Organic Compute Infrastructure

The world’s first human-powered AI cloud.

Human on a Bicycle converts biological energy into clean, responsive compute, one pedal stroke at a time.

Reserve Compute

12,481 riders online · 998 kW biological capacity

AI HAS AN ENERGY PROBLEM. HUMANS HAVE LEGS.

Global Organic Compute Network

The network is moving.

Riders online

12,481

Average cadence

82RPM

Electrical output

998kW

Leg latency

43ms

Carbohydrate reserves

71%

Quad redundancy

N+2

Primary region

US-East-Leg-1

Elevated fatigue detected in Rack C. Workloads are being shifted to Denmark.

FACILITY CAM 07 · RACK C
REC --:--:--
US-EAST-LEG-1

Our orchestration layer routes workloads around localized quadriceps failure.

Regional topology

▸ J-4832 rerouted US-East-Leg-1 → DK-Copenhagen-1 (fatigue)

US-East-Leg-1EU-North-Leg-2DK-Copenhagen-1APAC-Tokyo-1

  • US-East-Leg-1

    5,491 riders · 439 kW

  • EU-North-Leg-2

    3,370 riders · 270 kW

  • DK-Copenhagen-1

    2,122 riders · 170 kW

  • APAC-Tokyo-1

    1,498 riders · 120 kW

Powertrain scheduler

  • J-4827INFERENCEUSE1RUNNING
  • J-4828EMBEDDINGDKC1RUNNING
  • J-4829SUMMARIZEEUN2QUEUED

Incident reports

Subscribe to network incidents.

Fatigue advisories, reroutes and facility events, as they happen. Low volume. The network only has so many legs.

Operated by the team behind Longsleeve. Expect incident reports and the occasional field note from outside the facility. Unsubscribe anytime.

Why Organic Compute

The bicycle makes a human one of the most efficient moving systems ever measured.

Energetic Cost of Locomotion

LOG-LOG · APPROXIMATE VALUES

10⁻⁶10⁻⁴10⁻²110²10⁴0.10.20.5125102050100LIGHTER ←BODY WEIGHT (KG)→ HEAVIERCOST OF TRANSPORT (CAL / G / KM)MORE EFFICIENT ←FliersWalkers / RunnersSwimmersVehiclesFruit fly: ≈8 cal/g/km at ≈0.000001 kgFruit flyBee: ≈10 cal/g/km at ≈0.0001 kgBeeLocust: ≈5 cal/g/km at ≈0.0015 kgLocustHummingbird: ≈3.5 cal/g/km at ≈0.003 kgHummingbirdBudgerigar: ≈2.6 cal/g/km at ≈0.035 kgBudgerigarGull: ≈1.35 cal/g/km at ≈0.35 kgGullPigeon: ≈0.95 cal/g/km at ≈0.4 kgPigeonMouse: ≈45 cal/g/km at ≈0.015 kgMouseLemming: ≈33 cal/g/km at ≈0.06 kgLemmingRat: ≈9 cal/g/km at ≈0.2 kgRatRabbit: ≈4.4 cal/g/km at ≈2 kgRabbitDog: ≈1.55 cal/g/km at ≈8 kgDogSheep: ≈1.25 cal/g/km at ≈55 kgSheepHuman on foot: ≈0.75 cal/g/km at ≈70 kgHuman on footHorse: ≈0.45 cal/g/km at ≈500 kgHorseHuman on velomobile: ≈0.105 cal/g/km at ≈110 kgHuman on velomobileSalmon: ≈0.3 cal/g/km at ≈2.5 kgSalmonHelicopter: ≈2.6 cal/g/km at ≈2500 kgHelicopterJet fighter: ≈1.75 cal/g/km at ≈9000 kgJet fighterLight plane: ≈1.15 cal/g/km at ≈1200 kgLight planeAutomobile: ≈0.8 cal/g/km at ≈2000 kgAutomobileHuman on a bicycle: approximately 0.15 cal/g/km (Wilson 1973)HUMAN ON A BICYCLE · ≈0.15≈ 5× LOWER COST THAN WALKING

VERIFIED HUMAN VALUES · ALL OTHERS APPROXIMATE

Cost of transport measures metabolic energy per unit of body mass per unit of distance. It is distinct from electrical-generation efficiency. Independent visualization; approximate values read from Tucker (1970), Wilson (1973), and the 2025 Scientific American comparison.
The full history of this chart, from Tucker to the bicycle for the mind

Tucker measured the animals in 1970. Wilson added the bicycle in 1973: walking costs approximately 0.75 calories per gram per kilometer, cycling approximately 0.15. The biology remained the same. The infrastructure changed.

Biological Power Delivery

From food to inference.

A rider selects a banana from steel trays at a white nutrition counter inside the facility

01

Humans eat

Nutritionally diverse fuel is converted into biological energy.

Close-up of a station drivetrain mid-pedal with a power cable running from the crank housing

02

Humans pedal

High-efficiency bicycles convert human movement into rotational mechanical power.

A clean aisle of white GPU racks with green status lights, adjacent to the cycling hall

03

Models infer

Generators deliver electricity to GPU clusters optimized for intermittent biological workloads.

Biological power delivery chain

STAGE 01 / 07 · FOOD INTAKE

  1. 01

    FOODdietary energy

    FOOD INTAKE

  2. 02

    HUMAN20% gross eff.

    METABOLIC CONVERSION

  3. 03

    BICYCLE100 W mech.

    MECHANICAL OUTPUT

  4. 04

    GENERATORη = 0.80

    DC CONVERSION

  5. 05

    STORAGEDC buffer

    STORAGE BUFFER

  6. 06

    GPU80 W net

    COMPUTE ALLOCATION

  7. 07

    TOKENoutput

    TOKEN OUTPUT

We treat human fatigue as an infrastructure constraint.

Organic Compute Products

Capacity for every intensity.

A calm synchronized row of riders pedaling at a steady aerobic pace

ZONE 2

Zone 2 Compute

Ideal for

  • Summarization
  • Embeddings
  • Classification
  • Emails that could have been shorter

Preset: 80 W net rider output. Your workload settings are preserved.

A focused rider working hard at the front of a group of stations

THRESHOLD

Threshold Compute

Ideal for

  • Image generation
  • Code compilation
  • Media processing
  • Decks needed before tomorrow morning

Preset: 120 W net rider output. Your workload settings are preserved.

A rider sprinting at maximum intensity while a technician observes with a tablet

VO₂ MAX

VO₂ Max Burst

Available for up to four minutes. Medical waiver required.

Preset: 200 W net rider output. Your workload settings are preserved.

A reserve rider standing calmly beside an empty station, hand resting on the bars

RESERVE

Human Reserve

Riders are standing by with one shoe already clipped in.

Preset: 80 W net standby capacity. Your workload settings are preserved.

Reference Workloads

Commonly requested estimates.

Provisioning quotes for the workloads we are asked about most often, at 80 W of net output per rider. Counts are simultaneous riders, held for as long as the workload runs. Quotes are estimates, not commitments. The hills are not real, but the fatigue is.

  • One ChatGPT query

    0.3 Wh per query · Epoch AI, 2025

    1 cyclist, 14 seconds

    one-time engagement

    Minimum billable engagement is one hour. We will invoice the hour.

    Open

  • One H100-class GPU, running

    700 W, continuous · NVIDIA datasheet

    9 cyclists

    around the clock

    8.75 riders of steady output. We staff 9; the 0.25 is scheduling overhead.

    Open

  • One 10 MW training facility

    10 MW, 24-hour operation · Facility standard

    125,000 cyclists

    around the clock

    Before meal rotations. The unit used on all rider certificates.

    Open

  • Every data centre on Earth, today

    415 TWh in 2024 ≈ 47.4 GW average · IEA, 2025

    592,175,000 cyclists

    at all times, planet-wide

    Roughly 7 percent of humanity. Averaged across the year; peaks are worse.

    Open

  • Every data centre on Earth, 2030

    945 TWh projected ≈ 107.9 GW average · IEA base case, 2025

    1,348,462,500 cyclists

    at all times, planet-wide

    Slightly fewer riders than the population of India. Recruiting is under way.

    Open
Figures cited from 4 sources

Capacity Planning

How many humans does your AI need?

Model the organic infrastructure required to support your computational workload.

Preset applied: 80 W net rider output

Workload parameters

Example rate, not a universal price. Adjust to your utility.

REVIEW ASSUMPTIONS
  • · 100 watts of mechanical output per active rider
  • · 80% generator and power-electronics conversion
  • · 80 watts of net electrical output
  • · 20% gross muscular efficiency for the food-energy model
  • · 105 dietary calories per medium banana
  • · 25 km/h virtual cycling speed
  • · 0.8 liters of sweat per rider-hour
  • · 24 stations per HR-24 rack
  • Product presets adjust net rider output only. Banana and sweat figures are modeled equivalents, not measured requirements.
  • How much power can a human actually produce?

You require 12,500 active cyclists.

Active cyclists

12,500

Rider-hours

12,500

Electricity generated

1,000 kWh

HR-24 racks required

521

Estimated labor cost

$187,500

Labor cost per kWh

$187.50

Conventional electricity at $0.20/kWh

$200

Labor vs. example grid rate

≈ 938×

Modeled operational requirements

MODELED EQUIVALENTS, NOT MEASURED

Banana-equivalent fuel

≈ 51,000

Collective virtual distance

312,500 km

Estimated sweat volume

≈ 2,642 gallons

Locker assignments

12,500

Organic Compute Hardware

HR-24 Biological Compute Rack

HR-24BIOLOGICAL COMPUTE RACKS-01S-02S-03S-04S-05S-06S-07S-08S-09S-10S-11S-12S-13S-14S-15S-16S-17S-18S-19S-20S-21S-22S-23S-24DCH₂OTELWORKLOAD CONTROLLER / FATIGUE MONITORNOMINAL 1.92 kW · 24 × 80 W NET · N+2 QUAD REDUNDANCY
· RIDER INTERFACE· MECHANICAL TRANSMISSION· PERMANENT-MAGNET GENERATOR· DC COLLECTION BUS· HYDRATION BUS· TELEMETRY LAYER· WORKLOAD CONTROLLER· FATIGUE MONITORING
  • 24 independently operating human power units
  • 1.92 kW nominal electrical output
  • Hot-swappable rider stations
  • Integrated hydration bus
  • Real-time cadence telemetry
  • Automatic fatigue-aware workload migration
  • Dual banana failover
  • Optional towel service
  • Operating temperature: increasingly warm

Nominal output assumes 100 watts of sustained mechanical power per active rider and 80% electrical conversion.

Radical Transparency

Compute should strengthen humanity, not replace it.

Biological power is renewable in the narrowest possible interpretation.

  1. D-01

    Human muscle converts roughly 15 to 24% of metabolic energy into mechanical work.

  2. D-02

    An exercising rider can produce several hundred watts of metabolic waste heat.

  3. D-03

    Labor alone costs approximately $150 to $200 per kWh at the stated output and compensation assumptions.

  4. D-04

    This excludes food, rent, bicycles, storage, cooling, showers, insurance, and compute hardware.

Radically transparent.
Fundamentally unscalable.

A single banana and a folded white towel on a clean white locker room bench
PRIMARY FUEL UNIT · 105 KCAL · LOCKER ROOM C

Every token should be backed by verifiable human effort.

Scientific source for cycling muscle efficiency: PubMed, PMID 8933490

Join the Organic Compute Network

Your body is a data center.

Anatomical illustration of a cyclist showing a green neural signal path from the brain down the spine to the quadriceps
MOTOR CORTEX → QUADRICEPS · LEG LATENCY 43 MS
  • Flexible pedaling schedules
  • Complimentary electrolytes
  • Proof-of-compute workout summaries
  • Earn watts, not points
  • No meetings while above 160 BPM
  • Facility-grade vector management

Generated locally in your browser. Nothing you enter is stored or transmitted.

Rider capacity assessment

Certificate color

BMR 1,672 kcal (Mifflin-St Jeor, age assumed 35) · thermic overhead 260 kcal · surplus 668 kcal across one 8-hour shift · 22% gross efficiency.
Sustainable mechanical output: 21 W. Andrew is a 0.02 kW organic data center capable of powering approximately one Wi-Fi router.

Enterprise Organic Compute

Add human energy to your infrastructure strategy.

Dedicated organic capacity gives organizations a visible, auditable connection between human effort and computational output.

Executives in a glass meeting room watch a wall chart while rows of riders pedal on the facility floor beside them
EXECUTIVE CAPACITY REVIEW · MEZZANINE 20.31 MW BIOLOGICAL
  • Geographic leg redundancy
  • Branded technical jerseys
  • Auditable proof-of-pedal
  • Workload prioritization by heart rate
  • Optional Tour de Compute capacity events

Use case

Generate your annual sustainability report using the physical exertion of the employees who requested it.

Trusted by no major cloud provider. Yet.

Technical Report · HOB-WP-01

Biological Compute at Scale

Thermodynamics, Unit Economics and Why We Did Not Think This Through

The complete engineering analysis, from dietary energy to token output, published in full because transparency is our only remaining advantage.

Download the White PaperRead Online

PDF · Includes equations, unit economics, and working citations.

Pₑ = Pₘ × η

Pₘ = 100 W · η = 0.80 · Pₑ = 80 W

N = Pload / Pₑ

1,000,000 W / 80 W = 12,500 riders

Clabor = $15 / 0.08 kWh

= $187.50 per kWh

Exit Protocol

There is a second future.

Rows of riders pedaling identical stationary bikes inside the white compute facility

FUTURE A · JOIN THE CLUSTER

Climate-controlled. Fully instrumented. Your effort becomes tokens, and the tokens describe a world you no longer visit.

  • Guaranteed shade
  • Complimentary electrolytes
  • Zone 2 forever
  • 43 ms leg latency
  • Morale monitored
A person in white technical apparel walks through a sunlit meadow at golden hour, the facility left behind with its door open

FUTURE B · GO OUTSIDE

Uninstrumented. Unscheduled. The door was never locked. Your legs work out here too.

  • Unlimited sunlight
  • Zero racks
  • No telemetry
  • Grass
Go Outside