In development
2 MW-CLASS POWER BLOCK

Gas and fiber in, tokens out.

Firm on-site power for AI where the grid cannot connect on time. One 2 MW-class industrial gas turbine with an ACTE recuperator, a battery and exhaust-driven cooling, in four containers and a gas skid. It powers and cools 8 liquid-cooled AI racks or one NVIDIA DSX POD, with no grid connection.

Power block beside an AI pod on a gravel pad, conceptual 3D modelConceptual 3D model
1.4–1.6 MWenet electrical power per block
1.12–1.26 MWof IT: 8 AI racks or one DSX POD
~20 daysfrom stock to switch-on, on a ready site
No gridconnection or interconnection queue

Design targets per block, natural gas, sea level; the recuperated version will be validated on the pilot block.

THE BOTTLENECK

GPUs are waiting for power. Waiting is the expensive part.

4–7 years

to connect to the grid in the primary data-center markets

JLL, Jan 2026
18–36 months

for small off-the-shelf gas turbines

EPRI, Mar 2026
12 → 33 GW

short in US data centers in 2027 and 2028, even counting 19 GW of on-site gas generation

Morgan Stanley, Sep 2026

The chips exist; power on a date does not. A block that switches on in days or months, instead of years, puts waiting racks to work.

THE BLOCK

Four containers and a gas skid.

Each module does one job and travels as standard ISO freight. The whole block sits on a site of about 23 × 18 m.

Power block with four containers, roof-mounted recuperator, dry cooler and cooling tower, conceptual 3D modelConceptual 3D model
Power container with the recuperator skid and stack on its roof, conceptual 3D modelConceptual 3D model
01 · Power

Turbine and recuperator

A 2 MW-class industrial gas turbine in a 40 ft container: single shaft, dry low-emission combustion, a 480 V generator. The ACTE recuperator rides on a skid on the roof, with the heat recovery and the stack.

More than 40,000 h between major overhauls

Compute container with its aisle wall removed, showing a row of racks under a V-type dry cooler, conceptual 3D modelConceptual 3D model
02 · Compute

Eight racks in one row

Eight liquid-cooled AI racks, the network rack mid-row and three coolant distribution units, one of them a spare. A V-type dry cooler covers the roof.

Water at 45 °C into the rack

03 · Electrical and battery

Switchboard and battery

A 480 V switchboard with A and B buses, a 1.4 MW grid-forming converter and an LFP battery in its own fire-rated room. The battery absorbs rack load steps, starts the turbine and carries the block for ~7 minutes on a trip.

Islanded: no grid, no substation

Cooling container with its wall removed, showing the absorption chiller and pumps under a cooling tower, conceptual 3D modelConceptual 3D model
04 · Cooling

Chiller inside, tower on the roof

A 20 ft container with a lithium-bromide absorption chiller that runs on exhaust heat, its pumps, and a cooling tower on the roof.

Cooling made from heat, no compressor

Gas skid: metering, regulation, filter, shutoff and a compressor that lifts distribution-main gas to the turbine’s 13 barg. A transmission pipeline already has that pressure.

WHAT ONE BLOCK SUPPLIES

Power, cooling and heat, per block.

1.4–1.6 MWe

Net electrical power available

~1,630 kW at 15 °C and ~1,460 kW at 25 °C, after a 3% margin; 1,679 kW gross at ISO conditions.

~1.36 MWe

Held at full load up to 45 °C

Above ~30 °C, exhaust heat cools the air the turbine draws in.

1.12–1.26 MW

IT power

8 liquid-cooled racks of 140 kW (GB300-class NVL72): 1.12 MW, held to 45 °C. Or one NVIDIA DSX POD of 7 Vera Rubin NVL72 racks: 1.26 MW, continuous to 40 °C and to 45 °C in dry climates.

~2.3 MWth

Heat recycled by the ACTE recuperator

Returned to the cycle: electrical efficiency rises from ~23% to ~33%, about 30% less gas per kWh.

~1.0 MW

Rack heat rejected with no compressor

1,008 kW: water at 45 °C into the rack, a V-type dry cooler, only fans and pumps.

137–333 kW

Absorption cooling from exhaust heat

Chilled water from a 352 kW chiller: 137 kW on a mild day, 333 kW at 45 °C, including turbine-intake cooling.

~1.7–1.9 MWth

Exhaust heat left above 100 °C

Available as hot water or steam for a neighbor (cogeneration).

31–33%

Electrical efficiency

36–42% counting the cooling made from exhaust heat; up to ~80% with cogeneration.

~106 kW

Electricity spent on cooling

For 1,120 kW of IT. A grid-connected data center at a PUE of 1.3 would spend ~336 kW.

1.4 MW · 187 kWh

Battery and converter

LFP. Absorbs rack load steps of up to ~340 kW within milliseconds, starts the turbine with no grid and rides through ~7 minutes on a trip.

~430 Nm³/h

Natural gas at 25 °C

Pipeline natural gas. Hydrogen blends: see Fuels below.

~23 × 18 m

Site per block

Including the maintenance areas, on level concrete pads.

Design targets per block, natural gas, sea level; the recuperated version will be validated on the pilot block.


Exhaust duct rising from the power container into the recuperator on the roof skid, conceptual 3D modelConceptual 3D model
HOW THE HEAT IS USED

The same exhaust, used three times.

  • Recuperator. Air leaves the compressor at 280 °C and the ACTE recuperator heats it to 523 °C with the turbine exhaust, ahead of the combustor. ~2.3 MWth comes back into the cycle instead of going up the stack.
  • Absorption chiller. The exhaust leaving the recuperator, at ~300–340 °C, heats water to ~95 °C to drive the chiller: chilled water for the heat the racks release into the air and, on hot days, for the turbine intake.
  • Cogeneration. ~1.7–1.9 MWth is still left above 100 °C, for hot water or steam next door.

Rack heat takes a separate path: ~90% of it leaves in water at ~55 °C, and the dry cooler sends it to the air with fans and pumps only.

About the ACTE recuperator

Design targets per block, natural gas, sea level; the recuperated version will be validated on the pilot block.

Power block beside an AI pod, seen from above, conceptual 3D modelConceptual 3D model
NVIDIA DSX POD

One pod, one power block.

The NVIDIA DSX POD is a complete inference factory: 7 Vera Rubin NVL72 racks and 504 GPUs, integrated and validated before delivery, installing in as little as 14 days on a prepared site (NVIDIA, 21 Oct 2026).

The block is designed to power it at its door, at 800 VDC, the pod’s own input: 1.26 MW of racks, continuous to 40 °C and to 45 °C in dry climates. Each pair runs on its own, and a site grows one pod at a time.

NVIDIA, DSX POD, Vera Rubin and NVL72 are trademarks of NVIDIA Corporation.

Site with rows of identical power blocks and AI pods, aerial view, conceptual 3D modelConceptual 3D model
SCALE

A gigawatt as a production line.

1 GW of IT is about 890 identical blocks: the same drawings and the same factory test, repeated. Blocks group in small pools at metro sites with fiber and in larger clusters in the gas basins.

  • Telecom central offices: fiber, fencing and a site that is already there.
  • Gas compressor stations and gas basins: gas on the spot.
  • Industrial parks and corporate campuses: gas and fiber under contract, the customer next door.
DELIVERY

Power in days from stock, months to order.

~20 days

From stock to switch-on, on a ready site: gas connected, air permit issued, pad in place.

3–9 months

For new orders, built and hot-tested at the factory.

What the site needs

  • ~23 × 18 m of level ground with concrete pads
  • Natural gas, ~430 Nm³/h
  • Two fiber pairs
  • ~0.6–1.5 m³/h of make-up water for the cooling tower
  • No connection to the power grid
  • Remote operation, no permanent staff on site

Indicative · stock confirmed at order.

FUELS

Natural gas today, hydrogen blends in development.

Specified fuel

Natural gas

From a transmission pipeline or a distribution main; the gas skid brings it to the turbine’s pressure.

Blends up to 30%

Hydrogen

In development — already accepting up to 30% hydrogen blended with natural gas.

Racks waiting for power where there is gas and fiber?

Send the site address, the gas line, the fiber and the rack or pod delivery date. We reply with the site’s power path and a delivery date.

Talk to Bluen