The AI Spending Trap: How Unpredictable Infrastructure Costs Are Stalling Innovation

Businesses are investing heavily in artificial intelligence to develop new products, improve productivity, and compete in a rapidly changing market. But there's a problem. The infrastructure supporting those ambitions is becoming increasingly difficult to budget for.

Our new 2026 Cost of Innovation study, conducted with economic modeling from the Centre for Economics and Business Research (Cebr), found businesses across eight major economies (including six of the G7 nations) incur an estimated $63.4 billion annually in unexpected technology infrastructure costs attributable to unpredictable provider charges.

AI and machine learning account for $23.4 billion of that total, more than any other infrastructure category. In response, businesses are already cutting innovation budgets to cover the difference.

Why are AI infrastructure costs so difficult to predict?

The Cost of Innovation study examined six technology infrastructure layers: storage, compute, networking, data and databases, AI and machine learning, and hardware and supply chain. Among businesses that overspent, actual infrastructure spending averaged 17% to 21% above expectations, depending on the category.

Not only is the AI and machine learning layer the largest share of provider-attributable overspend, it's the hardest infrastructure category to forecast. Some 22% of businesses reported unpredictable AI costs, nearly twice the rate of other technology infrastructure layers.

AI projects involve multiple services and pricing models. As businesses move from testing applications to deploying them at scale, costs accumulate across licensing, compute, model operations, and data infrastructure.

Among businesses reporting AI infrastructure overspend, 38% cited higher-than-expected token, inference, or usage-based charges. Unexpected model licensing costs (49%) and integration or operational expenses (45%) were even more common.

Businesses know AI requires significant investment. They should also be able to estimate what it will cost before committing to a project.

How are unexpected infrastructure costs affecting AI investment?

The research found that 71% of businesses spent more on infrastructure than expected over the past year. Among businesses experiencing unpredictable infrastructure costs, 73% reported moderate or significant budget pressure. Of those facing that pressure, 36% had already delayed or scaled back AI, automation, or data science initiatives.

That should concern anyone responsible for technology investment. Companies are redirecting money away from projects intended to create future growth because their existing infrastructure costs more than anticipated.

And changing providers isn't always a realistic solution.

Why does cloud vendor lock-in make infrastructure overspending worse?

The Cost of Innovation study found 44% of innovation laggards and 22% of innovation leaders face difficulty switching infrastructure providers. Data transfer fees such as API and egress charges, fragmented infrastructure, and migration complexity can make it difficult for businesses to move to more affordable services, even when the potential savings are clear.

The research also found 89% of businesses incurred additional costs associated with fragmented data environments, while 84% experienced restricted or delayed access to data needed for a project or decision at least once in the past year.

For AI workloads, which frequently depend on accessing and moving large volumes of data between storage and compute environments, these restrictions can limit flexibility and increase costs.

Businesses should be able to choose infrastructure based on what works best for their applications. They shouldn't be financially penalized for moving their own data.

How much innovation is being lost to infrastructure cost overruns?

Across the eight economies studied, 84% of organizations made at least one cost-driven trade-off in the past year. Among businesses unable to switch infrastructure providers, that figure rose to 92%.

The consequences for innovation are tangible:

• 29% reduced or delayed planned hiring in technology, data, AI or innovation roles

• 28% delayed or scaled back an AI, automation or data science initiative

• 25% ran fewer AI, machine learning or data science experiments

• 22% avoided adopting a new technology or platform

• 16% delayed or canceled a product launch

A delayed AI initiative can mean losing months of development time. A canceled product launch can mean missing a market opportunity. These decisions affect far more than the IT budget.

Cebr's economic modeling puts the potential opportunity cost into perspective. Recovering the $63.4 billion in annual infrastructure overspend could support $21.8 billion in additional investment in R&D, AI, and new technology; $7.6 billion in workforce training; and 111,450 additional jobs.

The model also estimates a potential $20.4 billion increase in long-run GDP across the eight economies studied. That's a substantial amount of economic activity tied to a problem businesses should have more control over.

How can businesses make AI infrastructure costs more predictable?

Some AI costs will always fluctuate. Model pricing changes, inference demand varies, and compute requirements evolve. But organizations can make more deliberate decisions about how they store, access, and move the data supporting AI workloads.

This is the thinking behind AI Storagemaxxing, an approach to optimizing data storage and access across different stages of the AI pipeline to improve cost efficiency.

Liverpool FC, a Wasabi customer, is exploring this approach as it manages decades of video footage, with new content arriving every matchday. As Andy Fletcher, the club's Vice President, Technology & Digital Product, explained in the research: "AI costs are difficult to predict across licensing, deployment, and day-to-day operations, much of which is outside our control. But how we manage the data behind AI is something we can control."

He's right. Businesses should be asking straightforward questions about their infrastructure providers. What will it cost to store our data? Are we charged every time we access or move it? Can we switch environments without incurring substantial fees?

At Wasabi, we've long believed that businesses should have the freedom to store, access, and move their data without unpredictable fees getting in the way. The Cost of Innovation research shows why that matters, especially as AI workloads grow and businesses need greater control over their infrastructure spending.

To help businesses put real numbers behind those costs, we've introduced the AI Data Freedom TCO Calculator. It compares the total cost of hyperscaler cloud storage, including egress charges, API fees, and other hidden costs, with Wasabi's predictable pricing so organizations can see how much they could save.

Businesses should be taking risks on new ideas and technologies. They shouldn't have to gamble on what their infrastructure bill will look like.

Research methodology: Wasabi Technologies commissioned Sapio Research to survey 690 senior business and IT decision-makers across the US, UK, France, Germany, Italy, Japan, Australia, and New Zealand in July and August 2026. Economic modeling was conducted by the Centre for Economics and Business Research (Cebr).