The Cordova Electric Cooperative has spent nearly a decade building a smart microgrid that serves a city of 2,600 people in remote Alaska

Decision Lens

The Cordova pilot makes a concrete operational argument: surplus renewable generation — currently curtailed — can anchor a data center directly at the generation asset, bypassing the interconnection queue entirely. If DOE’s ARIES platform validates this architecture at scale, the queue-avoidance logic becomes difficult to ignore for greenfield site strategy.

90-Second Brief

Now, a remote Alaskan utility has deployed a modular data center physically inside a run-of-river hydroelectric facility, powered by renewable energy that would otherwise be spilled. The Cordova Electric Cooperative, working with edge data center provider Greensparc, installed 170 kilowatts of computing capacity at the Humpback Creek hydro plant, achieving near-optimal power usage effectiveness by placing servers 20 feet from the generators. In 2025, hydropower supplied more than 83% of Cordova’s electrical demand, with an estimated 20 to 25% of annual hydro output going unused. The U.S.

What’s Actually Happening

The Cordova Electric Cooperative has spent nearly a decade building a smart microgrid that serves a city of 2,600 people in remote Alaska. From 2017 to 2023, DOE investment helped Cordova integrate hydropower and battery energy storage, layering in advanced features including time-stamped phasor measurement, automated load management, and an updated metering network. The result was a highly resilient, low-import energy system validated through a digital twin built by the National Laboratory of the Rockies (NLR) alongside Idaho National Laboratory, Pacific Northwest National Laboratory, and Sandia National Laboratories.

Cordova’s leadership then applied the same localization logic to data. Greensparc, a provider of edge-based modular data centers, deployed 170 kilowatts of GPU-as-a-service computing capacity directly inside the Humpback Creek hydroelectric facility, several miles north of the city. Mountain meltwater provides passive cooling. The server racks sit 20 feet from the generators. PUE approaches 1.0. The entire deployment runs on surplus hydro that the cooperative would otherwise spill.

The energy math is straightforward: hydropower covered more than 83% of Cordova’s total electrical demand in 2025, yet 20 to 25% of annual hydro output goes to waste. That stranded generation now has an absorber. In winter months when river levels drop, diesel supplements the load — a frank acknowledgment that 100% renewable operation is seasonal in this geography.

The DOE Office of Electricity is explicitly connecting this pilot to larger ambitions. The OE Microgrid Program is researching advanced microgrid technologies designed to remove barriers to data center development and deliver reliable, affordable electricity. Using the ARIES platform — which couples data center simulations with real physical hardware — DOE researchers have demonstrated how mixed energy resource portfolios can manage the large, volatile power draws characteristic of AI compute workloads. ARIES also captures hardware degradation signals, including wear on spinning generators, which translates directly into lifecycle cost modeling for operators evaluating behind-the-meter generation assets.

Why It Matters for Global Heads of Data Center Energy?

  • From an operational standpoint, the Cordova model offers a tested blueprint for absorbing curtailed renewable generation at the generation asset itself, eliminating transmission dependency and the interconnection queue timelines that currently constrain expansion in most major markets.

  • From a budgetary standpoint, deploying compute inside an existing hydro or other generation facility can eliminate substation build costs, reduce transformer procurement exposure, and capture surplus energy at near-zero marginal cost — all of which directly compress energy cost per MWh for the workload.

  • From a competitive standpoint, Greensparc’s GPU-as-a-service model embedded in a generation asset is an early commercial signal that modular, co-located compute is moving from concept to contract. Operators who define their own co-location-with-generation strategy now will have more negotiating leverage with developers than those who wait.

  • From a regulatory standpoint, DOE’s active research mandate to “remove barriers to data center development” through microgrid pathways reveals federal policy tailwinds for off-grid or behind-the-meter data center configurations, particularly in underserved or remote jurisdictions.

  • From a workforce standpoint, ARIES-based pre-deployment simulation reduces the engineering risk of novel power configurations — meaning smaller, specialized teams can validate integration before committing capital, easing the technical burden on complex greenfield projects.

The Forward View

Over the next 30 to 90 days, watch for the DOE Office of Electricity to publish further ARIES findings on scaled microgrid-data center architectures, as the OE Microgrid Program has explicitly flagged this research direction. Greensparc’s GPU-as-a-service model will face its first real operational test through Cordova’s 2026 hydro season, producing the first credible performance data on availability, PUE, and diesel supplement rates. Any developer or operator evaluating stranded renewable capacity — hydro spill, wind curtailment, or otherwise — should treat this pilot’s performance results as a direct input to site selection criteria.

Peer Moves

No hyperscaler or major colo peer activity is referenced in the source material. The pilot remains at cooperative and edge-operator scale.

What We’re Uncertain About?

  • Scalability pathway: The Cordova deployment is 170 kilowatts. DOE signals it is exploring megawatt-and-beyond applications, but no specific project pipeline, capacity target, or commercial timeline is disclosed. What resolves it: OE Microgrid Program publications and ARIES demonstration results expected in 2026.

  • Diesel dependency magnitude: Cordova’s CEO acknowledges “probable diesel” in winter without quantifying volume or duration. The true annual carbon profile and energy cost of the facility remains unverifiable until seasonal operating data is published. What resolves it: Cooperative operating reports or Greensparc pilot disclosures following the first full year of operation.

  • Regulatory replicability: The Cordova model operates within a cooperative-owned, isolated microgrid with no ISO or RTO interconnection obligations. Whether the same co-location-with-generation architecture is permissible under FERC jurisdiction or in RTO-connected markets is not addressed. What resolves it: FERC interconnection rule clarification or a comparable pilot in a grid-connected jurisdiction.

  • ARIES commercial access: ARIES is positioned as available to industry partners, but terms, cost, and queue for access are not described in the source. What resolves it: Direct engagement with NLR’s Grid Integration program team.

One Question to Bring to Your Team

Which sites in our current portfolio or development pipeline have curtailed or stranded renewable generation within 10 miles — and have we modeled the economics of absorbing that surplus with a co-located compute load instead of waiting for a grid interconnection upgrade?

Sources

  • Nlr — In Alaska, a Data Center Inside a Power Plant, Inside a Microgrid (Link)

Tags