GEIOS GEOFLEX and Aiden Digital Labs Unveil the Dark Token Factory: Geothermal AI Compute on Idle Oil and Gas Wells

Lights-out, robotically serviced compute nodes generate native DC power and drive their own cooling from the well, with no grid connection and no fuel

HOUSTON, TX, UNITED STATES, November 13, 2026 /EINPresswire.com/ — GEIOS GEOFLEX, the well-repurposing branch of GEIOS Technologies, and data center design firm Aiden Digital Labs, partners under a memorandum of agreement, today unveiled the Dark Token Factory (DTF), which the partners announce to be the first autonomous lights-out AI token factories built on abandoned and orphan oil and gas wells: a distributed network of sealed compute facilities powered by the geothermal heat beneath them. Each node follows the “dark factory” model: sealed, unlit halls serviced by robots, with no grid connection and no fuel. (The “tokens” in the name are the units in which AI output is metered.) A first pilot node is targeted for early 2027 on an existing well.

The launch comes as power, rather than chips, has become the main constraint on AI capacity. New U.S. data center loads of 10 to 100 MW typically wait four to seven years for a grid connection, and 583 local and state restrictions on data centers are now active across 46 states. In Congress, the proposed Artificial Intelligence Data Center Moratorium Act (S.4214), introduced in March 2026 with a House companion in June, would halt construction of AI data centers above 20 MW until national safeguards are in place. On-site gas, the industry’s main workaround, faces its own bottleneck, with turbine delivery slots quoted into 2030 and beyond.

The DTF approaches the problem from the other direction by taking compute to where firm energy already exists. The U.S. EPA estimates that the country has roughly 3.7 to 3.9 million abandoned or non-producing oil and gas wells, many of which reach rock above 150 °C. Unlike new geothermal prospects, these wells have already been drilled and logged, with subsurface temperatures documented in existing well records, which sharply reduces exploration risk. The sites also come with roads, cleared pads and a wellbore, and because a DTF node re-enters an existing well, the work is permitted by the state oil-and-gas regulator rather than routed through a utility queue, shortening the path from well to compute.

At each site, GEIOS open-close loop Nano-Geocasing lifts heat through a sealed circuit, producing no reservoir fluid and using no water fracturing. A heat-driven power generator converts the heat into native direct current, designed with battery storage to keep power stable for AI computing. Power flows straight to the compute halls, immersion-cooled with a PFAS-free fluid, without the conversion stages of a grid-fed facility, and the same geothermal resource also supports cooling through redundant systems built to a Tier III+ design basis. Each module is designed for high energy efficiency and near-zero water use.

Inside the sealed halls, a rail-mounted robotic gantry guided by machine vision removes and replaces 8-GPU server sleds without anyone entering. Field crews maintain the surface plant and the well, while regional operations centers monitor every node around the clock. Because no one works inside the halls, nodes can operate on remote land where the wells are. Each pad is phased from 10 MW to a maximum of 20 MW, below the thresholds targeted by current moratorium proposals, and nodes linked by fiber operate as one distributed network.

Every other route to new AI capacity carries a trade-off: grid campuses face multi-year queues, gas depends on fuel and scarce turbines, and small modular reactors are not expected before around 2030. The DTF combines reused wellbores, direct-DC geothermal power and sealed robotic operation in a single system, with a planned deployment time of roughly 10 to 14 months per node.

The platform offers three classes of compute, based on NVIDIA H100 and H200, AMD Instinct MI350X and NVIDIA B200 accelerators, in the same module behind a single OpenAI-compatible interface. Data center operators, hyperscalers and GPU cloud providers can route workloads to DTF nodes over private fiber, without adding load to their own sites, and resell that capacity as Green Premium Tokens with a per-batch record of energy source and water use.

Each conversion also turns an environmental liability into an asset. Abandoned U.S. oil and gas wells leak methane and carbon dioxide, with methane alone estimated by the EPA at 303,000 tonnes in 2022. Rather than leaving a well to leak or be plugged at public expense, every DTF re-entry measures its leak rate, seals the wellbore against methane and CO₂ escape and verifies the result, putting the well back to work without consuming water.

“Data centers spent decades chasing the grid, and when the grid said no, the industry turned to gas and traded one queue for another,” said Stephane Mercier, CMO of GEIOS Technologies. “We are doing the opposite: putting compute where the energy already is and letting the same heat that makes the power run the cooling. We know of no other design that combines geothermal power, reused wells and fully unmanned operation. America has millions of non-producing wells; we see each one as a potential address for AI compute.”

GEIOS geothermal technology has been validated in the laboratory, and its key components have completed preliminary field tests. System-level field trials are now under way at three projects, two of them on existing wells. The facility design by Aiden Digital Labs is complete, and its OEM partner has begun fabricating structural components for a test module, including a small server room for extended run testing of the AI compute systems. The first DTF node will then move into its early-2027 pilot, where the lights-out operating model will be demonstrated through a structured commissioning program, including a continuous 168-hour run, ahead of commercial operation.

Companies will access DTF capacity through a dedicated API built for green compute, using the network for AI inference, model training, fine-tuning and other compute workloads. Ahead of commissioning, they can reserve tomorrow’s compute today by registering a non-binding letter of intent at www.darktokenfactory.com/loi, giving them priority access as capacity comes online. Owners of idle wells can also submit them for screening.

About GEIOS GEOFLEX
GEOFLEX a Texas corporation based in Greentown Lab Houston, develops technology that convert subsurface heat into firm, locally delivered power.

Ellica Mc Pherson
GEIOS, GEIOS GEOFLEX and AIDEN DIGITAL LABS
+1 651-317-7896
contact@geios.energy
Visit us on social media:
X

Dark Token Factory Ad

Media gallery