Global IT spending is expected to increase by 9% in 2025, driven by AI and cloud expenditures, according to a report from S&P Global Ratings.
But a divided picture is also emerging: As major cloud providers — AWS, Microsoft Azure and Google Cloud — along with Meta raise their AI investments to about $320 billion this year, businesses remain tentative about AI deployments.
“Enterprises are still figuring out use cases that can deliver good ROI. Explainability, security and legal risks remain key challenges,” S&P’s Technology Director Christian Frank told PYMNTS.
“Most enterprises are still in the proof-of-concept stage, and while we expect more to roll out broader implementations in 2025, we expect enterprises to remain cautious on AI implementation,” he added.
In the report, he wrote that companies are selectively picking areas for broader implementation of AI as they embark on a “more measured path.”
“This careful approach reflects ongoing challenges including a fragmented vendor landscape, scarce AI talent and established best practices, and emerging legal and compliance risks,” Frank added.
Last year, IT spending rose by 8.3%, according to the S&P report.
The search for ROI validation comes as Chinese AI startup DeepSeek recently surprised Wall Street and Silicon Valley with an AI foundation model that only cost $5.6 million to train and used 2,048 of slower Nvidia AI chips while performing at par with the best models from the U.S.
This stands in contrast to the hundreds of millions spent by U.S. AI companies for a large language model, which may also use tens of thousands of costly, top-of-the-line Nvidia AI chips.
Moreover, DeepSeek’s AI model is open source: It’s free to use and revise. Major U.S. AI companies typically charge for access to most of their AI models, including OpenAI, Google and AWS., with Meta being the exception. DeepSeek’s inference costs — what it charges to actually use the model — are also a lot lower than those for top U.S. AI companies.
However, Frank still is not fully convinced that DeepSeek’s savings will flow through U.S. company coffers.
DeepSeek’s “breakthrough signals a shift toward more efficient AI model training and deployment, lowering the cost to access AI models, which could improve return and accelerate adoption,” Frank wrote. However, “we need additional confirmation points to validate this hypothesis and its potential impact on global IT spending patterns and competitive dynamics.”
DeepSeek’s R1 reasoning model is being offered on the three major cloud providers’ platforms. However, several countries and states have also banned government employees from using its chatbot app.
While DeepSeek introduced a way to cost-effectively train AI models, instead of shrinking the capital expenditure of major cloud companies, it did the opposite.
“Hyperscalers are interpreting improved efficiency as a catalyst for broader AI adoption,” Frank said.
He cited Microsoft’s observation that lower AI input costs would increase the use of generative AI, or the Jevons paradox, “an economic theory that advancements in the efficiency of a resource increases its demand.”
IBM CEO Arvind Krishna recently predicted that AI adoption will “explode” as costs drop.
That’s what OpenAI could be banking on. It has partnered with SoftBank, Oracle, MGX on a project called Stargate that would invest $100 billion to $500 billion on AI infrastructure, which consists mostly of data centers.
However, S&P’s IT spending forecast could change depending on the outcome of any tariff fight. “Trade policy is a key risk to our IT spending forecast that we will monitor in the coming weeks and months,” Frank said.
Meanwhile, S&P said purchases of hardware such as PCs are accelerating partly due to customers looking to get ahead of the Trump tariffs.
On Feb. 4, an additional 10% tariff on Chinese imports took effect. There was a pause on U.S. tariffs of 25% on Mexican and Canadian goods for one month after they reached agreements with the U.S. on border control.
S&P expects PC sales to increase by 3% in 2025, up from 1% in 2024, as companies look to refresh “aging COVID-19-era” computers and Microsoft moves to end support for Windows 10 this October.
But AI has not been a catalyst for sales of PCs and smartphones, Frank said.
This year, S&P expects smartphone sales to struggle even after adding AI. “Despite the introduction of new AI features on many flagship smartphones, consumer demand has been relatively subdued,” Frank said.
This hasn’t stopped manufacturers from introducing AI-enabled devices, even if enterprises and consumers are “still evaluating use cases and value propositions before making significant hardware investments,” he added.
