Unexpected infrastructure bills leave UK firms less for innovation

Unexpected infrastructure bills can reduce innovation budgets

An AI project can consume money reserved for its next stage when infrastructure bills exceed expectations. Research commissioned by Wasabi estimates that UK businesses spend GBP £3.8 billion a year more than planned on technology infrastructure, with much of that overspend attributed to charges they found hard to forecast. Paying those bills can leave less money for AI development, hiring and new products.

The Centre for Economics and Business Research, or CEBR, conducted the research. Wasabi supplies infrastructure services, so it has a commercial interest in how businesses choose and pay for them. The research combined CEBR’s economic modelling with a survey of 690 senior business and IT decision-makers across the UK, US, France, Germany, Italy, Japan, Australia and New Zealand. Almost three-quarters of surveyed UK businesses reported infrastructure spending above expectations over the past 12 months; AI accounts for a substantial part of the estimated UK overspend.

AI costs accumulate across implementation and use

Forecasting AI spending is difficult because putting it to work involves several kinds of charges. More than one-fifth of British companies in the survey described AI infrastructure costs as unpredictable, identifying licensing, integration, deployment and usage charges. These costs arise at different stages, so the initial commitment gives budget owners only part of what they need to plan spending.

A complete forecast therefore has to follow the system from purchase into operation. Licensing covers the right to use the technology, integration connects it with existing systems, and deployment brings it into use. Usage charges add to the bill as people run the system. Budget owners need to account for both the work required to make AI usable and the costs that follow.

The difficulty continues after an organisation decides to proceed, as Andy Fletcher, Liverpool FC Vice President of Technology and Digital Product, described when discussing licensing, deployment and day-to-day operations. Running costs still have to fit within the available funds. If the eventual bill exceeds the forecast, the business has to decide which other commitments can wait.

For Wasabi Technologies President and Chief Operating Officer Marty Falaro, this uncertainty puts purchasing terms at the centre of the problem. His company commissioned the research and stands to benefit when businesses choose its infrastructure services. “Businesses are not short of ambition, but they are blindsided by unpredictable infrastructure costs that prevent them from driving their innovation agendas forward,” Falaro said.

Falaro’s proposed response is to establish data costs before committing to an AI project. He said: “There is a fundamental problem with how technology infrastructure is bought and priced. Businesses need to know what it will cost to store, access and move their data before they commit to an AI project, not after the bill arrives. Innovation should take risks, but the infrastructure bill shouldn’t.”

AI overruns compete with funding for future work

The scale of the forecasting problem becomes clearer when the survey’s two measures of AI overspending are separated. Among UK businesses that exceeded their infrastructure budgets, 61% identified AI and machine learning infrastructure as their largest overspend area. Machine learning uses data to learn patterns; the survey measure counts how often businesses named this infrastructure category. It does not mean AI accounted for 61% of the extra spending.

Measured in pounds, CEBR estimated that AI and machine learning infrastructure contributed GBP £1.2 billion, or 32%, of total UK infrastructure overspend. It was the largest category, though it represented less than half the overall amount. The monetary estimate shows AI’s contribution to total excess spending, while the business count shows how widely it was the largest overrun within affected organisations.

Those overruns matter because existing commitments compete with investment in future work. Over the past year, 77% of UK organisations in the survey reported at least one cost-driven trade-off. Among organisations making those trade-offs, 38% identified hiring in technology, data and AI roles as the area most likely to lose out. A further 37% pointed to technology and software investment, while 34% identified training and workforce development.

These choices affect the people, tools and skills available to deliver technology projects. Postponing a software purchase changes the resources a team can use, while delaying recruitment or training changes its ability to build and operate a system. An infrastructure overrun can therefore reduce delivery capacity beyond the period in which the bill is paid. The trade-off percentages apply to organisations already making cost-driven adjustments.

The global survey findings link unpredictable costs more directly to pressure on development budgets. Among businesses facing unpredictable infrastructure costs, 73% reported moderate or significant budget pressure. Within that pressured subgroup, 36% said they had already delayed or reduced AI, automation or data science initiatives. This last finding applies to businesses under that pressure, rather than the entire multinational survey population.

Owen Good, Head of Economics Advisory at CEBR, interprets these findings as a loss of investment capacity. CEBR conducted the research commissioned by Wasabi, and Good’s comments explain the economic consequences it identified. “Our research uncovered the hidden cost of innovation for some of the world’s largest economies and found the economic cost of unpredictable pricing is not simply that firms pay more,” he said.

Good linked that additional spending to the commitments businesses postpone: “This means hiring and investment in new technologies are postponed or reduced as budgets are redirected to cover unexpected charges. The technology generating the most unplanned costs is also among the investments businesses are cutting to pay for them.” He is interpreting the aggregate findings: AI is both the largest infrastructure overspend category and among the initiatives curtailed under budget pressure. For each business, the practical question is how much remains for development, recruitment and training after it pays its infrastructure bills.

Lower prices still have to support delivery

A business seeking to protect its remaining budget may consider changing suppliers, but switching can bring extra costs. Among businesses in the survey considering a supplier switch, 91% reported additional costs associated with fragmented data environments, where data is spread across separate systems. The quoted price tells only part of the story. Buyers also have to account for the work required to manage data across those systems.

Data access adds an operational constraint to the same purchasing decision. Among surveyed businesses considering a switch, 85% had experienced restricted or delayed access to data needed for a project or decision at least once in the previous year. People need that data when their work calls for it, so buyers have to weigh any savings against the proposed service’s access conditions. A cheaper arrangement can still delay the work it is meant to support.

The storage decision at tl;dv shows how retrieval performance can determine that choice. Allan Bettarel, the company’s Chief Technology Officer, said it had considered moving recordings to a lower-cost storage tier, a service option with different pricing and access characteristics. Retrieval would have been slower, so the company decided against the move. For tl;dv, the access delay that came with the lower price did not fit its operational requirements.

Even when an arrangement meets those requirements, suitable resources may be unavailable. Among surveyed UK companies, 84% had experienced delays because suitable hardware or compute capacity, the processing resources needed to run workloads, was unavailable. A business can therefore have funding allocated and still lack the resources to proceed. Resolving price uncertainty leaves availability to be addressed separately.

The survey also recorded consequences from hardware availability and data-access problems. Among respondents reporting consequences from those challenges, 32% said they had reduced or delayed planned hiring in technology, data, AI or innovation roles. Another 28% reported delayed or slower product-development cycles, and 26% reported postponed or reduced AI, automation or data science initiatives. These findings link delivery and staffing decisions specifically to hardware and access difficulties, keeping their cause distinct from unpredictable pricing.

Recovering overspend could release investment

The investment displaced by infrastructure costs gives cost control wider economic significance. CEBR modelled a scenario in which overspend attributable to providers was recovered and estimated that the released spending could support GBP £1.4 billion of productive investment in research and development, AI and new technology. It also estimated potential workforce-training spending of GBP £800 million. These figures describe possible uses of recovered funds under the model.

The recovery assumption matters because a lower-priced arrangement can introduce access or performance costs of its own, as tl;dv’s decision illustrates. CEBR’s estimates describe the potential opportunity from recovering overspend; they do not establish the benefit of a particular supplier change. Buyers still have to account for the replacement service’s costs and operating requirements.

Within its recovery scenario, CEBR also estimated support for 8,235 jobs and a contribution of GBP £1.4 billion to long-run gross domestic product, or GDP, the value of goods and services produced in an economy. The GDP figure measures an economic effect over time, while the identically sized productive-investment estimate measures a possible use of released spending. They are separate measures and cannot be added together as a single benefit.

Key executive takeaways

• Protect innovation budgets: Wasabi-commissioned CEBR research estimates UK infrastructure overspend at £3.8 billion annually. Finance and technology leaders need to assess how unexpected bills affect funds allocated to AI development, recruitment and new products.
• Forecast AI costs through operation: AI budgets need to cover licensing, integration, deployment and ongoing usage. Budget owners can reduce uncertainty by establishing data storage, access and transfer charges before committing to a project.
• Track the investment displaced by overruns: AI and machine learning account for an estimated £1.2 billion, or 32%, of UK infrastructure overspend. Finance teams need to track which hiring, training and development commitments are postponed to cover those bills.
• Test savings against delivery requirements: Lower-priced services can introduce slower retrieval, data-access restrictions and migration costs, while hardware shortages can delay funded projects. Procurement and technology teams need to assess total costs, access conditions and capacity availability together.
• Treat recovered spending as an investment opportunity: CEBR models potential productive investment of £1.4 billion and workforce-training spending of £800 million if provider-attributable overspend were recovered. Finance teams evaluating supplier changes need to establish achievable savings after transition and operating costs before reallocating funds.