Accepted into NVIDIA Inception
Alongside AWS Activate and the Google for Startups Cloud Program.
Umpireal Urban Computers
A digital free-flow junction
that pays for itself.
Traffic keeps moving on a gently raised road. People walk and cycle through an open, high-ceilinged underpass: daylit by day, well lit at night and covered by CCTV. The rooms beside it hold 1 to 20 MW of AI compute, batteries, EV charging and heat. Their income pays for the junction.
In one minute
A road junction and a data centre each fail on their own. Built as one structure, each pays for the other.
An Umpireal Urban Computer is a free-flow road junction with AI compute and battery storage in rooms within the raised road. Traffic never stops, so the fuel, time and emissions lost to stopping are prevented. The compute earns the income that pays for the junction, on land and a grid connection the road already has. Its heat warms nearby buildings, and its battery supports the local grid.
The junctionNo signals, no stops. Traffic keeps its momentum; people walk and cycle through an open, daylit underpass, well lit and covered by CCTV at night.
The compute1 to 20 MW of AI compute and storage, sited in towns, beside the people and machines that use it.
The return~€6M per MW silicon-ready, earning ~€3M a year per MW in the colocation base case.
~5,000 TWhLost every year to stopping, slowing and restarting
~$5 trillionThe annual momentum tax: the fuel, time and wear lost to stopping
A tax that only takes. Every stop at a junction, and every slow-down for a speed bump or a yield, turns momentum into brake heat. Then the vehicle burns up to three times that energy again to get back up to speed. Carmakers fit stop-start systems to save the idling in between, but idling is a small part of the loss, and many drivers switch the system off because the engine cuts out just as they need to pull away. The real waste is the stop itself. Umpireal removes it one digital free-flow junction at a time, starting with new masterplans, then retrofitting existing junctions.Modelled across the world's signalised junctions. Site-level figures are on our facts page.
Patent application IE 2025/0540
PCT application in progress.
First node in preparation
Council letter of interest, with no public capital asked.
A red light sends a stopping wave back down the road. Remove the conflict point and the wave never starts.
Stop signals ration capacity. Free-flow geometry creates it.The problems
Three problems that meet at the same junction.
01 · TRAFFIC
Every junction is a forced landing.
~14 GWhVehicle energy destroyed a year at one busy junction
~1.5M LOf fuel burned a year to buy that momentum back
Up to 3×The energy to restart, versus the energy lost stopping
€5–15MTypical cost of grade separation, with no revenue
- We call it the momentum tax. The stop line turns fuel into brake heat, then the vehicle buys the speed back with its engine at 12 to 20% efficiency, its worst operating band.
- Electric vehicles don't fix it: about 40% of the energy is still lost at a typical stop, and all of the time.
- A laden lorry throws away 24 times a car's energy at a stop, and road crashes kill 1.1 million people a year, about a quarter of them on foot or bike.
- No payout replaces a person. A moment’s distraction at a touchscreen or phone should never meet someone on a crossing, so the free-flow junction separates them by design.
Our answer: a free-flow junction, funded by the compute inside it.
02 · DATA CENTRES
Demand is soaring, but compute can't get built where it's needed.
82→219 GWGlobal data-centre demand, 2025 to 2030
4–10 yrsTypical grid connection queues
35%/yrGrowth in AI inference, to ~93 GW by 2030
14%Of US residents who want a data centre nearby
- Inference is overtaking training as the dominant AI workload, and it works best close to its users.
- Grid, land, water and consent push new campuses away: European projects for 2026 to 2028 average 175 km from a major city.
- In Ireland, data centres already use around 23% of the country's electricity.
Our answer: 1 to 20 MW of compute in town, on a local connection, with no land or water.
03 · THE GRID
Clean power is curtailed when the grid can't take it.
2,500+ GWOf energy projects stalled in grid queues worldwide
2+ yrsLead time for transformers in many markets
- Renewable generation is curtailed when the network cannot absorb it, then fossil plants cover the evening peak.
- Local grid constraints slow new renewable and EV charging connections in growing towns.
Our answer: about 4 MWh of battery storage for every MW of compute, sized to each node, at the edge of the network where the evening peak forms.
Sources: McKinsey, The future of AI workloads (Feb 2026) and Technology Trends Outlook 2026; JLL EMEA 2026; Reuters/Ipsos 2026; CSO; WHO. Traffic energy is from Umpireal's model at 18,000 vehicles a day in a 50 km/h zone, tank to wheel; see facts for assumptions.
The 747 test
22–30 tFuel for a 747 flying 2,000 km direct
54–66 tThe same flight, forced to land and take off again every 200 km
No aviation authority would allow that. It is exactly how road networks are designed, and the forced landings are called junctions.
Test case: the momentum tax, measured
Two roads, one journey. One flows. One stops 28 times.
Between Kells and Clonee, the M3 and the R147 cover almost the same distance. Driven at the same speed, the only difference is stopping. Using 2025 TII traffic counts, we measured what it costs.
~60%more fuel for a car on the R147, at the same speed
~7M litresof fuel a year lost to stop, idle and go on one corridor
83%of a car's motion energy lost slowing from 120 to 50 in a toll tag lane
Why the Urban Computer
One structure answers all three.
AI is moving from training to inference, which works best close to the people and machines using it. Towns need that compute, a road that flows and a grid that holds. A junction can deliver all three.
Compute where demand lives
Conventional data centres struggle for land, planning and grid capacity near towns. The Urban Computer puts inference capacity in the community it serves.
No new land
Built within the highway boundary, on land already set aside for the road. No field lost, no new site to buy.
A road that keeps moving
No traffic lights at the junction. Traffic flows over the top while people and bikes pass beneath in daylight.
One structure, six systems
Everything the junction does.
The road
Every stop charges a momentum tax: the fuel, time and emissions spent getting back up to speed, paid by every driver at every set of lights. A free-flow junction takes the stop out.
TrafficThe junction
The road rises about 2.5 m and the walking and cycling route drops about 1.5 m. People pass through in daylight rather than a subway, and the rooms between the two levels hold the computers and batteries.
StructureThe compute
Colocation for metro AI inference, agents, fine-tuning and sovereign capacity, from 1 to 20 MW set by the grid connection. Built for inference close to demand, not a training campus.
AI and data centresGrid storage
Battery storage scaled to each site, designed to support grid stability rather than only draw from it, helping growing towns get more from the network they have.
Battery energy storageEV charging
Charging at the junction where the site allows, sharing the same grid connection and battery storage as the compute.
MobilityHeat
Heat from the compute put to use in nearby buildings, pools and facilities, at roughly 55 to 65°C where there is a user for it, so it isn't wasted.
Community energyHow it is built
Split the levels, keep traffic moving, use the space.
- The road rises about 2.5 m and the path drops about 1.5 m, on a precast shell a road contractor can install.
- People and bikes pass through in daylight, with every floor inside on one continuous level.
- Direct liquid-to-chip cooling in a sealed loop: no water used, heat out at 55 to 65°C.
- Glazed rooms beside the path hold the compute, the batteries and, where the design allows, civic uses.
- Serviced by van. Built with new developments and scheduled road works, not as a surprise dig on a live crossing.
For towns and developers
The access junction a development needs anyway, paid for by compute.
- A free-flow junction funded by the compute tenant, not the council budget.
- A safe, daylit crossing for people on foot and bike, not a subway.
- No land, water, transmission capacity, rezoning or public capital asked of the town.
- Heat, storage, charging and ground rent where the site and contract allow.
- Three ways for a council to hold a node, none needing capital up front: a concession, a land lease with ground rent, or a shareholding in the node.
Our first market is new developments that already need a road access point. Retrofit of existing junctions follows.
The climate case
Prevented energy waste, not an offset.
Removing the stop prevents the fuel from being burned in the first place. A 1 MW node consumes roughly 8 to 12 GWh a year, while the junction it replaces was destroying about 14 GWh, so the site uses around 6 GWh a year less than before the compute arrived.
~3,000 tCO₂ a year addressed by removing the stop at one junction
~125×More than retiming the signals, which addresses about 24 t
~2,000 tCO₂ net negative per node per year
~5 weeksTo pay back the ~300 t of embodied carbon in the precast shell
ZeroWater used: sealed direct-to-chip cooling, no towers, no plume
~80% lessCopper than a hyperscale campus, with the transformer metres from the racks
Heat from the compute, about 0.7 to 0.8 MW per node at 55 to 65°C, is put to use in the buildings, pools and facilities around it rather than vented to the sky.
Prevented energy is reported as its own line, never netted against consumption. Modelled across the world's signalised junctions, the momentum tax runs to roughly 5,000 TWh and 1.3 gigatonnes of CO₂ a year. Counting the fuel, time and wear lost to stopping, it costs the world roughly $5 trillion a year, and about $6 to 7 trillion once crashes, pollution and freight delay are added.
Umpireal model figures, per node, approximate. Tank to wheel; well to wheel the junction figure rises to about 16.4 GWh a year. Global figures are modelled; the investor case uses site-level assumptions. See facts.
How it compares
No new land, no marine or orbital constraints.
The Urban Computer is not a replacement for a training campus. It is where inference, agents, fine-tuning and regional capacity can run close to the people using them.
| Land data centre | Sea / floating | Orbit | Urban Computer | |
|---|---|---|---|---|
| Capex per MW, excl. IT | ~€11–15M | ~€9M | ~2.5–3× land, every 4–5 years | ~€6M |
| New land | Yes | Marine or port | No | No, existing road |
| Process water | Often | Seawater | None | Zero |
| Service | Excellent | Complex | Replace, not repair | Excellent |
| Civil life | 20–40 yrs | 10–20 yrs | 5–7 yrs | 100-year shell |
| Consent | Contested | Marine | Space and debris | Road wayleave |
| Chip second life | Yes | Retrieve the pod | No, burns up | Yes, reused or resold |
| Revenue lines | Compute | Compute | Compute | Compute, plus storage, heat and charging |
| 2026 reality | Bans and queues | China state pilot | First satellite 2025 | Launching |
Capex excludes servers and GPUs. Land: McKinsey data-centre capex less IT, over ~137 GW added 2025–30, consistent with Cushman & Wakefield 2026. Orbit: BCG 2026, 20-year cost of ownership versus land. Sea: HiCloud Lingang, Shanghai, project budget for a planned 24 MW, state-backed and unaudited. Figures approximate.
Reference case
A worked example: one road crossing.
Larger junctions follow the same stack up to the power available, to around 20 MW.
~1 MWAI compute at a road crossing
~4:1Battery storage sized to compute: about 4 MWh per MW
~€6MCapital per MW, silicon-ready: before chips and racks
~€3MNet operating income a year per MW, colocation base case, with storage and other revenue
100 yrsCivil design life
ZeroLand bought, water used
A worked example, not a specification. Node one will prove delivered build cost, operating costs and net operating income. Hardware lease and ownership models, where umpireal also supplies the computers, are modelled separately. Figures are modelled; umpireal.com/facts is the canonical figure set.
Where it stands, plainly
Progress so far.
Precast tendered25 culvert units for the 350 m² shell, under €500k, built by a road contractor.
Cooling validatedWith Asperitas.
Council letter of interestNo public capital asked.
Campus deploymentIn discussion for 2027.
~€1M self-fundedOver four years.
TRL 9 componentsPrecast, liquid cooling, battery storage and road construction methods are all in use today.
Patent applicationIE 2025/0540, PCT in progress.
No node built yetThe integration is new; the first node is a pilot. Seed round in progress.
Campuses invent the model. Urban Computers run it, store it, and pay for the junction the city was going to buy anyway.
Programmes we are part of
AWS Activate
Google for Startups Cloud Program
AWS Activate
Google for Startups Cloud Program
Work with us
Three ways in.
Towns and developers
Turn the access junction your scheme needs into a funded, free-flow asset with community benefits built in.
Propose a site →Compute buyers
Colocation from 1 to 20 MW for inference, agents, fine-tuning and sovereign capacity, close to your users. Built for serving, not frontier training, which stays on large campuses.
Reserve capacity →Investors
Seed round in progress, funding the first live node, the regional operating structures and the technical team.
Request the deck → Request the overview →Further reading
The Momentum Papers.
Fifteen short papers on the problem and the answer: the physics of the stop, the hidden costs, the missing policy standard, the Urban Computer, concrete as a catalyst for cutting carbon, and why a junction destroys energy faster per square metre than any power station generates it.
The Momentum Tax calculatorPick a junction, turn the dial and see what it wastes every year in energy, carbon, time and collisions.
01Series introductionThree Pressures, One Root CauseEnergy costs, housing and congestion are treated as separate problems. They are not.02Energy economicsThe Hidden Tax at Every PumpCompounding fuel taxes, and the second tax the road collects.03The physics of lossThe Tax No Government CollectsThe physics of the stop, and why restarting costs up to three times more.04The EV mythWhy Your EV Doesn’t Fix ThisAbout 40% of a stop’s energy is still lost, and all of the time.05The shockwave effectThe Shockwave Cities Can’t SeeHow one stop becomes a city-wide jam.06Insurance and ESGOne Junction. Six Claims.Why a junction crash cascades across six insurance lines.07The full costThe $6 Trillion Price of Stop-Go~$5 trillion in momentum tax, ~$6 to 7 trillion in total, every year.08Nature’s designWhat Your Heart KnowsWhat the circulatory system teaches road design.09Policy critiqueThirty Years of Engine RulesThirty years of engine rules, none for the road.10The answerIntroducing the Urban ComputerA free-flow junction and AI compute in one structure, each paying for the other.11The business caseInfrastructure That Pays For ItselfHow the compute pays for the junction.12Policy briefMeasure the Road, Not Just the EngineThe infrastructure efficiency standard Europe is missing.13Materials and carbonConcrete: The CatalystEmbodied carbon repaid in about five weeks, then a century of prevented emissions.14Energy densityThe Most Destructive Square MetreA junction against gas, nuclear, solar, wind, the Hoover Dam and the river you know.15Fleet realityThe Electric Momentum TaxSixteen years on, EVs are 4% of Ireland’s fleet. Even when every vehicle is electric, a forced stop still loses about 40% of its energy, and all of the time.
Contact