of developing models will fall and companies will gravitate back to using internal models in order to prevent paying these fees.

Make sure to follow ITPro on Google News to keep tabs on all our latest news, analysis, and reviews.

MORE FROM ITPRO

• Conversational AI spending is going to skyrocket this year – here’s why
• AI is finally delivering bang for its buck, according to Microsoft
• Global cloud spending might be booming, but AWS is trailing Microsoft and Google