—·
The DOE’s PORTS Technology Campus model turns hyperscaler–chipmaker deals into sovereign, power-contracting arrangements with grid risk and site governance embedded.
On March 20, 2026, the U.S. Department of Energy (DOE) announced a public-private partnership to develop the “PORTS Technology Campus” at the former Portsmouth Gaseous Diffusion Plant site in Ohio—an effort built around an expected 10 gigawatts (GW) data center and up to 10 GW of new power generation, including 9.2 GW of natural gas generation. (https://apnews.com/article/4667fa1442ec1c652228337ab4eb68ee)
The headline number is eye-catching, but the real shift is in the deal structure DOE outlined: site control paired with dedicated generation, then a data-center buildout. That combination doesn’t resemble a standard “capacity lease” model. It points toward sovereign compute—where the identity of the party signing power procurement, controlling the critical site, and managing grid-related delays (interconnection) and reductions (curtailment) can matter as much as project finance.
DOE’s announcement is explicit about inviting private partners onto federal lands for AI data center and energy generation projects at selected federal sites, with DOE coordinating through consultation with states, local governments, and federally recognized tribes. (https://www.energy.gov/articles/doe-announces-site-selection-ai-data-center-and-energy-infrastructure-development-federal) In the Portsmouth announcement, DOE also tied the project to Japanese funding associated with natural gas generation components and named participating private sector entities in the power and grid upgrade portion of the work. (https://apnews.com/article/4667fa1442ec1c652228337ab4eb68ee)
For “sovereign compute” governance, the key is the contractual division of responsibilities that standard leasing models often leave blurry. In PORTS-style federal-site development, the governance stack commonly separates into at least four roles: (1) the entity holding site control (or long-term control of land and supporting facilities), (2) the entity procuring generation (and negotiating dispatchability, fuel risk, and offtake terms), (3) the entity responsible for grid and interconnection scope (upgrades, engineering, and energization timing), and (4) the entity that must convert those infrastructure milestones into enforceable IT availability claims to tenants.
In practice, sovereign compute governance becomes less about “who wants AI” and more about “who can credibly guarantee conversion.” When one counterparty controls site-readiness milestones and the power delivery pathway, remedies can track the real bottleneck—interconnection and curtailment—not just uptime. When responsibilities are split across counterparties, remedies often collapse into more generic tools: discretionary liquidated damages, force majeure around grid events, and tariff pass-throughs—outcomes regulators later struggle to translate into “reliability sovereignty.”
That’s also why DOE’s coordination role has strategic weight. When federal processes influence which projects are deemed “ready” for interconnection and which parties gain early access to site footprints, they indirectly shape negotiation leverage—especially whether private partners can secure firm, scheduleable service commitments before they’re locked into interconnection queues or tariff constraints.
So what: Treat PORTS as a signal that future “compute partnership” term sheets will lean less on pure capacity leasing and more on governance documents—contracts that allocate responsibility for proving readiness (site), securing delivery (power), and withstanding grid constraints (interconnection/curtailment) without renegotiation at the moment power is needed.
Classic “capacity leases” often center on data center uptime commitments for recurring fees, while electricity procurement typically blends market purchases and standard utility arrangements. PORTS-style models reorder that ledger. They suggest the partnership will control more of the critical chain: power procurement becomes a form of infrastructure rights—and those rights carry political and regulatory salience.
You can see the same power-binding logic in other hyperscaler-adjacent AI infrastructure deals that describe arrangements as “lease agreements” on paper. Applied Digital, for example, disclosed that it entered lease agreements delivering 250 megawatts (MW) of critical IT load to host CoreWeave’s AI and high-performance computing infrastructure at its Ellendale, North Dakota campus. (https://www.cnbc.com/2025/06/02/applied-digital-coreweave-ai.html) The same deal class is also described in SEC filings as an arrangement in which Applied Digital is dependent on delivering services on time and managing power or other supply disruptions—risks that sit at the reliability-governance layer PORTS is moving into a more governmental frame. (https://www.sec.gov/Archives/edgar/data/2071778/000164117225017704/filename1.htm)
One reason sovereign compute governance can shift economics is that electricity and grid access behave like constrained resources governed by queue-driven delays. In the United States, generator interconnection processes determine how quickly new capacity can connect, and FERC has been reforming the system to reduce delay and speculative queue behavior. In Order No. 2023 and subsequent clarifications, FERC required evidence of 90% site control at the time of interconnection request submission and 100% site control at the time of execution of the facilities study agreement. (https://www.ferc.gov/explainer-interconnection-final-rule)
When governments sponsor or coordinate site control and generation buildout, they can shape who is “ready” to enter interconnection studies. That readiness becomes leverage: power procurement terms start to function as timeline discipline, because converting contracted generation into synchronized grid connection is the gating item.
So what: For hyperscaler partnerships, power procurement clauses increasingly determine whether the project is enforceable on schedule. For chipmakers, it signals a shift from “supply availability” conversations toward “compute availability rights” that can be constrained by grid and interconnection structure.
Two reliability risks often get lumped together in public discussion: grid interconnection delays (when capacity can’t connect because studies and upgrades lag) and curtailment (when connected capacity must be reduced despite being operationally ready). PORTS’s site-control-plus-generation concept points toward a governance solution: contract around grid-imposed risks rather than assuming they wash out at the operating layer.
Interconnection delays produce measurable contract-operational outcomes because interconnection milestones define when a facility can legally and physically synchronize. “Delay” isn’t just inconvenience—it becomes a condition precedent to claiming availability. Under FERC’s interconnection reform framework, the governance lever is early proof and earlier accountability. The rules tighten when interconnection customers must demonstrate site control—90% at request submission and 100% before facilities study agreement execution—paired with queue-efficiency measures intended to reduce speculative capacity hoarding. (https://www.ferc.gov/explainer-interconnection-final-rule)
For contract drafters, the practical significance is direct: the party holding or influencing site control and interconnection milestones is better positioned to write remedies tied to “who caused the milestone to slip.” Without alignment, schedules become unenforceable abstractions—leaving tenants exposed to a provider’s inability to secure synchronized service, and leaving neither party clear fault allocation once system delays arrive.
Curtailment is the later risk—when a connected facility can’t deliver the full contracted output under grid conditions (congestion, voltage constraints, or operational contingencies). Unlike interconnection delays, curtailment turns on operational reality—dispatcher constraints, transmission limitations, and grid contingencies that can emerge even after commissioning. The policy implication of energy-first compute is that curtailment risk must be explicitly priced and governed across the partnership, because it affects the effective availability of AI data centers and therefore accelerator utilization economics.
That same governance lesson shows up in other AI infrastructure announcements that emphasize reliability as something to contract, not hope for. For instance, data center power deals sometimes include behind-the-meter or power-purchase structures tied to “always-on, mission-critical” blocks; when those arrangements are described publicly, it reflects a market belief that reliability must be contracted, not hoped for. A DatacenterDynamics report described LS Power’s plan to sell up to 300 MW from a gas power plant to a planned behind-the-meter data center under a five-year power purchase agreement, with the campus expected to consist of “always-on, mission-critical” data center buildings. (https://www.datacenterdynamics.com/en/news/ls-power-to-sell-300mw-from-virginian-gas-plant-to-be-behind-the-meter-data-center/)
The interconnection reforms and reliability-driven power deals share the same message: if interconnection and curtailment risk aren’t allocated up front, the partnership defaults to whichever party is best able to absorb loss—raising the risk that, in sovereign contexts, the government absorbs costs by accident. PORTS signals a model where government involvement can reduce that likelihood.
So what: Regulators should require that “power-backed” AI compute rights include explicit curtailment and interconnection allocation language, including who triggers schedule remedies and who bears cost when grid constraints reappear.
When power procurement shifts from commodity purchase to infrastructure rights, it becomes sovereignty. That sovereignty isn’t abstract national pride—it’s contract enforceability around time, quantity, and delivery conditions. In PORTS’s case, the announcement highlighted dedicated generation and grid upgrade scope, including new transmission lines and associated upgrade work intended to bring large loads onto the system. (https://apnews.com/article/4667fa1442ec1c652228337ab4eb68ee)
The governance logic is straightforward. AI data centers need consistent operational power because downtime isn’t just inconvenient—it directly affects compute throughput and can cascade into customer service level obligations. When governments embed dedicated generation and grid upgrade scope into a sovereign compute initiative, they also embed themselves into future questions of pricing fairness, rate recovery, and public-interest balancing for critical sites.
This dynamic matters for hyperscaler–chipmaker–government partnerships beyond the United States. “Trusted” compute localization strategies (where compute is located within specified jurisdictions and protected from supply-chain disruption) depend on more than chip manufacturing policy. They also depend on electricity procurement stability and grid readiness—conditions that can become the bottleneck that determines whether compute is actually usable.
For policy readers, the takeaway is to treat power procurement terms as de facto sovereignty clauses. If the power contract includes take-or-pay or firm delivery characteristics, the party holding those clauses gains leverage over operational availability. If grid interconnection is delayed, whoever controls site readiness can claim schedule remedy rights. If curtailment occurs, the contract determines who absorbs energy delivery shortfalls.
So what: When reviewing AI infrastructure partnerships, treat power procurement language as an instrument of sovereignty assessed with the same seriousness as data governance. Ask which entity controls delivery terms, which entity controls remedies, and which entity controls the ability to recover costs from customers or taxpayers.
PORTS is a new anchor point, but sovereign compute governance is already visible in earlier deal structures and regulatory reforms. Four documented examples clarify where hyperscaler partnerships with energy and grid governance are headed.
Outcome: A public-private partnership concept with expected 10 GW data center and up to 10 GW generation, including 9.2 GW natural gas, plus grid upgrade scope and named participating entities.
Timeline signal: Announcement date is March 20, 2026, marking a live initiative rather than a speculative program.
Source: Associated Press reporting on DOE announcement. (https://apnews.com/article/4667fa1442ec1c652228337ab4eb68ee)
Outcome: Disclosure that Applied Digital will deliver 250 MW of critical IT load for CoreWeave’s AI and high-performance computing at Ellendale, with an option for additional capacity described in public coverage.
Timeline signal: Lease deal disclosed publicly in 2025, with investor reaction reported the same day.
Source: CNBC coverage and Applied Digital SEC disclosure context. (https://www.cnbc.com/2025/06/02/applied-digital-coreweave-ai.html, https://www.sec.gov/Archives/edgar/data/2071778/000164117225017704/filename1.htm)
Outcome: A reported plan to sell up to 300 MW from a Virginian gas plant to a planned behind-the-meter data center under a five-year power purchase agreement, with a campus concept framed around mission-critical availability.
Timeline signal: Report published 11 months ago (relative to the search crawl), representing ongoing deal-market movement.
Source: DatacenterDynamics. (https://www.datacenterdynamics.com/en/news/ls-power-to-sell-300mw-from-virginian-gas-plant-to-behind-the-meter-data-center/)
Outcome: Governance reform that requires interconnection customers to provide evidence of 90% site control at interconnection request submission and 100% site control at the facilities study agreement stage.
Timeline signal: Rule issued July 28, 2023, and subsequent clarifications continue to shape how projects structure early commitments.
Source: FERC explainer pages. (https://www.ferc.gov/explainer-interconnection-final-rule, https://www.ferc.gov/explainer-interconnection-final-rule-2023-A)
Taken together, these cases point to a governance shift: even when partners don’t brand deals as “sovereign compute,” the contracting reality increasingly ties compute availability to grid readiness, generation rights, and early site control.
For investors and regulators, interconnection and power purchase obligations should be treated as the core due-diligence objects. The technology supply chain matters, but it’s secondary to whether electricity delivery and grid connection can be contractually secured.
Sovereign compute partnerships will become more explicit as grid constraints and interconnection reforms raise the “proof burden” earlier. FERC’s site control milestones effectively require project proponents to demonstrate physical readiness sooner than older models demanded. (https://www.ferc.gov/explainer-interconnection-final-rule)
At the partnership level, that changes who sits at the table. Utilities and independent power producers are not merely service providers; they can become co-architects of the reliability package, shaping power procurement terms, delivery schedules, and curtailment assumptions. National governments are no longer only grantors or conveners; when they provide site control or sponsor grid upgrades, they influence the enforceability of timelines and the distribution of risk.
For chipmakers, there’s a downstream implication for cross-border supply. “Trusted” compute localization strategies may fail not because chips cannot be shipped, but because electricity and grid readiness in the target jurisdiction can’t support operational throughput. When compute is anchored to physical power infrastructure, the chip supply chain must be synchronized to the energy infrastructure timeline—turning synchronization itself into a policy issue that creates new dependency patterns between industrial policy and power procurement.
Energy-first sovereign compute also raises a governance question: how to prevent power contracts from becoming opaque sovereignty transfer mechanisms that evade normal competition and ratepayer protections. If governments backstop dedicated generation or transmission upgrades, investors will likely seek stronger remedies and firmer delivery obligations, and regulators should expect those contract pressures.
So what: Build an oversight framework now that maps contract rights to public-interest constraints—interconnection responsibility, curtailment allocation, and power procurement transparency must be reviewable before projects scale.
A mature approach should treat energy-first sovereign compute as a regulated partnership category, not a generic infrastructure investment. For the United States, the most direct policy lever lies with agencies that govern energy markets and interconnection, coupled with national security-oriented economic policy bodies.
Concrete recommendation (start within 6 months): The U.S. Federal Energy Regulatory Commission (FERC), in coordination with the DOE and state public utility commissions, should issue a guidance framework for government-supported AI compute projects that explicitly requires contract-level disclosure of (1) interconnection schedule accountability, (2) curtailment allocation and remedies, and (3) how power procurement terms map to availability claims. The objective is to prevent “sovereign compute” from being defined only in press releases while risk remains contractually ambiguous.
This is feasible because FERC already sets enforceable interconnection milestones, including site control evidence requirements. (https://www.ferc.gov/explainer-interconnection-final-rule) The missing step is a partnership-level contract governance template tailored to AI data centers and sovereign compute.
Forward-looking forecast with timeline: By Q4 2026, expect AI infrastructure partnerships in the United States to increasingly adopt “power-backed infrastructure rights” language in project documentation, not just capacity leasing. By Q2 2027, interconnection and power procurement risk allocation clauses should become standard items in regulatory review for large government-influenced projects, because FERC’s site control framework pushes project readiness earlier in the process. (https://www.ferc.gov/explainer-interconnection-final-rule)
In the end, energy-first sovereign compute won’t be decided in a server room—it will be decided in contracts that assign costs when the grid says no and define who earns the schedule when power procurement becomes the real bottleneck.
The next AI infrastructure deal is less about chips than about who finances substations, secures grid access, and absorbs the risk of 24/7 power demand.
Gigawatts of AI compute are being marketed like campus projects. The investment truth is subtations, interconnection queues, and 24/7 power reliability.
Planned US data centers face power delays tied to grid hardware lead times and interconnection limits, forcing hyperscalers and utilities into new PPA and reliability fights.