Data Center
Powering
Models
August 2025
Public version
This is the public preview of the full report
on the Sightline platform. To become a
Sightline Climate client, talk to our expert
team at hello@sightlineclimate.com.
About Sightline
Sightline is a tactical intelligence platform purpose-built for
decision-makers looking to stay ahead in the transition. We
provide AI-powered data, tools, and research to help
investors, energy and utility leaders, and banks evaluate
technologies, monitor tipping points, and act with confidence
across sectors in power, grid, fuels, carbon, and industrials.
We work with 80+ leading organizations — including Southern
Company, HSBC, BHP, B Capital Group, Tokyo Gas, and
Energy Impact Partners — to inform strategic decisions and
capital allocation. Sightline is also the team behind CTVC, the
industryʼs go-to transition innovation briefing read by 75,000+
professionals globally.
Report contributors
Julia Attwood Charles Bondu
Research Director, Research Associate,
Sightline Climate Sightline Climate
Olivia Wang Holly Secon
Research Analyst, Editor,
Sightline Climate Sightline Climate
Contents
Key takeaways ⟶
Data center development 101 ⟶
Powering data centers today ⟶
Powering data centers tomorrow ⟶
Deployment progress ⟶
Case studies* ⟶
Recommendations ⟶
* Only included in the full report available for Sightline clients. To become a Sightline client, talk to our team.
1/7
Key takeaways
Key takeaways
Data center development is now ‘power first’
Data centers are driving the first meaningful growth in electricity demand in developed economies in decades. The
rising strategic and commercial importance of AI has sparked an arms race for computing power, between the big
tech powers and between nations.
Power is now the greatest bottleneck to deploying data centers, and the first priority for developers when choosing
a site. Utilities, grid operators and power producers are caught in the middle — thereʼs an enormous opportunity
here to use data center demand from big tech to accelerate clean power, bolster the grid, and add capacity, but
equally large challenges from interconnection and permitting timelines, community pushback to rising rates, and
the risk of stranded assets.
Novel powering strategies, new tariff structures, and close partnerships between tech and energy players are all
options for overcoming these challenges and accelerating both AI and clean power capacity.
From tracking down case studies, talking to companies in the data center value chain, and applying what we know
of clean firm power development, these are our key takeaways:
Data centers are offering Clean tariffs shield
Power first
generation, grid services, to ratepayers and smooth
development
court local governments access
Power is the biggest constraint for In the US, and congested grids Grid and capacity investments for
data centers. Developers are generally, data centers are offering data centers can raise prices for
prioritizing sites with grid access grid services, additional capacity, or other electricity consumers.
and generation over clusters and local investment to make their Specially designed tariffs for data
fiber. projects more attractive to utilities centers can offer long-term clean
and governments. Itʼs a different firm power, and allow large loads to
story in Europe though, where take on the extra costs.
utilities are still keen to woo new
large loads.
More backup power = Off-grid power gives speed, Utilities and governments
more options but comes with more risk can guide the siting process
Large backup power systems Giving up on the grid allows data Highlighting regions with excess
increase costs, but allow data centers to have more control over generation that could provide
centers to come online before a grid their generation; the source, when power immediately, new generation
connection is approved, offer grid it comes online, and how much they in need of demand, or freed-up grid
services once connected, and give pay for it. But it also requires active connections can pull data centers
added security for outages. management and comes with to less-stressed portions of the
higher outage risks. grid, and deliver economic benefits.
2/7
Data center
development 101
Data center development 101
How data centers get built
There are several factors for a data center developer to consider when choosing a site. In order of
importance:
Power — The main determining factor in data center development today. Finding a short interconnection
queue, or an existing connection with spare capacity, paired with the potential to develop clean power, or
any power, quickly are paramount.
Clusters — Data center operators prefer to expand capacity locally. Siting new data centers near older
ones allows them to share workloads and keep sensitive data close. Experience with local authorities and
utilities also cuts development costs and timelines. New clusters are forming in regions with good grid
access.
Fiber — Reliable, high-speed connections are needed to train AI models and run inference jobs. This is
seen as less of a bottleneck than power, as new fiber can be more quickly rolled out.
Latency — Much less of a concern for data centers training and running AI models than it was for data
centers processing high-speed transactions. The limiting factor for speed is now how long it takes the
model to run, not the data to transfer.
Cost — The largest costs in building a data center are the chips, especially with three to five year
replacement cycles to take advantage of performance improvements. Energy is a small but growing
portion of data center costs, making tech companies relatively insensitive to it when considering sites.
However, if electricity prices rise above $150/MWh, opex could become a greater concern.
Ownership structures
Landowners typically operate within the data center value chain in one of three ways:
● Selling land directly to hyperscalers. Owners may apply for power and grid access in advance,
or simply identify good sites, buy them, and then sell them on to large tech companies and data
center developers. Lightest touch option.
● Building the shell to sell on. Larger landowners, such as real estate developers or private equity
investors may build out more of the infrastructure before selling the site on. This can be attractive
to smaller data center developers, as they become more like middle-men selling capacity in an
existing data center without having to manage the construction and supply chain. These ‘shellsʼ
can be dark (just the warehouse, some permits in progress for power) or powered (also including
racks, electrical infrastructure, and cooling).
● Full ownership with the intention to lease capacity. Real estate developers with more data
center experience could take on the role of a colocation operator, usually in collaboration with a
specialist data center developer, and with input on compute needs from tenants. They build the
structure, source power, and install the computing equipment.
Data center development 101
Roles for owners, tenants, developers
Different types of data center owners and users interact with developers and the value chain slightly
differently:
Big tech (e.g. AWS, Google, Microsoft)
● Typically own and operate the majority of their data center capacity, though this could start to
change as tech companies seek more flexibility in sites and to lower their capex commitments
● Buy land from real estate developers or private landowners and contract out the EPC for the
core infrastructure
● Heavily involved in power procurement, even working with clean firm FOAK providers on PPAs,
or utilities on passing novel tariff structures
● Can work with either large HVAC suppliers or startups to design and implement cooling systems
● Spec and purchase the compute systems from large players like NVIDIA
● Continue to own and operate the data center once itʼs fully operational
Non-tech hyperscalers (e.g. Equinix, Digital Realty, NTT Data, Prometheus Hyperscale, CoreWeave,
Crusoe)
● Typically own and operate their data centers
● Buy land, though potentially with more of the core infrastructure installed than big tech owners
● Can be involved in power procurement (e.g. Prometheus) but more often just take whatever
power is fastest, or is preferred by their tenants
● Contract out the EPC for the core infrastructure, can take instructions from large tenants on
compute requirements, or design it themselves for colocation facilities for smaller players
● Continue to own and operate the data center once itʼs fully operational
Enterprise-focused tech companies (e.g. Yahoo, Oracle)
● Typically lease their data center capacity from third-parties, such as the non-tech
hyperscalers
● Involved in setting the compute requirements and potentially selecting preferred regions for the
site, but not heavily involved in power procurement or site design
● In some cases, they will use a sell to lease model, where they will sell data centers they owned
and built to third party operators, paired with a long-term lease for the capacity
Small AI players, startups
● Lease a portion of a data centerʼs capacity based on how much computing power (measured in
the number of GPUs) their models require
● Usually customers of the non-tech hyperscalers
● No involvement in data center design or operation, effectively a subscription model
Data center development 101
Powering model basics
Data center powering models can be on or off-grid, and use clean or fossil power. This combination determines
their speed.
Grid connected systems have balancing built-in, but can result in long waits for interconnection and the hundreds
of GW of new generation required. They also require carefully structured electricity tariffs to avoid increasing costs
for other ratepayers from upgraded infrastructure. Re-starting recently retired generation can be a shortcut to
many of these challenges.
Off-grid systems are complex and expensive to balance, to ensure the 99.999% uptime data centers strive for,
but can be much faster to build and give the developer much more control over the source of the power. Data
centers can be built in 18 months, renewables projects in 2-3 years with simple permitting regimes, while
interconnection queues can stretch to 5-10 years.
Clean firm power is increasingly the goal for hyperscalers, particularly the big tech companies. Developers are
looking beyond virtual PPAs to hourly matched, dedicated contracts. Additionality is important to them, particularly
if it helps to bring a technology online that could scale and help them later, such as geothermal.
Fossil power is usually the fastest route to powering data centers. There are still natural gas connections available
in many regions, and small gas turbines do not have the wait times and supply chain challenges of larger systems.
Some developers are promising hydrogen-ready turbines or CCS projects to eventually decarbonize the power.
Source: Sightline Climate
Backup power is almost always installed alongside a data center. A small UPS (uninterruptable power system)
ensures that very brief outages or fluctuations in power quality are covered, while backup generation can supply a
portion of the systemʼs power needs while it is fulfilling its demand response obligations, to balance on-site power,
or to run the entire system for months or years while waiting for a grid connection. In this last case, the backup
power system is sized to the expected full system demand.
Data center development 101
Clean power electricity tariff design
In the US, most states have some form of large load electricity tariff that data centers fall under. These were
designed to prevent stranded generation assets and allow grid operators to leverage large chunks of power for
demand response at peak times or in emergencies. They have a few elements in common:
De-risking new generation and infrastructure
● Contracted capacity and duration (anywhere from 2-20 years, sometimes tied to the assetʼs lifetime)
● Exit fees (penalty payments for ending the contract early) and capacity reduction notice periods, typically
for a maximum of 20% of the contracted capacity
Additional services
● Demand response and load shedding obligations at peak times or emergencies
● Investment requirements (e.g. new substation infrastructure, or contributions to funds for low-income
ratepayers to offset additional demand charges)
Managing costs
● Minimum demand and load factor requirements
● Electricity price premium to cover novel clean power, grid upgrades, and infrastructure
More recently, new large load tariffs have been developed to allow data center operators to buy or fund clean
power and grid upgrades. This allows them to access clean firm power managed by utilities and grid operators, but
assume the additional costs, shielding other ratepayers and avoiding community pushback. The Clean Transition
Tariff (CTT) in Nevada is a good example of this, as is the Green Source Advantage Choice program in North
Carolina. Duke Energy is planning its own CTT, known as the Accelerating Clean Energy tariff (ACE). It is currently
pending approval with the regulator.
American utilities are trying to find ways to leverage the enormous demand for power theyʼre seeing from
data centers to improve the grid and increase capacity. European utilities have the opposite problem. They
have excess clean firm power from large resources that are difficult to ramp down (hydropower, nuclear). For this
reason, European electricity tariffs for data centers are typically lower than for other consumers, thanks to tax
breaks (France), network charge reductions (Germany), or the ability to skip interconnection queues if theyʼre
classed as critical infrastructure (UK). Only the most stressed grids are starting to add conditions to new data
center loads. Ireland, one of the first data center hubs in Europe, put a moratorium on new data centers near Dublin
in 2022, and currently has a proposal to require data centers to install matching onsite generation and storage
before being granted a grid connection.
In addition to tariffs, there are also different types of connections:
● Firm service. The ‘traditionalʼ type of grid connection, ensuring 24/7 power. It means that the grid operator
has to be prepared to accommodate this additional demand at peak times.
● Interruptible service. A type of grid connection where the load accepts that at certain times it may not
receive power from the grid. These are tricky for the load to manage, as grid operators often canʼt predict
how long outages will be or how often they will occur. However, they can provide a faster connection, as it
doesnʼt require the utility to build more generation for peak times.
3/7
Powering
data centers today
Powering data centers today
Fast firm power
Most of the options for powering data centers rapidly involve tapping into existing, excess generation, or using a
significant amount of backup generation. In order for sites to come online quickly, backup power must use a fuel
that is separate from the grid. In other words, small turbines running on natural gas or RNG are needed, as batteries
or LDES would still need a large grid connection to charge.
Powering options and case studies
Excess or curtailed generation on managed sites
South Koreaʼs data center campus*
Repurposing an existing grid connection
Homer Cityʼs coal-to-gas switch for data centers*
Backup power as a stopgap for grid connections*
Microsoftʼs RNG microgrid with Enchanted Rock*
xAIʼs Colossus data center*
Microsoftʼs backup for grid services in Ireland*
Off-grid renewables and storage*
Google and Intersect Powerʼs energy parks*
Key for rankings
The powering model or project doesnʼt meet this requirement.
The powering model or project could meet this requirement in the future, or partially does today.
The powering model or project meets this requirement.
Categories:
CLEAN Is the system low-carbon
FIRM Is the power available close to 24/7 with this combination of generation and backup
FAST Is it one of the fastest options for powering data centers
LOW-COST Is it one of the cheapest options
SCALABLE Is this broadly replicable, or does it require specific geographic conditions or infrastructure
ADDITIVE Is this adding clean power to the grid
* Only included in the full report available for Sightline clients. To become a Sightline client, talk to our team.
Powering data centers today
Powering model summary
Most of the near-term options have similar scores, as theyʼre all fast, and theyʼre all not very scalable or additive, as
theyʼre about repurposing whatʼs already there. These models are mostly differentiated on how clean they are, or
how expensive.
Thereʼs much more variation in the long-term options. Most of these are at least partly additive, and almost all are
clean. But they vary from 2 year timeframes (gas without CCS) to potentially over 10 years (advanced nuclear).
Many of them donʼt have firm cost ranges, so could end up being very expensive.
Near-term options
Powering option Clean Firm Fast Low-cost Scalable Additive
Excess or curtailed
generation
Repurposing an existing
grid connection
Backup power as a
stopgap*
Off-grid renewables
and storage*
Long-term options
Powering option Clean Firm Fast Low-cost Scalable Additive
Gas, potentially with
CCS
Commercial-scale
geothermal
Restarting recently
retired nuclear*
Commercial-scale
advanced nuclear*
* Only included in the full report available for Sightline clients. To become a Sightline client, talk to our team.
Powering data centers today
Excess or curtailed generation on managed sites
CLEAN FIRM FAST LOW-COST SCALABLE ADDITIVE
Timeframe
Immediate – 3 years
Overview
Some utilities and governments have identified regions or even specific sites where there are curtailed renewables
or existing large-scale generation like nuclear, hydropower, or coal that is underutilized. Data centers that are able
to scale their demand to whatʼs available (and balance intermittent power) can move onto these sites and receive
power immediately. Local governments and utilities are usually involved to smooth the permitting process.
The ‘idealʼ range for data center electricity costs is around $70-80/MWh. This is the only powering option that
could get close to that, by plugging into the existing system and taking typical industrial electricity prices. Flexible
systems like those being trialled through the DCFlex Initiative could access even lower prices by using curtailed
power.
Sightline take
Where this helps to keep large-scale low carbon power online (e.g. nuclear), or improves the economics of
intermittent renewables, it can be a big benefit to the local grid. It only works with partnerships across the value
chain and government support or coordination, making it hard for the private sector to scale alone.
Powering data centers today
Repurposing an existing grid connection
CLEAN FIRM FAST LOW-COST SCALABLE ADDITIVE
Timeframe
6 months – 2 years
Overview
In regions where interconnection queues are stretching to five to ten years, finding sites with large power plants about
to retire and repurposing their grid connections can be a way to cut down on waiting times. In theory the existing
generation could be kept online with investment in refurbishment and a PPA, but in most cases there are better and
cheaper forms of generation to install. However, if the developer also needs to avoid a long wait for additional
generation from the utility, then an on-site, grid independent form of power like natural gas is the only option.
Data center clusters
Retiring coal plants
Source: EPA, Data Center Map
Source: Sightline Climate, Data Center Map, EIA
There is roughly 60GW of coal capacity due to retire in the US by 2030, mostly in the Midwest and the Southeast. This
doesnʼt match well with current data center clusters, but could become an option for new clusters to link to the Virginia
hub and serve demand for AI on the east coast. How to repurpose these grid connections is a bigger challenge. There
are good wind resources in the region, but this requires large amounts of land. Small gas turbines are a faster and more
compact solution, but decarbonization will be difficult, with little existing CCS or hydrogen infrastructure in the region.
Sightline take
This solves the grid problem, but limits the solutions to the generation problem. With most retiring capacity either
nuclear or coal, most of these sites wouldnʼt have enough space for on-site renewables. Without coordination with the
utility on off-site renewables projects, which could run into the same grid constraints elsewhere, itʼs either worse or not
much better in terms of emissions.
4/7
Powering
data centers
tomorrow
Powering data centers tomorrow
FOAKs for the future
Earlier-stage generation sources can take up to a decade to be deployed, but with tech companies expecting their
AI models to grow in adoption and sophistication, their power needs will only increase. Companies committed to
clean power are looking for sources that can provide firmer power than renewables with much greater energy
densities. That leads them to gas, geothermal, and nuclear.
Powering options
Gas, potentially with carbon capture and storage
Prometheusʼ GW-scale all of the above system*
Funding commercial-scale geothermal
Google, Fervo and NV Energyʼs Clean Transition Tariff*
Restarting recently retired nuclear*
Microsoftʼs deal with the Crane Clean Energy Center*
Funding commercial-scale advanced nuclear*
Amazonʼs future SMR fleet*
Key for rankings
The powering model or project doesnʼt meet this requirement.
The powering model or project could meet this requirement in the future, or partially does today.
The powering model or project meets this requirement.
Categories:
CLEAN Is the system low-carbon
FIRM Is the power available close to 24/7 with this combination of generation and backup
FAST Is it one of the fastest options for powering data centers
LOW-COST Is it one of the cheapest options
SCALABLE Is this broadly replicable, or does it require specific geographic conditions or infrastructure
ADDITIVE Is this adding clean power to the grid
* Only included in the full report available for Sightline clients. To become a Sightline client, talk to our team.
Powering data centers tomorrow
Powering model summary
Most of the near-term options have similar scores, as theyʼre all fast, and theyʼre all not very scalable or additive, as
theyʼre about repurposing whatʼs already there. These models are mostly differentiated on how clean they are, or
how expensive.
Thereʼs much more variation in the long-term options. Most of these are at least partly additive, and almost all are
clean. But they vary from 2 year timeframes (gas without CCS) to potentially over 10 years (advanced nuclear).
Many of them donʼt have firm cost ranges, so they could end up being very expensive.
Near-term options
Powering option Clean Firm Fast Low-cost Scalable Additive
Excess or curtailed
generation
Repurposing an existing
grid connection
Backup power as a
stopgap*
Off-grid renewables
and storage*
Long-term options
Powering option Clean Firm Fast Low-cost Scalable Additive
Gas, potentially with
CCS
Commercial-scale
geothermal
Restarting recently
retired nuclear*
Commercial-scale
advanced nuclear*
* Only included in the full report available for Sightline clients. To become a Sightline client, talk to our team.
Powering data centers tomorrow
Gas, potentially with carbon capture and storage
CLEAN FIRM FAST LOW-COST SCALABLE ADDITIVE
Timeframe
2 – 10 years
Overview
Gas with CCS is being touted as a source of fast, clean firm power for data centers. Exxon, Chevron and Eni have all
announced gas+CCS projects to power data centers. Carbon capture systems can be added to existing or
greenfield gas plants, but need access to CO2 transport and storage in order to be effective. The cost of power can
rise exponentially if new pipelines and wells must be built. It can also impact on timelines: CO2 storage sites can take
more than three years to be approved, and transport networks face lengthy permitting and development processes.
CCS is best used on large baseload power plants, to spread the high fixed capex costs across large amounts of CO2
stored and tax credits claimed. LCOEs for gas+CCS can be close to $100/MWh. Only the largest data centers would
be able to justify their own large-scale gas plants and their own storage site. Even if a developer makes it through all
of these hurdles, the wait for a new gas turbine is currently five to seven years. CCS equipment can also take as long
as two years to be delivered. Small gas turbines could be delivered faster, but would increase the cost of capture.
Data center clusters
Source: Sightline Climate, EPA, Data Center Map. Sightline clients can see more information on gas+CCS here.*
The need for transport and storage infrastructure will limit gas+CCS projects for data centers to Texas, Wyoming,
and the Midwest. Cancelled pipeline projects and long wait times for CO2 storage sites even make the biogenic CO2
hubs in the Midwest a risky place to plan gas+CCS and data centers.
Sightline take
Gas+CCS isnʼt a near term option for powering data centers, except in a few locations (Texas, Illinois, Wyoming)
where there are existing CO2 storage sites and transport networks. Data centers that are flexible on location could
move to suit these resources and use smaller gas turbines, but will still need to wait for capture equipment and
transport networks. CCS is mostly being offered as an olive branch, to show that this power could be decarbonized.
Any large scale capacity would still be close to a FOAK, as thereʼs only been one significant trial of a pilot gas+CCS
plant, at Bellingham. Most data centers leveraging this power will start with unabated gas and punt the CCS unit to
some time in the future, or use carbon credits to cover the emissions.
* For the most current data on gas +CCS projects and costs, access the full database on the Sightline platform. To
become a Sightline client, talk to our team.
Powering data centers tomorrow
Funding commercial-scale geothermal
CLEAN FIRM FAST LOW-COST SCALABLE ADDITIVE
Timeframe
5 – 10 years
Overview
Geothermal power has come back into vogue with startups like Fervo, Eavor, and Quaise promising cheaper
systems that can operate in regions with lower-quality resources. However, these large-scale projects require
long-term offtake agreements to de-risk their demand and allow them to access tens or hundreds of millions of
dollars in financing. Hyperscalers can provide this, and regions with good geothermal resources tend to line up well
with emerging or existing western US data center clusters. However, unless the data centers are sited near to the
plant, it doesnʼt solve the grid challenges.
Data
center
clusters
Next-gen
geothermal
projects
Source: Sightline Climate, Data Center Map
Capacity in development in the US: 860MW
Capacity in development in Europe: 8MW (does not include capacity for heat)
Sightline clients can see our geothermal projects data here.*
Advanced geothermal is a western and southern US story for now, though there are a few projects announced in
Europe. The next generation of data center clusters are already being built here (Wyoming, Nevada), making it a
natural fit as a powering option. Fervoʼs PPA with NV Energy states that it will sell power at $107/MWh. Sightline
expects geothermal to generally be in the range of $77-202/MWh.
Sightline take
While the capacity factor of geothermal power plants depends on the technology and individual resource, itʼs
usually much higher than wind or solar, and therefore a firmer option for data centers. Most offtakers are waiting
for Fervoʼs commercial-scale projects to come online before making the leap, but if these projects are successful,
this could become the powering model of choice for data centers in Nevada, California, Texas and Utah.
* For the most current data on geothermal projects and costs, access the full database on the Sightline platform.
To become a Sightline client, talk to our team.
5/7
Deployment
progress
Deployment progress
The race to deployment*
The competition for AI dominance is a data center race to scale, and major players are already sprinting. Over 130
projects target operation by end-2026, with 58 already under construction. Over 6.7GW of capacity is due online in
the next 18 months, a robust near-term pipeline. Since data centers themselves only take 12-18 months to build,
power availability will be the key factor in determining whether the projects come online on time. More
announcements are expected, and no widespread cancellations have come yet, reflecting strong financial and
policy backing.
Meanwhile, geographic diversification is accelerating, with more activity in Asia and Europe — particularly as US
grids get congested. The US leads with 106 announced sites, while mega-projects in the UK and France signal a
wave of strategic AI infrastructure in Europe and Asia.
Fifteen announced projects exceed 1GW capacity as AI drives demand for hyperscale buildouts. Facilities designed
for cloud, big data, and AI workloads dominate the pipeline. Next-gen centers optimized for AI training and
inference are central to this wave. Five projects top 3GW, with hundreds of billions in investment at stake.
● Microsoft (21 projects), Google (19), Amazon (18) and Meta (12) lead Big Tech buildouts.
● Governments are also scaling AI infrastructure. The US DOE (16 projects) and EUʼs EuroHPC JU (11) are
funding strategic projects to increase their nations' compute capacity and ensure they maintain an edge in
new tech.
● Blackstoneʼs 1GW Northumberland project will increase the UKʼs data center capacity by 63% while
NVIDIAʼs AI Factory will more than double (+145%) data center capacity in France.
Recently commissioned data centers and pipeline by status (MW)*
Source: Sightline Climate. Note that data is current as of 1 July 2025. These numbers are changing daily. Sightline
clients can see our data centers data here.*
* For the most current data on global data center projects and costs, access the full database on the Sightline
platform. To become a Sightline client, talk to our team.
Deployment progress
Project economics: What it costs*
Data centers are a well-established technology and a booming market, but costs still vary wildly. Location,
application, power stack and the developerʼs own place in the value chain all determine the investment required to
build a data center. ‘Merchantʼ developers that build a shell for someone else to power and fill with computing
power will invest much less than a hyperscaler designing an energy microgrid in a developed economy for AI model
development. Either way, billions of dollars will flow into this market over the next ten years.
● Median project investment: $800m, averaging $5.5m/MW for recent announcements.
● AI-specific builds: 5–6x more capital intensive than enterprise or colocation facilities due to GPU costs,
high-density racks, and advanced liquid cooling.
● Retrofits: Offer ~35% cost savings vs. greenfield builds by leveraging existing power interconnections,
water access, and pre-approved zoning.
● Regional variability: Projects in high-wage or dense regions (e.g., Jeollanam-Do, Northumberland) face
significantly higher costs than builds in emerging markets like Johor or Kuala Lumpur.
● Full-stack power vs. grid reliance: Developers building on-site generation bear higher upfront costs but
secure long-term energy reliability, bypassing grid bottlenecks and interconnection delays.
Data center investment vs. capacity (all projects)* Data center investment vs. capacity (projects
under 1 GW capacity and $10bn)*
Source: Sightline Climate. Note that data is current as of 1 July 2025. Sightline clients can see more information
on data centers here.*
* For the most current data on global data center projects and costs, access the full database on the Sightline
platform. To become a Sightline client, talk to our team.
7/7
Recommendations
Recommendations
For data center developers
● Come with your own generation capacity. Utilities facing rising data center
demand have two challenges: building more generation and building more grid
capacity. If data center operators can remove one of these hurdles (generation),
their applications become much more attractive.
● Large backup power systems give much greater flexibility. While this increases
upfront costs, it also gives operators the ability to speed up grid connections,
operate on backup until grid power comes through, and offer flexibility services (and
earn revenues). If speed is a priority, large backup power systems are essential.
● Prioritize regions where utilities have data center experience and creative
tariffs. Utilities that have new data center-specific tariffs approved by their PUCs
and integrated other large loads will be able to provide guidance and ready-made
programs for data centers.
Recommendations
For utilities and power producers
● Proactively highlight regions with excess generation. Streamlined permitting,
substation upgrades, and additional generation already in planning make sites
attractive to data centers. Providing a ‘package dealʼ can bring new demand to
regions that have lost industrial or commercial loads and jobs.
● Make interruptible connections available. Many data centers can accommodate
some interruptions, but utilities need to highlight this option when they receive
applications, and provide detailed data and forecasts for interruptions to allow data
center developers to plan and properly scale their backup systems.
● Shore up the distribution grid. Thereʼs a lot of attention on the transmission grid
and interconnection queues, but greater resiliency and flexibility in the distribution
grid will also be important to incorporating data center loads. Smart substations,
rapid restoration systems and software-controlled circuit breakers could all help.
Recommendations
For policymakers
● Streamline permitting in regions with excess power. Where large generators are
at risk of closing due to declining demand, faster permitting for large loads like data
centers can help to keep them online.
● Mandate high levels of backup power in regions with congested grids. Most
data centers already install significant backup power. By specifying how much of
their total generation must be covered by onsite backup (and ensuring that this
backup power is low-carbon) the effects on the local grids can be mitigated. Watch
to see how data centers in Ireland respond to the governmentʼs proposal on this.
● Encourage long-term data center decarbonization with tax breaks and
incentives. Corporate targets are driving immediate clean power purchasing, but
governments can give data center operators a long term signal by giving tax credits
or preferential access to the grid for companies that commit to building long-term
clean generation.
This is the public preview of the full report
on the Sightline platform. To become a
Sightline Climate client, talk to our expert
team at hello@sightlineclimate.com
About Sightline
Sightline is a tactical intelligence platform purpose-built for
decision-makers looking to stay ahead in the transition. We
provide AI-powered data, tools, and research to help
investors, energy and utility leaders, and banks evaluate
technologies, monitor tipping points, and act with confidence
across sectors in power, grid, fuels, carbon, and industrials.
We work with 80+ leading organizations — including Southern
Company, HSBC, BHP, B Capital Group, Tokyo Gas, and
Energy Impact Partners — to inform strategic decisions and
capital allocation. Sightline is also the team behind CTVC, the
industryʼs go-to transition innovation briefing read by 75,000+
professionals globally.
Ready to
navigate
disruption?
Talk to our expert team
hello@sightlineclimate.com
www.sightlineclimate.com
