The
artificial intelligence boom is often measured in GPUs, model parameters, and
data centre investment announcements. Yet behind every new AI cluster sits a
less visible requirement: a large, reliable and increasingly
difficult-to-secure supply of electricity. This is changing the economics of
data centre development. A site can have land, fibre connectivity, financing
and access to advanced computing hardware and still be years away from becoming
operational if it cannot secure sufficient power.
The
scale of the challenge is becoming clearer. The International Energy Agency
estimates that global data centre electricity consumption was around 415
terawatt-hours (TWh) in 2024, equivalent to about 1.5% of global electricity
consumption. Under its base case, data centre electricity demand is expected to
more than double to around 945 TWh by 2030. AI-optimised data centres are
expected to grow particularly rapidly. The numbers illustrate why power has
become a strategic issue for AI infrastructure. But they do not necessarily
reveal where the constraints will emerge. That is where expert networks can add
a different layer of insight.
The Megawatt Race Behind the AI Boom
AI
Data Centres are Changing the Shape of Electricity Demand:
Traditional
data centres already consume substantial amounts of electricity, but AI
workloads are increasing the density of computing infrastructure inside
individual facilities. Training and inference workloads require large
quantities of specialised accelerators operating in tightly interconnected
clusters. As more computational capacity is concentrated into fewer facilities,
the power requirement of an individual campus can become comparable to the
electricity consumption of a sizeable community.
EPRI's
2026 analysis estimates that a single large data centre can require between 100
MW and 1,000 MW, depending on its scale and configuration. EPRI estimates U.S.
data centres consumed approximately 177–192 TWh of electricity in 2024 and
could reach roughly 380–790 TWh by 2030 under different development scenarios.
A
Data Centre Can Be “Planned” Without Being Power-Ready:
One
of the most important distinctions in data centre diligence is between a
project that has been announced and one that has secured a viable path to
electricity. A developer may control land and announce hundreds of megawatts of
planned capacity. That does not necessarily mean the local transmission network
can deliver those megawatts when the facility is ready.
Grid
interconnection can require new substations, transmission upgrades,
transformers, protection systems and network studies. In heavily developed data
centre markets, multiple large-load projects may also be competing for the same
grid capacity. This is one reason why public project pipelines can give a
different impression from the amount of capacity that is realistically
deliverable.
Expert Networks Can Help Separate
Announced Capacity from Deliverable Capacity
The
most useful expert-network research in this market is unlikely to focus solely
on whether AI demand will continue to grow. The more valuable question is what
stands between demand and physical deployment. Interviews with former utility
executives can shed light on grid connection processes.
Data
centre development specialists can explain how projects progress from land
acquisition to energisation. Electrical engineers can identify equipment
constraints. Power-market experts can explain regional electricity economics.
Equipment manufacturers can provide insight into supply-chain pressures.
Together, these perspectives can help investors build a more granular view of
data centre capacity.
Turning Expert Interviews into a
Diligence Framework
The
strongest expert-network research does not rely on a single interview. Instead,
researchers can build a cross-functional network in which different experts
address different parts of the same infrastructure chain. A utility expert
might describe the expected grid connection timeline. A transmission engineer
can assess the infrastructure requirements. A data centre developer can explain
whether that timeline fits normal project execution. An electrical equipment
specialist can identify potential procurement risks.
When
these perspectives converge, researchers can gain greater confidence in the
underlying thesis. When they conflict, the disagreement itself becomes a
diligence signal. For example, if a developer expects a campus to become
operational within two years but several former utility professionals describe
the required grid upgrades as a four-year process, the discrepancy warrants
deeper investigation.
Expert Network Questions That Can
Expose Hidden Power Constraints
Questions
for Utility Experts: Rather
than asking broad questions about data centre growth, researchers can focus on
the mechanics of large-load connections. How much incremental capacity can the
utility realistically provide in the relevant service territory? What
infrastructure upgrades would a 100 MW, 300 MW, or 1 GW customer typically
require? How long can those upgrades take? Which components are controlled by
the utility and which must be funded or constructed by the developer? These
questions can reveal constraints that are difficult to identify from public
company presentations alone.
Questions
for Data Centre Developers: Developers can be asked to explain which milestones
indicate genuine progress. Has the project secured land and permits? Has the
utility confirmed a connection date? Has the substation design been completed?
Have long-lead electrical components been ordered? Is construction dependent on
transmission upgrades? These questions help separate a conceptual development
pipeline from capacity that is moving toward commercial operation.
Questions
for Equipment Experts: Equipment
specialists can provide insight into a less visible part of the AI
infrastructure bottleneck. Researchers can explore transformer lead times,
switchgear availability, manufacturing capacity, order backlogs, and the extent
to which data centre developers are placing orders well ahead of construction.
This can help identify whether equipment procurement could become a constraint
even when electricity generation itself is available.
What Investors Should Listen for in
Expert Interviews
One
of the most important distinctions in data centre diligence is terminology.
Experts can help clarify the difference between power that is theoretically
available, power that has been requested, power that has been contracted, and
power that can actually be delivered to a facility.
The
same distinction applies to data centre capacity. A project can be announced,
permitted, financed, under construction, or operational. Each stage carries a
different level of execution certainty. Expert networks can help researchers
understand what each milestone means in practice.
The Expert Network Advantage: Finding
the Bottleneck Before it Appears in the Numbers
The
biggest value of expert-network research may ultimately be its ability to
identify bottlenecks before they become visible in reported financial results.
A transformer shortage may first appear as a procurement concern. A
transmission constraint may initially appear as a revised connection date. A
shortage of skilled electrical contractors may manifest as a construction
delay.
By
the time the issue becomes visible in company results, the underlying problem
may already have affected project economics. Expert interviews can provide an
earlier window into these developments because professionals working within the
infrastructure ecosystem often encounter constraints before they appear in
headline industry statistics.
Expert
interviews can provide an earlier window into these developments because
professionals working within the infrastructure ecosystem often encounter
constraints before they appear in headline industry statistics.
Conclusion: The Real AI
Infrastructure Question is Who Can Deliver the Power
The
AI data centre boom is creating a new research challenge. It is no longer
sufficient to ask how much computing capacity companies plan to build.
Researchers increasingly need to understand whether the electricity,
transmission infrastructure, electrical equipment and construction capabilities
required to support that capacity will be available at the right time.
Expert networks are particularly well suited to this problem because the relevant knowledge is distributed across multiple industries. Utilities understand the grid. Data centre developers understand project execution. Electrical engineers understand infrastructure requirements. Equipment suppliers understand manufacturing constraints. Power-market specialists understand electricity economics.