Data Center Developers Meet Nuclear Operators

In partnership with

Following last week's analysis of grid queue arbitrage and behind-the-meter nuclear strategies, this week we examine the partnership structures that could unlock £142bn ($179bn, €166bn) in value, yet remain largely unexplored.

Whilst France's Data4 signs a 12-year nuclear allocation contract with EDF and Amazon commits £14.3bn ($18bn, €16.7bn) to Talen Energy's Susquehanna plant, the partnership model driving these deals remains fundamentally misunderstood. Most industry observers see power purchase agreements. We see revenue diversification strategies that nuclear operators desperately need and data center developers barely comprehend.

The UAE's Barakah nuclear complex generates 40 TWh annually, powering 25% of the nation's electricity whilst supporting a planned 5GW AI campus. Not through traditional PPAs, but through infrastructure-first planning that treats nuclear and data centers as symbiotic assets from day one. This approach, rare globally, suggests a partnership model the industry keeps reinventing poorly.

Here's the disconnect. Nuclear operators face revenue volatility in deregulated markets whilst sitting on the world's most reliable baseload capacity. Data center developers chase power certainty whilst navigating 5 to 10 year grid connection queues. The solution isn't more PPAs. It's rethinking who owns what, who risks what, and who profits when.

The PPA Structures Nobody Optimises

Talen Energy's restructured agreement with Amazon Web Services reveals the complexity. Originally structured as behind-the-meter co-location, FERC's November 2024 rejection forced a pivot to front-of-the-meter grid-connected supply. The £14.3bn ($18bn, €16.7bn) deal now operates as traditional power purchase, abandoning the cost advantages that made co-location attractive initially.

Exelon and AEP's challenge centred on £111m ($140m, €130m) in annual transmission costs the original structure would avoid. Their argument: AWS receives grid services without paying grid costs. FERC agreed. The revised structure forces both parties into a less economically efficient model because the partnership framework itself lacks sophistication.

France's Data4 and EDF demonstrate an alternative. The Nuclear Production Allocation Contract shares both costs and risks of actual energy produced from EDF's operational fleet. Data4 commits to 40MW over 12 years, approximately 230 GWh annually, beginning 2026. More importantly, the contract structure explicitly allocates production risk between parties. When reactors generate, both parties benefit. When maintenance reduces output, both parties absorb impact proportionally.

Korea Hydro and Nuclear Power's partnership with Amazon, X-energy, and Doosan Enerbility suggests a third model: tri-party developer agreements. KHNP provides operational expertise and existing fleet access. X-energy supplies Xe-100 SMR technology. Amazon guarantees offtake. Doosan manufactures components. The partnership mobilises up to £39.7bn ($50bn, €46.5bn) in combined public and private investment, but more significantly, distributes development risk across four entities rather than forcing binary developer-operator arrangements.

Revenue Diversification for Nuclear Operators

Constellation Energy's June 2025 agreement with Meta demonstrates what revenue diversification actually means for merchant nuclear operators. The 20-year virtual PPA covers 1,092 MW from Clinton Clean Energy Center, but Meta purchases renewable energy credits, not direct power. This structure provides Constellation stable revenue supporting relicensing and operations whilst Meta achieves sustainability targets without physical power delivery complexity.

Compare this to Microsoft's Three Mile Island restart arrangement, priced at £82 ($104, €97) per MWh before tax credits. The premium pricing reflects restart risk and capital requirements. Constellation captures development upside whilst Microsoft locks in 20-year supply certainty. Both parties optimise for different objectives through contract structure, not just pricing.

The operators managing US merchant nuclear fleets, Constellation, Vistra, NRG Energy, and PSEG, collectively control capacity that could serve data center demand for decades. Their challenge isn't technical capability. It's contract structure sophistication. Virtual PPAs, capacity-plus-energy agreements, shared ownership models, and allocation contracts each optimise different risk-return profiles. Most operators default to traditional PPAs because partnership structure innovation remains underdeveloped.

EDF's February 2025 announcement of six sites for data center development, totalling 2GW potential capacity, reveals a fourth revenue stream: land monetisation. Nuclear sites typically occupy substantial acreage with existing grid connections, cooling infrastructure, and security perimeters. Rather than selling power alone, EDF offers integrated site development where data center operators build adjacent to nuclear plants, purchasing power through allocation contracts whilst EDF captures land development value.

Crusoe Energy's expansion into the Middle East through partnerships with Oman Investment Authority and Mubadala Investment Company demonstrates how geography shapes revenue models. Crusoe's energy-first development approach, beginning with stranded or flared gas and expanding toward nuclear, treats power source and compute facility as single integrated asset. Their model: control the energy, monetise the compute. Traditional data center developers reverse this: secure compute customers, then solve power.

Risk Allocation Between Parties

TerraPower's January 2025 memorandum with Sabey Data Centers highlights the risk allocation challenge advanced nuclear introduces. Natrium reactor deployment for data centers requires one party to absorb first-of-a-kind technology risk. TerraPower provides reactor technology and development expertise. Sabey commits to site development and data center operations. The agreement doesn't eliminate technology risk. It explicitly assigns it to TerraPower whilst Sabey assumes market and operational risk for computing infrastructure.

Google's partnership with NextEra to restart the 615 MW Duane Arnold Energy Center by 2029 allocates risk differently. Google provides offtake certainty through a 25-year PPA. NextEra absorbs restart execution risk, including regulatory approvals, refurbishment costs, and schedule. The plant shuttered in 2020. Restarting adds complexity traditional development avoids. NextEra's expertise managing nuclear operations makes them the rational party to own restart risk. Google's scale and power demand make them the rational offtaker. The partnership works because risk allocation matches capability.

The challenge most partnerships fail to optimise: who absorbs queue risk? PJM's interconnection queue contains 2,600 GW awaiting connection. Projects entering today face 5 to 10 year waits. Traditional PPA structures treat queue risk as developer problem. But nuclear operators possess existing interconnection rights worth potentially millions. Sophisticated partnership structures could monetise existing interconnection capacity separately from energy supply, creating a third revenue stream whilst solving developer queue challenges.

Behind-the-meter structures attempted to bypass queue risk entirely by avoiding grid connection. FERC's rejection of expanded BTM arrangements forces the industry toward front-of-the-meter models that reintroduce queue exposure. The regulatory preference for grid-connected supply, driven by cost allocation concerns, pushes partnerships away from economically optimal structures toward politically acceptable ones.

The Tri-Party Model: Utility, Operator, Developer

South Korea's deployment framework, linking KHNP, equipment manufacturers, technology providers, and hyperscale customers, suggests tri-party arrangements might actually require four or five parties for optimal risk distribution. KHNP operates existing reactors and contributes operational excellence. Equipment manufacturers like Doosan manage supply chain and fabrication risk. Technology companies like X-energy provide SMR designs and licensing expertise. Hyperscalers guarantee offtake and provide capital. No single entity absorbs disproportionate risk.

The UK's AI Growth Zones, expected to streamline data center planning and grid access, create opportunity for similar multi-party frameworks. Developers provide compute infrastructure. Grid operators manage connection and transmission. Nuclear operators supply baseload power. The government's National Policy Statement EN-7, anticipated late 2025, will incorporate SMRs and AMRs into planning frameworks, potentially enabling zone-based development where partnerships form around geographic clusters rather than bilateral agreements.

Norway's feasibility study for KHNP i-SMRs in Aure and Heim demonstrates how national energy strategy influences partnership structures. Norwegian grid operators, municipal authorities, KHNP, and potential data center tenants must align on planning, financing, operation, and revenue sharing. The complexity increases, but so does risk distribution. When partnerships span jurisdictions and involve sovereign entities, contract structures must accommodate political risk alongside technical and commercial risk.

France's approach, leveraging EDF's integrated utility model, simplifies partnerships by centralising nuclear operation, transmission access, and site control within one entity. Data center developers negotiate with EDF alone, not separate generator, transmission, and distribution entities. This structural advantage explains why France attracts data center investment despite higher power costs than Nordic alternatives. Partnership simplicity has value.

The Crusoe Insight: Energy-First Development

Crusoe's business model inverts traditional data center development. Instead of identifying compute demand then securing power, Crusoe identifies stranded energy assets then monetises them through computing. Their initial focus on flared natural gas established the approach: take free or negative-cost energy, deploy modular data centers on-site, sell compute services.

Their expansion into the Middle East with backing from Oman Investment Authority and Mubadala targets the region's substantial flared gas volumes. Saudi Arabia alone flares approximately 1.6 billion cubic metres annually. Crusoe's model: partner with oil producers to capture flared gas, build adjacent data centers, share revenue from compute services. The oil producer eliminates a liability. Crusoe acquires free fuel. Both parties create new revenue from previously wasted assets.

Extending this model to nuclear requires similar thinking. Nuclear plants, particularly those in merchant markets, sometimes curtail output during low-demand periods because selling into negative-price markets loses money. Data centers consuming that curtailed capacity convert waste into revenue. The partnership structure: nuclear operator provides discounted power during curtailment periods, data center operator provides flexible load, both parties share margin from compute services during those periods.

Crusoe's announced SMR plans, though details remain limited, suggest intention to own both generation and compute infrastructure. This vertical integration eliminates partnership complexity by consolidating risk and return in one entity. However, it also concentrates capital requirements and restricts scale. Pure partnership models distribute risk but introduce coordination complexity. Crusoe's hybrid approach, owning some generation whilst partnering for other assets, attempts to optimise both.

The Middle East expansion demonstrates geographic arbitrage. Regions with stranded energy assets and limited alternative monetisation options offer better partnership economics than developed markets where generation assets have established revenue streams. Nuclear operators in competitive markets facing revenue pressure might find Crusoe's model more attractive than operators in regulated markets with guaranteed returns.

Investment Implications for Partnership Structures

FTI Consulting tracks over 20 proposed nuclear-data center projects as of February 2025. Each requires partnership structures allocating billions in capital across multiple decades. Current PPA templates, designed for renewable projects with 20-year life cycles and predictable output, fit nuclear's 60 to 80 year operational life and baseload characteristics poorly.

Nuclear projects require different risk allocation across development, construction, operation, and decommissioning phases. Partnership structures should reflect these distinct risk profiles. Development phase: technology and regulatory risk dominate. Construction phase: execution and schedule risk peak. Operation phase: market and operational risk matter most. Decommissioning phase: cost uncertainty and long-term liability risk emerge. Traditional PPAs collapse all phases into single contract terms, misallocating risk and capital cost.

Sophisticated investors should evaluate partnership structures separately from project fundamentals. A strong nuclear project with poor partnership terms may underperform a moderate project with optimised risk allocation. Specific elements to assess: interconnection risk assignment, curtailment revenue sharing, capacity payment structures, decommissioning liability allocation, technology risk ownership, and regulatory change provisions.

The £39.7bn ($50bn, €46.5bn) commitment from the KHNP partnership demonstrates capital scale required for advanced nuclear deployment. No single entity will fund this alone. Partnership structures that efficiently aggregate capital from multiple sources, equipment vendors providing vendor financing, hyperscalers prepaying through take-or-pay agreements, governments offering loan guarantees, will enable deployment. Those requiring single-entity balance sheets will bottleneck.

Geographic arbitrage opportunities favour partnerships that leverage cross-border expertise. KHNP's operational excellence, proven through APR-1400 exports, partners naturally with American or European data center developers lacking nuclear experience. French operators like EDF, managing 56 reactors, possess load-following and operational sophistication that American merchant operators, managing far fewer units, might lack. Partnerships bridging these capability gaps capture arbitrage value.

The Path Forward for Industry Stakeholders

Nuclear operators should develop partnership templates for multiple structures: traditional PPAs, virtual PPAs, allocation contracts, shared ownership models, and energy-first development. Each optimises different scenarios. Forcing all partnerships into PPA frameworks leaves value uncaptured. Operators with merchant exposure benefit most from revenue diversification through innovative partnerships rather than defending traditional utility business models.

Data center developers must build nuclear literacy. Understanding capacity factors, refuelling cycles, regulatory requirements, and operational constraints determines partnership success. Treating nuclear plants as interchangeable power sources misses optimisation opportunities. Developers who invest in nuclear expertise capture better contract terms and identify opportunities others miss.

Regulators face the hardest challenge: balancing innovation against ratepayer protection whilst enabling efficient market structures. FERC's BTM co-location rejection protects transmission cost allocation but constrains economically efficient structures. Future regulatory frameworks should explicitly accommodate nuclear-data center partnerships rather than forcing them into rules designed for different technologies and use cases.

Policymakers should consider zone-based development frameworks similar to UK AI Growth Zones. Clustering data centers near existing or planned nuclear sites reduces interconnection complexity, shares infrastructure costs, and creates ecosystem benefits. France's site-based approach and the UAE's integrated planning demonstrate advantages of treating energy and compute infrastructure holistically rather than separately.

The Bottom Line

The partnership structures supporting nuclear-data center development will determine which projects succeed financially, not just technically. Whilst the industry focuses on SMR technology and safety debates, partnership innovation lags decades behind. Traditional PPA frameworks, designed for different technologies and market structures, inefficiently allocate risk and constrain capital formation.

France's allocation contracts, Korea's multi-party development partnerships, and Crusoe's energy-first model each offer insights. Yet no standard framework has emerged. Each partnership remains bespoke, expensive to structure, and difficult to finance. The operators and developers who standardise effective partnership models, creating templates that efficiently allocate risk whilst attracting capital, will dominate the coming decades.

As one senior executive at a European nuclear operator noted privately: "We spent 18 months negotiating partnership terms that both parties knew were suboptimal. But neither had better templates, and lawyers only know traditional PPAs. We left £79m to £158m ($100m to $200m, €93m to €186m) on the table because we couldn't structure the deal properly."

The question isn't whether nuclear will power data centers. It's whether the partnerships enabling this will capture the value created or waste it on inefficient structures designed for different markets. The £142bn ($179bn, €166bn) at stake demands better answers than traditional PPAs provide.

Next week: We examine Saudi Arabia's nuclear reset: why Vision 2030's success requires Barakah-style capacity by 2030, and what the Kingdom's £397bn ($500bn, €465bn) infrastructure investment means for regional AI development.

You Don’t Need to Be Technical. Just Informed

AI isn’t optional anymore—but coding isn’t required.

The AI Report gives business leaders the edge with daily insights, use cases, and implementation guides across ops, sales, and strategy.

Trusted by professionals at Google, OpenAI, and Microsoft.

👉 Get the newsletter and make smarter AI decisions.

Lead with AI—no coding needed