Why CIOs face rising AI costs without new projects

A chief information officer (CIO) can postpone a large AI project and still face AI-related cost increases. Some arrive through capabilities included in software the organization already buys; others come from infrastructure demand that pushes up memory prices. A cautious rollout limits direct investment in AI. But costs can still rise through these other purchases, so CIOs need to separate the projects they approve from the wider spending affected by AI.

AI spending includes purchases with different intentions

In the spending outlook described by Lovelock, global AI spending is forecast to nearly double in 2026, but that total combines purchases made for different reasons. A company may deliberately select an AI capability, or it may buy a broader product that already includes one. Both count toward AI spending. Only the first, however, reflects a specific decision to adopt the capability.

Lovelock describes some of this growth as a change in how existing technology spending is classified. “There’s a bit of rebranding of the CIO’s dollars to be AI,” he said. Buyers may choose a product for other business needs and get AI with it. As Lovelock put it: “They’re not intentionally going out and buying it, they just can’t get out of the way easily.”

Companies also continue to make deliberate AI investments. Lovelock points to businesses exploring agents, AI systems that carry out tasks, and pursuing automation. He also describes investment in large language models, which generate and work with language, to improve quality, consistency, reliability and processes. These projects reflect an organization’s priorities, while AI bundled into other purchases can enter the budget through a separate decision.

Infrastructure demand can raise memory costs

One route into that budget starts with technology providers building capacity for AI. According to Lovelock, providers are buying servers optimized for AI, and those purchases are helping drive up memory costs. Enterprises then face higher infrastructure costs, even when their AI deployment plans remain limited. The CIO can choose an internal project’s workload and scope, but those choices do not control the wider demand affecting memory prices.

Providers are also introducing products for deploying AI infrastructure. Broadcom, which has a commercial interest in adoption of its infrastructure offerings, introduced VMware Private AI Cloud alongside VMware AI Factory. Broadcom describes VMware AI Factory as the cloud offering’s software-defined foundation: infrastructure managed through software, with capabilities that automate its deployment. AI offerings thus extend into the tools used to set up a computing environment, as well as the applications that run there.

Software can carry AI costs into existing purchases

Software offers another route because providers can add AI capabilities to products enterprises already need. Lovelock says software is becoming more expensive as providers introduce these capabilities. Customers can therefore pay more for an ordinary software purchase, even when the added AI capability falls outside their planned investment. Here, a provider’s product decisions drive the cost increase rather than memory-market demand.

Buyers must judge the whole product, which can make that decision difficult. An organization may need the broader software package while placing little value on the added capability, yet the purchase brings them together. So spending classified as AI can rise faster than deliberate adoption alone would suggest. The total includes customers actively choosing AI and those buying it within software that meets existing business needs.

Salesforce’s AIforce illustrates how providers are adding AI to enterprise software environments. Salesforce, the customer relationship management company, has a commercial interest in customers adopting its AI offerings. It unveiled AIforce as an interface layer for its agentic enterprise, an environment in which AI agents perform work. Salesforce describes AIforce as combining model intelligence with context from its company data to work with core enterprise platforms.

The Salesforce and Broadcom offerings address different parts of the technology environment: one adds an AI interface to enterprise platforms, while the other supports infrastructure deployment. Their availability expands what customers can buy and deploy, but buyers still have to judge their value. That judgment becomes harder when a purchase combines capabilities the organization needs with AI capabilities it values less.

CIOs are choosing smaller projects while they establish value

The difficulty of demonstrating value is already shaping those project decisions. Lovelock says measuring and showing AI’s value is becoming harder, so CIOs are increasingly choosing smaller, incremental use cases over larger projects. Smaller projects limit the commitment while returns remain difficult to demonstrate. They let CIOs control the work they authorize while learning what produces useful results.

Providers’ infrastructure investment leaves room for that deliberate pace. According to Lovelock, CIOs can begin projects on their own schedules even as providers build capacity and expand their offerings. Capacity growth therefore gives CIOs room to choose when to begin; it creates no requirement to match providers’ investment pace.

Within the projects they authorize, Lovelock recommends developing the ability to turn AI concepts into useful results. Leaders should begin with modest expectations, then build skills and repeatable experience through practical work. His advice is to start “building up the muscle memory and the skills for bringing AI from a concept, bringing AI from the lowest level of expectations, to building value.” Capability development becomes part of the project itself, with experience guiding the next commitment.

Key takeaways for leaders

Separate AI spending by intent: CIOs need to distinguish deliberate AI investments from capabilities bundled into broader purchases. This separation helps budget reviews show which costs reflect approved AI priorities.
Account for rising memory costs: Technology providers’ purchases of AI-optimized servers are helping push up memory prices. Infrastructure budget owners need to account for this pressure even when internal AI deployment remains limited.
Assess bundled software value: Software buyers need to weigh higher prices against the value of the full package, including added AI capabilities. A product may meet existing business needs while including AI the organization values less.
Build evidence through smaller projects: CIOs are choosing incremental use cases as AI returns become harder to demonstrate. Modest commitments give project teams room to develop skills, establish repeatable practices and use results to guide further investment.