AI-only data centers · Alaska

Compute, priced energy-first.

We build AI data centers where power is abundant and cooling is nearly free: contracted Alaska natural gas, generated on site, and cooled by Arctic air.

Talk to us Read the thesis
~1.10Design PUE target, against an industry average of about 1.54Uptime Institute 20253 · target, not measured
50MWFirst campus, IT load, built for AI training and inferenceTarget
3–4Alaska sites planned across the platformCorridors under evaluation
~7.7Bcf/dProduced gas reinjected at Prudhoe Bay, largely for lack of a pipeline to marketU.S. DOE2

01 Thesis

Energy is the cost of compute. So start with energy.

Most data centers are sited near users and then go looking for power. We do the reverse. We start where energy is stranded and the air is cold, and we price every megawatt of compute from there.

Where the power goes

Facility overhead for every 100 MW of IT load

Industry averagePUE ~1.54
Our design targetPUE ~1.10
Most overhead is cooling. Arctic free air removes most of it, so more of every megawatt reaches the chips. Industry average from Uptime Institute (2025).3 Our figure is a design target, not a measured result.
A

Priced energy-first

Data centers already use about 1.5% of the world's electricity, and that share is growing.1 We treat energy as the line item that sets the price of compute, and we control it.

B

Arctic cooling is the edge

Cooling is the biggest overhead in a data center. Outside-air economization in the Arctic cuts the energy spent on chillers, so more power goes to GPUs.

C

No waiting on the grid

We buy natural gas under contract and generate power behind the meter, next to the load. No transmission line to build and no interconnection queue to wait in.

02 Platform

One stack, from gas to GPU.

We control each layer that sets the cost of compute: the fuel, the power plant, the building and the capacity on top. Fewer handoffs, fewer margins stacked on one another.

  1. 03Capacity

    Capacity, then cloud

    Power and space first: reserved capacity for hyperscalers, Compute Vaults for venture funds and direct AI colocation. Over time, Arctic Cloud: contracted GPU clusters on our own power.

    • Reserved capacity
    • Compute Vaults
    • Colocation
    • Arctic Cloud coming
  2. 02Facilities

    Arctic Data Centers

    AI-only campuses designed around Arctic air. A 50 MW first campus, then a multi-site Alaska platform of three to four sites.

    • 50 MW first campus
    • Free-air cooling
    • Liquid-ready halls
  3. 01Energy

    Arctic Power

    Gas-to-power. We contract Alaska natural gas and turn it into electricity on site, behind the meter, sized to the load it serves.

    • Contracted gas
    • On-site generation
    • No grid queue

Energy enters at the bottom of the stack. Every layer above it is priced from that cost.

03 Offerings

Three ways to use Arctic capacity.

Power and space, built for AI and HPC only. Each offering is for workloads that care more about cost per megawatt than distance to end users.

01

Reserved capacity for hyperscalers

Backup and overflow capacity, reserved in advance, for training runs, fine-tuning and batch inference: work that needs power, not proximity. We plan to offer up to about half of our capacity this way.

  • Committed, reserved capacity
  • Latency-tolerant AI workloads
  • A second region outside power-constrained markets
02New model

Compute Vaults for VCs

Self-storage for compute. A venture fund owns dedicated capacity on our campus and assigns it to its portfolio startups. If a startup fails, the fund backfills the space with another one.

  • Fund-owned, dedicated capacity
  • Startups lease from their investor
  • Capacity stays in the portfolio
03

AI colocation

Power, cooling, space and connectivity for teams that bring their own GPUs and servers, priced from our energy cost. That includes a home for GPU fleets from stalled projects that still need power and cooling.

  • Racks and modular halls
  • Built for dense GPU loads
  • A landing spot for stranded GPU fleets

How a Compute Vault works

Startups burn their raise on compute. A vault keeps that value with the fund.

The investor owns the capacity and the portfolio uses it. The legal structure, including which entity holds the hardware, is still being designed.

  1. The fund reserves a vaultIt commits to dedicated capacity on our campus.
  2. Startups lease from itPortfolio companies run on the fund's vault instead of paying cloud rates.
  3. Winners buy inA startup that scales can buy its allocation and own its compute.
  4. Backfill if one failsThe fund puts another startup in the same space. The capacity keeps working.

04 Sites

A multi-site Alaska platform.

We plan three to four sites where gas, roads and fiber meet. Three corridors are under evaluation today; we don't own or lease any of them yet. Alaska's main utilities have said they can't serve a large data center today,5 which is why we generate our own power.

Map of Alaska showing three target corridors Alaska with markers at the Deadhorse and Dalton Highway corridor on the North Slope, Kenai to Nikiski, and Anchor Point on the Kenai Peninsula, with an inset of Cook Inlet. Fairbanks and Anchorage are shown for reference. Corridors are approximate. CANADAArctic OceanBering SeaGulf of Alaska Fairbanks Anchorage 1 Deadhorse Dalton Hwy + fiber COOK INLET Anchorage 2 Kenai–Nikiski 3 Anchor Point Dalton Hwy / fiber corridor Parks Hwy Target corridor
Approximate locations. Markers show target corridors, not owned or leased land. Blue dashes: Dalton Highway and fiber corridor; gray dots: Parks Highway (simplified). Base map: Natural Earth.

1North Slope · Lead corridor

Deadhorse / Dalton Highway

Along the Dalton Highway from Franklin Bluffs to Deadhorse, next to the largest pool of stranded gas in the state.

  • GasPrudhoe Bay area produced gas, bought under contract (to be negotiated)
  • AccessDalton Highway and Deadhorse airport
  • FiberQuintillion's network runs along the corridor
  • PrecedentThe state issued a preliminary decision in 2026 to lease state land for a gas-powered, behind-the-meter data center here4

2Cook Inlet

Kenai–Nikiski

An established industrial corridor on the Kenai Peninsula with infrastructure already in place.

  • InfrastructureIndustrial brownfields, deep-water port, Railbelt grid
  • GasCook Inlet producers today; the planned Alaska LNG pipeline terminus over time
  • DiligenceFirm gas in a tight Cook Inlet market

3Kenai Peninsula

Anchor Point

Onshore Cook Inlet gas near the Sterling Highway, where new gas development has been planned.

  • GasA nearby operator has announced a new well targeting about 25 MMcf/d (operator's plan, not yet proven)6
  • AccessSterling Highway
  • DiligenceGas contracts, land, fiber and competing utility demand

05 Roadmap

Lock the inputs. Then build at scale.

Each phase depends on the one before it. Timelines and sizes are targets.

  1. Phase 0 · Now

    Secure the inputs

    • Gas supply agreements
    • Land and site control
    • Fiber capacity
    • Permits: state land, air, wetlands
    • Engineering design and cost estimate
  2. Phase 1 · Target

    50 MW first campus

    • On-site gas generation, behind the meter
    • AI-only halls designed for free-air cooling, liquid-ready
    • Target: about 18 months from start of construction to first power
  3. Phase 2 · Target

    The platform

    • Expansion on the first campus
    • Three to four Alaska sites
    • Arctic Cloud capacity on our own power

06 Team

Builders who've been thinking about this for years.

Brad Robinson

Brad Robinson

Co-founder & CEO

Formerly at Bernstein and co-founder of Zevv and Soho Sheds. Leads the company, capital and partnerships.

Rory Greenberg

Rory Greenberg

Co-founder, President & COO

Leads operations, construction and site delivery. Data-center developer behind Project Growler, a proposed 600,000 sq ft data-center campus in Indiantown, Florida, with hands-on experience in site development, permitting and community engagement.

Max Crespo

Max Crespo

Co-founder & Head of Development

Originated the energy-first compute thesis and the Compute Vault model. Leads development: sites, gas supply, permitting, partners, and customer and VC relationships.

Drew Popkin

Drew Popkin

Co-founder & Head of Real Estate Acquisitions

Co-founder and Principal of Highpoint Property Group, a New York real estate developer and owner focused on multifamily and mixed-use assets. Previously Head of Acquisitions at Naftali Group. Leads land and site acquisition.

Talk to us

Hyperscaler, fund, AI team or gas partner?

We're talking with capacity buyers, venture funds, gas producers, builders and investors now. Tell us what you need.

  • Reserved capacity
  • Compute Vaults
  • AI colocation
  • Stranded GPU fleets
  • Gas supply

We use your details only to reply. No cookies, no tracking.

Prefer email? contact@arcticdatacenters.com