European AI that regenerates buildings to create
mini AI data centers, eco-sustainable.

Big Tech performance on hardware the size of a napkin, on sovereign European infrastructure.

At over 170 times less than the competition.

We bring together technological innovation, environmental sustainability, and social cohesion.

Three principles, one machine: the same efficiency that cuts costs also cuts waste and gives value back to communities.

01

Innovation

A proprietary AI model, world-leading in efficiency, running on tiny hardware.

02

Sustainability

Air cooling, renewable power, existing buildings: zero extra environmental impact.

03

Social cohesion

Urban regeneration and social housing within the very same refurbished buildings.

Our AI already exists.

An AI with Big Tech performance, running on hardware the size of a napkin and drawing power like an old lightbulb.

Ele isn’t a promise of future research: it’s already running, today. Thanks to a proprietary quantization and efficiency process and our industrial secrets, it keeps 99% of its original pre-compression performance, with no perceptible degradation. This efficiency is why a Mini AIDC costs 500K instead of 85 million, and why the entire infrastructure project is possible.

15 × 15 × 5 cm
A whole AI data center, in the palm of your hand.
Performance retained
99%
Power · Mini AIDC
15 kW
Efficiency · PT/B
0,490
Efficiency ranking
1st
Inside Ele, our engines

Not a bought-in model: two engines, invented by us line by line.

One serves it, the other trains it. Neither is bought off the shelf: we wrote them bespoke for Ele 5.

01The engine that serves it

OmniFast

It runs Big Tech performance on a module the size of a napkin. Depth where it matters, speed where it doesn’t the rest is one of our secrets.

02The engine that trains it

OmniFast Train

It slashed the cost of training and sent quality soaring, inside tiny, low-power hardware. The how stays within our walls.

World-leading in efficiency. And not by a little.

Size for size, Ele 5 produces more value than any other model. That margin becomes a lower cost per user and higher margins, measured on the public benchmarks used across the industry.

#Model (MoE, public data)PT / B
  1. 1
    Ele 5first in the world
    0,490
  2. 2
    Qwen3.8-Flash-Next
    0,317
  3. 3
    MiniMax M2.7
    0,240
  4. 4
    Qwen3 235B 2507 (Reasoning)
    0,239
  5. 5
    dots3-note Preview
    0,223

Efficiency = value per parameter (PT/B). Comparison models from benchlm.ai public scores, Ele 5 computed with the same method, 1st among MoE models with available data.

1,5×
More value per parameter
versus the 2nd model for efficiency (0.490 vs 0.317 PT/B).
1ª
Efficiency ranking
among MoE models with public data.
84,1/100
Visual understanding (MMMU)
images, charts, and technical data at university level.
84,4/100
Knowledge and reasoning (MMLU-Pro)
reliable competence across complex subjects.

And it stays reliable on everything else: coding, advanced math, understanding across 29 languages, executing complex instructions, and autonomous tool use.

Today’s AI data centers aren’t sustainable.

A standard AI data center has a huge environmental, economic, and social impact, and these aren’t abstract risks: they’re already documented problems, side by side. The more AI demand grows, the more they multiply.

A standard AI data center: huge rows of racks in an industrial warehouse
Today’s data center, the AIDC standard.
Electrical power
1,4 GW
Water use
2 Mln L/h
Footprint
70.000 m²
Build time
72 mesi
Build cost
€ 50 Mld
01

Energy

Around 1.4 GW drawn, as much as a city: the local grid under stress and residents’ bills rising.

02

Water

Millions of liters an hour for cooling, often in areas where water is already scarce.

03

Land and landscape

Warehouses of tens of thousands of square meters: land consumed and landscape disrupted.

04

Time and capital

Years of construction and billions in capex locked into a single, enormous site.

05

Relentless noise

A background hum that never stops, not even at night; infrasound escapes sound meters but is still “felt.”

06

Diesel generators

Tens of thousands of liters of diesel per machine, with spills already documented in recent years.

07

Polluted air

Particulates and nitrogen oxides from the generators: tons of pollutants in the surrounding air.

08

Residents’ health

People living nearby report disturbed sleep, headaches, and stress; hundreds exposed.

Reference profile for the scale figures: new-generation industrial-scale AI data centers (e.g. Paris AI Campus, Fouju, Seine-et-Marne, France). The harms listed are phenomena observed and reported in data-center news over recent years.

The very same work. In a module the size of a napkin.

All those problems, brought to zero. The Elestre Mini AIDC does the job of an industrial AI data center, with no water, no diesel, no noise, no new concrete.

15 × 15 × 5 cm
0 L
water, air cooling
0
diesel generators, no spills
0 m²
new land, in existing buildings
15 kW
of power, not 3 MW
~6 mesi
to build, not 72

For 15,000 users: a standard AI data center versus an Elestre Mini AIDC.

Machine and build cost
€ 85 M€ 500 K
170×
Electrical power drawn
3 MW15 kW
200×
Hardware footprint
150 m²2 m²
75×
Build time
~72 months~6 months
12×
Cooling, water used
Water, huge consumptionAir
0 L
lo standard150 m²Elestre2 m²75× meno spazio

The same work, for the same 15,000 users, in two square meters instead of a hundred and fifty.

Real area scale, 150 m² against 2 m². The m² indicate the hardware footprint; the values are project estimates.

This is the product we’ll build: not an AI model to invent, but a physical infrastructure to replicate, node after node, across Europe.

Transparency note: Elestre describes an infrastructure project in development. The cost, consumption, space, and time values are project estimates, for a comparison over 15,000 users, and are not a guarantee of results. The m² indicate the hardware footprint. The standard AIDC profile refers to new-generation industrial-scale AI data centers (e.g. Paris AI Campus, Fouju, Seine-et-Marne, France). The estimates come from our compact, math-based tests, not from assumptions.

Choosing Ele pays off. For the planet too.

Unlimited tokens on our dedicated hardware, instead of paying per use for a same-class cloud AI. Here’s how much you save in money and resources per 15,000 users a year.

1,98Mln €
Not spent per use
vs a same-class AI served per use.
288t CO₂
Emissions avoided
tons of CO₂e less in a year.
1,3Mln L
Water saved
≈ 1296 m³ of data-center cooling.
13.7Kalberi
Equivalent trees
the CO₂ they absorb in a year (~21 kg each).

Transparency note: these are estimates, declared as such, with a transparent calculation chain. Basis: a typical active user processes ~2 million tokens a month; the factors (energy per token, grid CO₂ intensity, cooling water, a same-class competitor’s per-use price) are the same as the in-app “Context and usage” card. The trees are the CO₂ avoided divided by how much a tree absorbs in a year (~21 kg): it’s the equivalent, not trees planted.

A building shown half and half: ruined on one side, regenerated on the other, with the Ele mascot on the roof
The same building, split in two: yesterday’s ruin and, on that very roof, the AI network bringing it back to life.

Small enough to fit in any building.

Not a new eyesore, but an existing building. Every Mini AIDC rises inside a regenerated property: no new construction, no land consumed, powered mainly by renewable sources.

Where the reclaimed space exceeds the technical need, the same building houses homes and social housing at sustainable rents. AI infrastructure and local regeneration come from the same property operation, creating a new system of community.

01

Regeneration

Abandoned buildings come back to life, with no land consumed.

02

Renewables

Mainly green power, air cooling.

03

Social housing

The spare space becomes homes at sustainable rents.

04

Repopulation

Urban regeneration of otherwise abandoned areas.

What European AI really means.

Data that stays in Europe. Native compliance. No dependence on the US or China.

The European Union flag breaking the chains of the US and China
The thesis

Being “European AI” isn’t a slogan: it’s owning, physically, the infrastructure that AI runs on.

01

Data in the EU

Physical and legal residence of the data on European soil.

02

Native compliance

EU requirements met by design, not bolted on.

03

Own infrastructure

We own the hardware: real sovereignty, not delegated.

The context

ChatGPT, Claude, Gemini, DeepSeek, and Qwen are born outside the EU. Mistral is the most direct European competitor and builds its own data centers in Europe, but on the centralized mega-center model. Elestre’s difference isn’t the flag: it’s distributed, owned infrastructure low-impact Mini AIDCs, profitable from the first node.

An infrastructure Europe doesn’t have yet.

The European AI market is growing, without a sovereign alternative. 33 million SMEs, 100K local authorities, 150K public bodies, and over 450 million citizens are looking today for exactly what Elestre will build.

400Mld €
EU data-center investment
by 2031, up from €5.3bn in 2026.
42Mld €
EU AI investment
by 2031, up from €11bn in 2026.
33Mln
European SMEs in the target
plus 100K local authorities and 150K public bodies.
450Mln
Reachable citizens
the entire population of the Union.

From the first node to the European network.

It starts today, from a single Mini AIDC, and replicates, year after year, into a network covering Europe.

  1. year0

    Today, the first Mini AIDC

    The network starts here, from a single node. You are here.

  2. year1

    The first node live

    B2B AI and API, 5,000 users.

  3. year2

    51 Mini AIDC

    B2B + B2C, 500,000 users.

  4. year3

    2,000 Mini AIDC

    European market, 20 million users.

  5. year4

    10,000 Mini AIDC

    B2B + B2C, 60 million users.

  6. year5

    20,000 Mini AIDC

    B2B + B2C, 100 million users.

Growth trajectory from the Elestre Financial Plan: projections, not guarantees of results.

One brain, for every format.

Images, video, music, websites, games, presentations, voice. Ele 5 creates in every field, and the best way to believe it is to come in and generate freely.

Join the Elestre early access now.

Try the Ele model from the first node of the network: chat, documents, images, music, and video, on European infrastructure. Early-access spots are limited.

Elestre is a project by Asincrona S.r.l.

Early access

Entra in Elestre, basta un tap

Accesso anticipato al modello Ele 5. Semplice, veloce e senza carta di credito.

Continuando, accetti l’Informativa sulla Privacy e i termini d’uso.

On-premise

Elestre on-premise, at your site

Bring a compact version of the modules that make up the Mini AIDC Elestre inside your organization: the Ele model runs on your own hardware and your data never leaves your walls. Full sovereignty for public sector, healthcare, finance and industry.

Investors

Want to invest in Elestre?

We’re building Europe’s first AI network, sovereign and profitable from the very first node. If you want to be part of the project as an investor, let’s talk.