The Cost of Innovation
A $63.4 billion infrastructure bill is slowing innovation
Analysis from “The Economic Cost of Unpredictable Technology Infrastructure Report”
AI Infrastructure
AI is the biggest infrastructure bet, with the least predictable bill
Across eight markets, AI is also where cost unpredictability and unexpected spending are most acute.
AI costs are difficult to predict across licensing, deployment and day-to-day operations, much of which is outside our control."
Andy Fletcher, Vice President, Technology & Digital Product, Liverpool FC
When infrastructure costs rise, innovation pays the price
Recovered budget spurs economic and job growth
Infrastructure costs cannot be controlled
71%
of businesses spent more on infrastructure than they expected over the past 12 months — 21% significantly more
Companies' budget pressures build
73%
of those hit by unpredictable or unplanned costs say it created significant or moderate budget pressure
Businesses cut back on innovation
36%
of those companies facing budget pressure have cut back on AI initiatives in the past year
Opportunity cost if budget is recovered
$21.8bn
of provider-attributable overspend could be redirected towards capital-forming investment in R&D, AI initiatives and new technology
The innovation opportunity could become narrower
Every dollar absorbed by prices, tariffs and scarcity is a dollar that cannot fund the innovation opportunity.
Infrastructure constraints are driving up the costs and slowing the pace of innovation.
With 82% of organizations facing delays in the past 12 months due to lack of access to suitable hardware or compute capacity, it's hard to innovate.
Vendor Lock-in Constraints
Lock-in makes rising costs harder to escape
Thinking of switching providers to gain predictability? Moving data, applications, and workflows carries costs and complexity.
Types of lock-in experienced at least once in the last year:
Innovation levels determine gravity
Budget and technical pressure isn’t just slowing innovation, it’s stopping it
Data Movement Pressures
Data movement pressures are halting innovation
Innovation initiatives impacted by data access and infrastructure costs
We nearly made a trade-off based on cost. Our storage costs were climbing and the obvious fix was to move older recordings to a cheaper, colder tier, but it would be slow to retrieve data. For a product where the recording must be ready the moment a meeting ends, that would have meant sacrificing the instant access our customers rely on."
Allan Bettarel, CTO at tl;dv
The Big Picture. . .
The innovation bleed is reversible.
How reclaiming provider-attributable overspend impacts on a broader economic scale.
When infrastructure becomes more predictable, money flows back to innovation
Businesses say that if provider-attributable overspend were avoided, they would reinvest the savings in AI, technology, people and productivity.
&w=1920&q=75)
Download the complete report
Download the full report to access the complete report and analysis from the Centre for Economics and Business Research (CEBR).
Get the Report
Frequently asked questions
The research found among organizations with AI overspend, the leading causes were higher-than-expected licensing costs for AI platforms, models and tools (49%) and integration, deployment and model operations expenses (45%). Token, inference and usage-based charges ranked third at 38%. This matters because usage-based costs are the ones most organizations try to forecast. The commercial terms and the cost of getting models into production are harder to see in advance and harder to control once committed.
Global AI infrastructure spending reached $1.43 trillion in 2026, up from $976 billion in 2025, with AI spending growing more than three times faster than overall IT budgets. The five largest US cloud and technology providers alone have committed between $660 and $690 billion in capital expenditure for 2026. For individual organizations, average monthly AI spend was projected to reach $85,521 in 2025 — up 36% year over year. Flexera, CloudZero
Vendor lock-in makes it structurally difficult for organizations to respond to rising costs. When monthly AI infrastructure spend varies significantly, finance teams cannot forecast accurately — leading to either over-budgeting, which ties up capital, or under-budgeting, which forces mid-year budget requests and creates friction between engineering and finance. Organizations that are locked in through data egress fees, proprietary APIs, and workflow dependencies cannot easily switch providers to find better pricing even when costs become unsustainable.
When costs are unpredictable, engineering teams spend more time analyzing bills, optimizing resource allocation, and managing cloud configurations — time that could be spent on model development, data quality, and AI product features. The infrastructure effectively becomes a tax on the AI team's productivity.
Beyond compute and storage, AI cloud environments carry several layers of hidden costs — managed Kubernetes services, container registries, logging and monitoring, identity management, load balancers, and data egress fees. Individually each service may appear inexpensive, but collectively they can represent a meaningful percentage of total infrastructure spend, and their costs scale with usage in ways that are not always obvious during planning.
The research identifies where the unpredictability comes from: the largest share of AI overspend is attributed to opaque or difficult-to-forecast provider charges rather than businesses’ own usage. It also shows that lock-in makes the problem harder to escape, with the cost of moving data the single biggest barrier to switching provider.
Wasabi’s view is to reduce dependence on any one cloud provider so switching remains possible. Choose pricing you can forecast a year out, with no per-request or egress charges, so every API call, data read and training cycle runs without triggering a surprise bill.
Provider-attributable overspend refers to infrastructure costs that organizations incur not because of their own usage growth, but because of pricing decisions made by their cloud providers — including price increases, opaque fee structures, egress charges, and API request fees. The CEBR research featured on this page estimates $63.4 billion in provider-attributable overspend annually that could be redirected toward R&D, AI initiatives, and new technology investment if eliminated.
The CEBR research commissioned for this report estimates that eliminating provider-attributable overspend could unlock $21.8 billion for reinvestment in innovation, R&D, and AI hiring. Beyond individual organizations, this has a broader macroeconomic effect — when infrastructure becomes more predictable and affordable, money flows back into capital-forming investments that drive economic and job growth.