Power Options for Hyperscale Datacenters

What's Realistic in 2025? An Analysis of Energy Sources from Nuclear to Space-Based Solutions

3x Projected growth in U.S. datacenter power by 2030
460 TWh Additional power demand expected by 2030
22% More grid power needed by end of 2025

The Datacenter Energy Crisis

Hyperscale datacenters are experiencing unprecedented growth, driven primarily by artificial intelligence workloads that require massive amounts of computing power. A typical AI-focused hyperscaler consumes as much electricity annually as 100,000 households, and the larger facilities under construction are expected to use 20 times as much.

⚡ Key Context

  • U.S. datacenter power demand is projected to grow from 50 GW in 2023 to 134 GW by 2030
  • Natural gas currently supplies over 40% of datacenter electricity, followed by renewables (24%), nuclear (20%), and coal (15%)
  • Hyperscalers are signing major power purchase agreements, with Microsoft, Google, and Amazon all announcing ambitious nuclear and renewable energy deals
  • The power crisis is so acute that utilities in some regions cannot meet demand until 2025-2026 while new infrastructure is built

This analysis examines the realistic power options available to hyperscalers today, emerging technologies on the horizon, and more speculative solutions that capture headlines but may not be practical for years—or ever.

Current Datacenter Energy Mix (2024)

Source: International Energy Agency (IEA), Pew Research Center (2024)

Realistic Power Options Available Today

These are proven, commercially available power sources that hyperscalers are deploying right now to meet their energy needs.

✓ DEPLOYED NOW

Natural Gas Generation

Status: Most scalable and immediately available option for meeting demand

Advantages:
  • Highly reliable baseload power (80%+ capacity factor)
  • Fast deployment compared to other baseload sources
  • Lower emissions than coal (50% less CO2 per kWh)
  • Can be used for both grid power and on-site generation
  • Combined-cycle plants achieve 50-60% efficiency
Challenges:
  • Still a fossil fuel with carbon emissions
  • Conflicts with hyperscalers' net-zero commitments
  • Price volatility based on natural gas markets

Real-world deployment: Utilities serving Virginia, Georgia, and Carolina markets have announced 20 GW of new natural gas capacity by 2040, with two-thirds tied to datacenter growth.

✓ DEPLOYED NOW

Grid Power (Mixed Sources)

Status: Standard approach, but increasingly constrained

Advantages:
  • Established infrastructure in key markets
  • No upfront power generation investment needed
  • Can supplement with renewable energy credits (RECs)
  • Most cost-effective for initial deployment
Challenges:
  • Major transmission constraints in high-demand areas
  • 2-5 year interconnection queues in many regions
  • Rising costs as datacenters strain grid capacity
  • Limited control over energy mix/sustainability

Key constraint: In Northern Virginia's "Data Center Alley," power demand could rise from 4 GW to 15 GW by 2030—potentially half the state's total electricity load.

✓ DEPLOYED NOW

Solar & Wind (with Storage)

Status: Widely deployed but requires complementary baseload power

Advantages:
  • Zero emissions during operation
  • Costs continue to decline ($25-26/MWh for solar)
  • Aligns with corporate sustainability goals
  • Can meet 80% of datacenter demand with storage
  • Amazon has financed 500+ solar/wind projects globally
Challenges:
  • Intermittent—solar ~6 hrs/day, wind ~9 hrs/day average
  • Battery storage still expensive and limited duration
  • Requires baseload backup (typically natural gas)
  • Large land footprint requirements
  • Transmission costs for remote installations

Industry consensus: Renewable developers indicate solar and wind can serve roughly 80% of datacenter demand when paired with storage, but baseload generation is still essential for 24/7 reliability.

✓ DEPLOYED NOW

Geothermal Energy

Status: Limited deployment but gaining significant interest

Advantages:
  • Consistent 24/7 baseload power
  • Cheapest form of renewable energy
  • Compact footprint (~11% of coal/solar per GWh)
  • Available nearly anywhere with drilling technology
  • Can be used for both power and cooling
Challenges:
  • High upfront drilling and development costs
  • Location-dependent (best near geothermal hotspots)
  • New drilling technology still scaling
  • Limited current capacity and expertise

Future potential: Google is exploring Clean Transition Tariffs in Nevada for geothermal, and analysts suggest it could meet up to 64% of datacenter demand growth by the early 2030s if datacenters locate in optimal areas.

Nuclear Power: The Emerging Baseload Solution

Nuclear energy has emerged as a focal point for hyperscalers seeking reliable, carbon-free power. Multiple major announcements in 2024-2025 signal a nuclear renaissance driven by datacenter demand.

◐ EMERGING (2-5 Years)

Existing Nuclear Plants (Reactivation)

Status: First plants coming online 2026-2028

Advantages:
  • Proven technology with 93% capacity factor
  • Zero carbon emissions during operation
  • Operates 24/7 regardless of weather
  • 800+ MW capacity per reactor
  • Faster than building new nuclear plants
Challenges:
  • Significant refurbishment costs required
  • Regulatory approval processes (NRC, state, local)
  • Limited number of recently-closed plants available
  • Public perception and political challenges

Major deals: Microsoft signed a 20-year PPA to restart Three Mile Island Unit 1 (835 MW) by 2028; Amazon acquired a 960 MW datacenter at Pennsylvania's Susquehanna nuclear plant.

◐ EMERGING (5-10 Years)

Small Modular Reactors (SMRs)

Status: At least 5 years from commercial operation in the U.S.

Advantages:
  • Scalable designs (20-300 MW modules)
  • Factory-built, potentially faster deployment
  • Enhanced safety features
  • Can be built closer to transmission lines
  • Modular—add capacity as needed
Challenges:
  • No commercial SMRs operating in U.S. yet
  • First planned U.S. SMR was canceled in 2023 due to costs
  • Regulatory pathway still being established
  • Unproven economics at scale
  • Nuclear fuel supply constraints (especially HALEU)

Industry investment: Google and Amazon announced investments in SMR startups (Kairos, X-Energy) in October 2024. Tech companies have signed contracts for 10+ GW of potential SMR capacity, though successful development remains uncertain.

Projected Nuclear Capacity for Datacenters

Source: Goldman Sachs Research, IEEE Spectrum, IEA (2024-2025)

Backup Power Solutions

While primary power sources keep datacenters running, backup systems ensure 99.999% uptime ("five nines") during grid failures. Traditional diesel generators dominate, but alternatives are emerging.

✓ STANDARD TODAY

Diesel Generators

Status: 95% of operators still rely on diesel backup

Advantages:
  • Proven reliability over decades
  • Fast startup (~1 minute to full load)
  • High energy density fuel storage
  • Well-understood maintenance procedures
  • Cost-effective (~$1/W capital cost)
Challenges:
  • Significant carbon emissions when operating
  • Increasingly difficult air permit approvals
  • Conflicts with carbon-negative goals
  • Fuel degradation over time (24-month usability)

Industry trend: Microsoft has committed to eliminate diesel by 2030 as part of its carbon-negative pledge, spurring interest in alternatives.

✓ AVAILABLE NOW

Natural Gas Generators

Status: Growing alternative to diesel for backup power

Advantages:
  • Burns cleaner than diesel (lower CO2, NOx, SOx)
  • Fuel delivered via pipeline (always available)
  • Can monetize excess capacity (sell to grid)
  • Easier air permit approvals than diesel
  • Less noise pollution
Challenges:
  • Still a fossil fuel with emissions
  • Requires natural gas pipeline access
  • More complex than diesel systems

Emerging use case: Some datacenters are using natural gas for both primary and backup power, with Combined Heat and Power (CHP) systems achieving up to 80% efficiency.

◐ EMERGING (1-5 Years)

Hydrogen Fuel Cells

Status: In testing phase, approaching cost parity with diesel

Advantages:
  • Zero emissions (only water vapor produced)
  • Scalable to megawatt requirements
  • Quiet operation
  • Can potentially provide grid services
  • Approaching cost parity in 3-5 years
Challenges:
  • Slower startup than diesel (~7 minutes for PEM)
  • Green hydrogen supply chain still developing
  • Higher upfront capital costs currently
  • Hydrogen storage and distribution infrastructure needed
  • Some fuel cell types require 100% utilization for efficiency

Testing progress: Microsoft successfully demonstrated 3 MW hydrogen fuel cell backup systems in 2023-2024. Bloom Energy has announced plans to deploy 1 GW of fuel cells at datacenter sites as interim backup while grid infrastructure expands.

✓ DEPLOYED NOW

Battery Energy Storage (BESS)

Status: Used for UPS and short-duration backup

Advantages:
  • Instantaneous power delivery (UPS function)
  • Zero emissions during operation
  • Fast response to load changes
  • Declining costs (lithium-ion)
Challenges:
  • Limited duration (minutes to hours, not days)
  • Prohibitively expensive for 100-1000 MW sustained loads
  • Large physical footprint for multi-hour backup
  • Not viable alone for datacenter-scale extended outages

Current role: Batteries are critical for UPS systems that provide instantaneous power while backup generators spin up, but aren't economical for the sustained multi-day backup datacenters require.

Coal Power: A Temporary Extension, Not a Solution

🔥 The Coal Question

Headlines vs. Reality: Coal has appeared in news as a datacenter power source, leading to concerns. However, the reality is more nuanced—and the outlook is clear that coal is not the future.

Current Status (2024-2025): Coal generation for datacenters has increased nearly 20% year-to-date, currently supplying about 15% of U.S. datacenter electricity. Some coal plants scheduled for closure have been kept online or closure dates delayed.

Why Coal Is Extending (Temporarily)

Short-term factors:
  • Urgent power needs outpacing new clean energy construction
  • Existing coal plants already grid-connected (no 2-5 year queue)
  • Natural gas price volatility making coal economically competitive in some markets
  • Utilities facing massive projected demand (20%+ load growth by 2035)
Why it won't last:
  • Coal plants built in 1970s-1980s, facing increasing unplanned outages
  • Capacity accreditation only 83% (vs. 93% for nuclear)
  • Some plants like Montana's Colstrip operate at only 54% availability
  • High maintenance costs as aging infrastructure degrades
  • Hyperscalers' carbon-negative commitments incompatible with long-term coal use
  • IEA projects absolute decline in coal-fired generation for datacenters by 2035

The Real Trend: Old coal plant sites are being repurposed—not kept as coal plants, but converted to other generation sources. Pennsylvania's Homer City coal plant (closed 2023) is being transformed into a 4.5 GW natural gas-powered datacenter campus, opening in 2027. This pattern of "coal site reuse" rather than "coal continuation" is the realistic path forward.

Bottom Line: Coal is a stop-gap measure that addresses short-term grid constraints, not a strategic power solution. The combination of aging infrastructure, unreliability, and corporate sustainability commitments means coal's role will diminish after 2030 as nuclear, natural gas with carbon capture, and advanced renewables scale up.

Space-Based Datacenters: Science Fiction or Future Reality?

The idea of placing datacenters in orbit has captured significant media attention in 2025, with startups raising millions and tech giants conducting feasibility studies. But how realistic is this option?

⚠ HIGHLY EXPERIMENTAL (10+ Years)

Orbital Datacenters

Status: First small-scale demonstrations planned 2025-2026; commercial viability uncertain

Theoretical Advantages:
  • 24/7 solar power availability (no night, no clouds)
  • Radiative cooling in vacuum of space (no water needed)
  • No land use on Earth
  • 8x more solar power per year vs. mid-latitude Earth panels
  • Potential 10x lower CO2 emissions over lifecycle
  • Unlimited expansion space
Fundamental Challenges:
  • Launch costs: $8+ million per satellite (Starcloud estimate), hundreds of launches needed
  • Latency: Light-speed delays rule out time-sensitive applications (financial transactions, real-time services)
  • Radiation: Solar particles degrade electronics faster than on Earth
  • Maintenance: Physical repairs extremely difficult/impossible
  • Data transfer: Optical links show promise but unproven at scale
  • Component failure: No proven longevity for datacenter hardware in space environment
  • Economics: Unclear if costs will ever be competitive with terrestrial options
Current Projects:
  • Starcloud (formerly Lumen Orbit): Planning 5 GW orbital datacenter; first demonstration satellite with NVIDIA H100 GPUs planned late 2025
  • Google: Published research paper (Nov 2025) on launching AI chips into sun-synchronous orbit; ran radiation tests at UC Davis
  • EU ASCEND Study: Thales Alenia Space study found orbital datacenters could reduce energy/carbon vs. Earth, targeting 50 kW proof-of-concept by 2031, 1 GW by 2050
  • Axiom Space: Partnering with Skyloom for world's first orbital datacenter prototype
Realistic Assessment:

Best-case scenario: Niche applications emerge by 2030s for specific workloads like Earth observation data processing, satellite AI inference, or batch processing that's latency-tolerant. Primary use would be processing data already in space rather than serving Earth-based users.

Most likely: Demonstrations succeed technically but economics don't work out vs. rapidly improving terrestrial options (cheaper SMRs, advanced geothermal, fusion eventually). Remains a research curiosity rather than mainstream solution.

Quote from industry: "In 10 years, nearly all new data centers will be being built in outer space," predicts Starcloud CEO Philip Johnston. However, most datacenter industry experts view this timeline as highly optimistic given the fundamental challenges.

Verdict: Space-based datacenters are fascinating from an engineering perspective and may find niche applications, but they are not a realistic solution for the massive hyperscaler power needs of the 2025-2035 timeframe. The core datacenter energy challenge will be solved on Earth.

Power Source Comparison Matrix

Power Source Availability Today 24/7 Reliability Zero Carbon Cost Competitive Scalability Deployment Speed
Grid Power (Mixed) ~ (2-5yr queue)
Natural Gas
Solar + Wind (needs backup) ~
Geothermal ~ (limited) ~ ~
Nuclear (Existing) ~ (limited plants) ~ (3-5 years)
SMRs (uncertain) (5-10 years)
Coal ~ (declining) (aging fleet) ~
Hydrogen Fuel Cells (testing) (3-5 yrs to parity) ~
Space-Based (10+ years)
Legend: ✓ = Yes/Strong | ~ = Partial/Moderate | ✗ = No/Weak
Analysis based on industry reports from Goldman Sachs, IEA, McKinsey, S&P Global, and company announcements (2024-2025)

Realistic Power Timeline: 2025-2035

What hyperscalers will actually be using to power their datacenters over the next decade:

2025-2027
(NOW)

Bridging the Gap

Primary sources: Grid power (mixed), natural gas, solar/wind with RECs

Reality: Hyperscalers scrambling to secure any available power. Natural gas becomes the "necessary evil" to meet immediate demand while waiting for cleaner baseload. Some coal plants extended temporarily. Interconnection queues remain 2-5 years in key markets.

Key developments: First hydrogen fuel cell demonstrations for backup power; Microsoft's Three Mile Island restart announced; Google signs first SMR agreements.

2028-2030

Nuclear Renaissance Begins

Primary sources: Natural gas (40-45%), renewables (25-30%), nuclear (15-20%)

Reality: First reactivated nuclear plants come online (Three Mile Island, Duane Arnold). Solar and wind deployments accelerate with improved storage. Natural gas still dominant but percentage declining. Hydrogen fuel cells achieve cost parity with diesel for backup power.

Key developments: Geothermal projects gain momentum; first commercial SMR site prep begins (though operation still years away); coal generation for datacenters peaks then begins decline.

2031-2033

Diversification Era

Primary sources: Natural gas (35%), renewables (30-35%), nuclear (20-25%), geothermal (5-10%)

Reality: First Small Modular Reactors enter commercial operation. Advanced geothermal provides baseload in specific regions. Battery storage improves significantly but still not viable alone for multi-day datacenter backup. Natural gas plants increasingly paired with carbon capture technology.

Key developments: Coal generation for datacenters enters absolute decline; hydrogen supply chain matures enabling fuel cell deployments; datacenter locations increasingly chosen based on clean power availability rather than just connectivity.

2034-2035

Clean Baseload Maturity

Primary sources: Nuclear (30-35%), renewables (35-40%), natural gas (20-25%), geothermal (10%)

Reality: Multiple SMRs operational across the U.S.; nuclear provides reliable clean baseload complemented by renewables and storage. Natural gas still essential but increasingly with carbon capture. Geothermal serves 10% of demand in optimal regions. Hydrogen fuel cells standard for backup power.

Key developments: Hyperscalers approach carbon neutrality goals; coal essentially eliminated from datacenter power mix; space-based datacenter demonstrations may occur but remain niche curiosity rather than practical solution.

Key Takeaways: The Realistic Path Forward

💡 What's Actually Going to Happen

  • No silver bullet exists. Hyperscalers will use a portfolio approach combining multiple power sources based on location, availability, and cost.
  • Natural gas is the bridge. Like it or not, natural gas will power the datacenter boom through 2030 while cleaner baseload sources scale up. It's the only option that's both scalable and immediately available.
  • Nuclear is the long-term answer for clean baseload. The industry has decisively shifted toward nuclear for 24/7 carbon-free power. Reactivated plants arrive 2026-2028; SMRs follow in the 2030s.
  • Renewables are essential but insufficient alone. Solar and wind can provide 80% of demand with storage, but that last 20% requires dispatchable baseload—likely nuclear or gas with carbon capture.
  • Coal is a footnote, not a chapter. Despite headlines, coal's role is temporary stopgap, not strategic direction. Aging infrastructure and sustainability commitments ensure its decline by 2030.
  • Space datacenters are not the answer—at least not this decade. Fascinating technology demonstration, but fundamental challenges (latency, cost, maintenance) make terrestrial solutions far more practical for the foreseeable future.
  • Location, location, location. Datacenter siting is shifting from "near fiber" to "near power." Expect growth in regions with available power capacity, renewable resources, or nuclear plants.
  • Backup power is evolving. Hydrogen fuel cells will replace diesel over the next 5-10 years as costs fall and supply chains develop, enabling true zero-emission operations.
  • The speed bottleneck is real. Even with unlimited capital, physical infrastructure (transmission lines, nuclear plants, pipelines) takes years to build. The 2025-2027 period will be the most constrained.

The Bottom Line

The realistic power mix for hyperscale datacenters in 2030 will be approximately:

This isn't what environmental advocates want to hear, and it's not what nuclear enthusiasts hoped for, but it's what the physics, economics, and timelines dictate. The path to truly clean datacenter power is a marathon, not a sprint—and we're still in the early miles.