Is AI the main cause of business power pressure?
A PwC survey of 1,003 US large-energy users found widespread concern about reliability and AI-related computing growth. Respondents placed AI alongside expansion, electrification, automation and data centres — and the survey measures expectations, not actual demand or future savings.
The short version
- What happened
- PwC surveyed 1,003 US directors and senior leaders between 27 January and 5 February 2026; 77% expected computing-power use to rise over two years, largely because of AI and machine-learning workloads.
- Who it affects
- Workers may benefit from more reliable facilities, but automation and energy-management systems can also change roles, monitoring and staffing requirements.
- What remains unproven
- The survey represents 1,003 senior US energy decision-makers, not all firms, workers, households or grid operators.
Keep in mind
The survey represents 1,003 senior US energy decision-makers, not all firms, workers, households or grid operators.
Recruitment, response rate, weighting, full question wording and uncertainty intervals are not disclosed on the public page.
Reliability problems, investment intentions and expected AI savings are self-reported rather than independently audited.
PwC has a commercial advisory interest in energy, infrastructure and AI services.
LBNL scenarios corroborate demand growth direction but do not validate PwC’s respondent percentages or causal claims.
Multi-source analysis · 2 linked sources. Primary analysed source date: 15 October 2026.
Read the full report · 7 min The evidence, methods, implications and limitations
The answer: AI is part of a broader squeeze, not the only cause
US companies that use large amounts of electricity are planning for tighter supply, higher costs and more computing demand, according to PwC’s new survey. AI is clearly part of that picture: 77% of respondents expected their computing-power use to rise over the next two years, largely because of AI and machine-learning workloads, while 84% called power interruptions a significant risk to AI investment. But the survey does not support a claim that AI alone created the pressure.
Respondents placed five demand drivers within a narrow 41%–42% band: AI and machine learning, facility expansion, electrification of industrial processes, automation and data-centre growth. That convergence is the more useful finding. Manufacturers, warehouses, campuses, retailers and data centres are adding different loads for different reasons. Public modelling from Lawrence Berkeley National Laboratory independently projects strong US data-centre electricity growth, but it does not turn business opinions into measured demand or prove which driver caused a particular outage or delay.[1][2]
Who answered the survey and when
PwC collected responses from 1,003 leaders across five US regions between 27 January and 5 February 2026. Every respondent held a director-level or more senior role in sustainability, facilities or energy management. The sample covered petroleum refining, chemicals and plastics manufacturing (15%); other manufacturing (15%); distribution centres (14%); corporate parks or campuses (14%); data centres (12%); large retail (15%); and universities (15%). Those denominators make the broad comparisons easier to understand.
The public methods stop there. The page does not say how organisations were recruited, how many invitations were sent, whether results were weighted, how questions were worded or whether one person could speak for several facilities. It also does not supply uncertainty intervals. Percentages separated by only one or two points may therefore reflect sampling and question design rather than meaningful rank differences. The respondents are energy decision-makers, not a representative sample of US businesses, households or grid operators.[1]
Reliability concerns are real to respondents, but still self-reported
Three quarters of respondents said unreliable electricity supply had caused operational or financial impacts, 70% were less confident in grid reliability than two years earlier and 35% named energy supply as their largest growth constraint. Nearly all reported at least one supply problem in the preceding two years, including connection delays, long waits for infrastructure upgrades or increased outages. These answers show that reliability is influencing investment conversations across the sampled organisations.
They do not provide outage duration, lost production, interconnection-queue records or audited financial losses. PwC itself notes that perceptions may not align with system-wide reliability data. A company waiting for a new high-capacity connection faces a different problem from an existing site experiencing repeated service interruptions, yet both can report that power limits growth. Policy and investment decisions should therefore pair survey evidence with utility records, regional load forecasts and facility-level measurements.[1]
Companies expect AI to consume more power and help save it
The survey captures a striking tension. Eighty-seven per cent expected energy demand to exceed available supply as AI adoption accelerates; 78% considered electricity prices a significant barrier to AI adoption; and 88% said energy resilience was critical to rising computational demand. At the same time, nine in ten expected AI’s long-term energy savings to justify its initial consumption, and 89% expected AI-powered systems to reduce operational energy costs within two or three years.
Those savings are forecasts by interested organisations, not measured outcomes from deployed systems. AI-enabled monitoring, forecasting and controls can improve scheduling, detect faults and shift flexible loads, but savings depend on baseline efficiency, equipment, tariffs, staff and rebound effects. A system that makes each computation more efficient can still raise total electricity use if the organisation runs many more computations. The survey does not quantify that balance or compare AI projects with non-AI efficiency investments.[1]
Independent modelling supports growth but exposes uncertainty
Lawrence Berkeley National Laboratory’s June 2026 update provides a different kind of evidence. Its bottom-up model combines expected shipments of data-centre equipment, per-device annual energy use, cooling simulations and information about facility type and location. The reference case estimates 649 terawatt-hours of US data-centre electricity consumption in 2030, or 11.8% of national use. Sensitivity and compounded-uncertainty scenarios range from 521 to 843 terawatt-hours, equivalent to 9.5%–15.3%.
That wide range is a warning against false precision. The scenarios change assumptions about equipment installations, specialised graphics-chip shipments, AI-chip operating life, idle power and utilisation. They support the direction of PwC’s concern — data-centre electricity demand is likely to grow materially — while showing how much the scale depends on technology and deployment choices. They do not validate PwC’s company-level percentages, reliability perceptions or forecast that AI controls will pay back within three years.[2]
Investment plans could affect workers, customers and communities
Seventy per cent of respondents planned at least $10 million in energy-related investment over two years, and 90% said they were likely to partner with private providers rather than rely only on utilities. More on-site generation, batteries, private power agreements and demand-management software could protect operations and jobs from interruptions. It could also move costs, pollution, land use and reliability risks between companies, ratepayers and host communities depending on fuel, location and tariff design.
Workers may gain more reliable facilities and new roles in energy management, controls and maintenance, while also facing automated monitoring and changed operating practices. Customers may benefit from fewer disruptions but pay more if infrastructure costs are socialised without safeguards. Communities need transparent information about who pays, which generation is added, water and emissions impacts, local employment and whether large loads can reduce consumption during grid stress. The survey asks business leaders; it does not measure those public outcomes.[1][2]
Evidence limits and what would change the assessment
PwC publishes the sample size, respondent seniority, sector mix and field dates, which is more useful than an undated headline survey. It is nevertheless a consultancy report written partly as advice to utilities, oil and gas companies, technology vendors and energy-orchestration providers. PwC has a commercial interest in the market it describes. The page does not provide a questionnaire, microdata, weighting, response rate or independent peer review, so each percentage should remain attributed to PwC’s respondents.
Confidence would rise if PwC released full questions, recruitment and weighting details, subgroup denominators and uncertainty intervals. Independent researchers could then link company expectations to measured load, outages, connection times, investment completion and energy savings. The most important future evidence is longitudinal: whether planned projects are built, whether AI optimisation reduces total rather than unit consumption, and whether costs and reliability benefits reach workers, customers and communities rather than only the large-load owner.[1][2]