HIGH UNCERTAINTY ON PRODUCTIVITY; HIGH CONFIDENCE ON INFRASTRUCTURE DEMAND | STRATEGIC JUDGEMENT
Is AI a productivity revolution, labour-market shock, or infrastructure crisis?
Verdict
AI is all three, but on different timescales and with different confidence. Task-level productivity and occupational disruption are already visible; economy-wide gains remain modest and conditional; infrastructure demand is rapid and geographically concentrated. For TCE, the immediate issues are power, cooling, water, fibre, cybersecurity, data governance and the changing purpose of physical places not a simple forecast of fewer offices.
Evidence anchorsInternational Energy Agency, 2025 · ILO, 2025 · IMF, 2025
AWhat is changing
AI is moving from a specialist tool to a general layer across knowledge work, operations, customer service and physical systems. The strongest evidence is at task level: drafting, coding, search, scheduling and pattern recognition can become faster, especially for less experienced workers and clearly defined tasks. Aggregate productivity remains uncertain because gains require organisational redesign, complementary investment and diffusion beyond leading firms.
Labour-market exposure is broad but not equivalent to job loss. The ILO's refined 2025 index estimates that one in four workers globally is in an occupation with some generative-AI exposure, while transformation is more likely than complete replacement. Clerical and routine cognitive work is most exposed; human judgement, care, negotiation, physical dexterity and accountability remain complementary. The distributional risk is that productivity gains accrue to owners and highly skilled users while transition costs fall on routine workers.
The physical footprint is immediate. Data-centre electricity use is projected to rise from about 415 TWh in 2024 to roughly 945 TWh by 2030, while local water, transmission and land constraints can delay projects. AI also increases demand for advanced chips, secure networks and cyber capability. The "cloud" is becoming more visibly territorial: large loads, cable routes, cooling systems, backup generation and contested planning decisions.
FORECAST
The IMF's preferred modelling for Europe suggests around 1% cumulative productivity gain over five years, with larger upside if adoption and labour reallocation are effective. This is a modelled scenario, not observed productivity (IMF, 2025).
OBSERVED
Data-centre demand is already concentrated in particular grid areas. The global figure understates the local effect on substations, water and transmission (IEA, 2025).
BThe bolder interpretation
The edge view is that AI increases the scarcity value of trusted human and physical interaction. As synthetic content, automated service and remote work proliferate, high-quality places may become verification environments: people meet to build trust, experience culture, receive care, learn socially and make consequential decisions. This could strengthen the value of curated public realm and hospitality even as routine office attendance declines.
A second edge is that "smart place" optimisation becomes a surveillance and legitimacy risk. Digital twins, sensors and algorithmic management can improve energy, maintenance and crowd flows, but they also create questions about who owns place data, how people are profiled, and whether public-facing spaces become privately governed behavioural systems. Efficiency gains may be rejected if users cannot understand or contest decisions.
A third edge is infrastructure triage. Where grid and water capacity are scarce, governments may need to decide which AI workloads merit priority. High-value national capability may coexist with low local employment and high resource use. The issue is not whether data centres are beneficial in the abstract, but what service, sovereignty and local value they provide relative to other uses.
CGeographic evidence
| Lens | Trajectory | Challenge to UK assumptions |
|---|---|---|
| UK / Europe | Services create high exposure; regulation and grid constraints may slow diffusion. Strong universities and finance coexist with dependence on foreign compute and platforms. | AI sovereignty is partly an energy, cloud and procurement question. |
| China | State-backed power, data hubs and domestic platform ecosystems support scale; surveillance and carbon intensity create trade-offs. | Infrastructure coordination can accelerate AI, but governance conditions are not transferable. |
| India | Digital public infrastructure and a young services workforce support rapid application, while informal work, skills and power reliability create uneven benefits. | AI can diffuse through public digital rails without equal physical or social capability. |
| Japan | AI and robotics address labour scarcity in care, logistics and manufacturing, with cautious, human-centred governance. | Demographic need can drive adoption even without a headline productivity boom. |
EvidenceILO, 2025 · NDRC China, 2023 · METI Japan, 2025
DStakeholder effects
| Segment | Potential value | Vulnerability |
|---|---|---|
| Knowledge workers | Augmentation and flexibility | Work intensification, deskilling and fewer entry routes |
| Frontline / place-bound workers | Safer operations and assistive tools | Monitoring, algorithmic scheduling and weak bargaining power |
| Young workers / Gen Z | AI-enabled learning and entrepreneurship | Credential erosion and disappearance of junior tasks |
| Commercial occupiers | Smart operations and productivity | Cyber dependency and uncertain space demand |
| Data-centre operators | Connected power, fibre, cooling and security | Consent, water, grid allocation and sovereignty scrutiny |
| Visitors / public | Convenience and personalisation | Privacy, manipulation and synthetic-content distrust |
EImplications for The Crown Estate
- Urban portfolio: design for more variable attendance and higher expectations of experience, trust, learning and social connection. Avoid assuming a linear collapse or recovery in office demand.
- Data and technology: adopt digital twins and automation with explicit data minimisation, cyber resilience, human oversight and public-facing transparency. Define ownership and permitted use of tenant, visitor and operational data.
- Infrastructure: treat data-centre or AI occupiers as system users requiring a power-water-cooling-benefit case, not simply high-credit tenants.
- Rural and Marine: use AI for ecological monitoring, maintenance and planning, but retain auditability and recognise that models can reproduce weak baselines or hide uncertainty.
- Capability: build enough internal technical and procurement expertise to challenge vendors. Outsourcing AI judgement without internal competence would replicate the institutional hollowing-out identified in Q9.
Core trade-off
Innovation can increase productivity and system intelligence, but may concentrate power, consume scarce infrastructure and weaken trust. TCE must decide where optimisation justifies data collection and automation and where human accountability and public legitimacy require restraint.
FStress test
The original three-way classification is useful but encourages a single verdict. The evidence shows different outcomes by sector, worker and place. The sharper question is what conditions determine diffusion, who captures gains and which physical and governance constraints bind first.
Rewritten question
Under what infrastructure, organisational and trust conditions will AI create broad productivity gains rather than concentrated rents, workforce disruption and new pressure on power, water and place?
Provocations
- What if AI's most important property impact is demand for trust-rich places, not fewer offices?
- Which AI workloads are worth allocating scarce electricity and water to?
- What place data should TCE refuse to collect, even when it could improve efficiency?
Board prompts
- Where could AI materially improve TCE decisions, and where must human accountability remain explicit?
- How should TCE assess local value and resource intensity of data-centre demand?
- What capabilities must remain in-house to govern vendors and cyber risk?
- How might AI alter entry-level employment and the future occupier base?
Signpostsmeasured firm-level productivity; entry-level vacancies; data-centre connection requests; water and energy intensity; AI regulation; cyber incidents; public attitudes to smart-place data; concentration of cloud and model providers.