Fleet compute

The fleet's brain belongs on the floor.

Shared GPUs on your factory floor, run for you. One pool, every robot, inside the loop.

Ask for a link See how it works

Backed by & building with

01
On the robot Too small. Carries the reflexes, not the reasoning.
02
In the cloud Too far. Every action waits on the network.
03
On the floor Shared. One pool, every robot, inside the loop.
The loop runs many times a second. The brain has to be inside it.
Why believe it

The loop has a budget. The cloud doesn't fit in it.

5-33 ms
the control-loop budget
30-200 Hz loops. Practical cloud round-trips measure 50-150 ms.
control-loop physics · outside the loop before egress cost
99.96%
System-1 actions on time, one pool
32 robots sharing one server-class GPU pool; 0% without a scheduler.
ROSA · NVIDIA Research + Stanford · 8× H200, GR00T N1.6 · arXiv:2607.01088 · the paper's numbers, not ours yet
The whole argument, with the diagramWhy the brain belongs on the floorPer-robot GPUs idle by design, the published pooling pattern, and what we run on top of it.Read the long page
What we run for you

The pattern is public. The operating layer is the product.

Bring your models. Connect your fleet. We run the pool. The Box on your floor, the reliability loop underneath it, and 24/7 ops - run for you.

Inside the node on your floor · on your LAN
Hardware

The Box

The enclosure and the GPUs inside it - the shared pool that serves your fleet's inference, milliseconds away. Sized in robots, not chips: your whole model set resident, every loop in budget, with headroom for the fleet you're growing into.

Software · reliability loop

The Brain

The on-box control loop that keeps the pool dependable - it fuses GPU, power, cooling, network, and workload signals, autoscales within reserved headroom, and holds tail latency in band.

Procure

Opex only. Capital stays in your product, not depreciating GPUs.

Reliability

24/7 ops, run for you. No NOC to staff, no alert fatigue to own.

Capacity

Headroom from day one. Reserved capacity for the fleet you're growing into.

Staffing

Your team builds robots. We run the infrastructure.

Your compute, your data, your models - our loop to keep them production-grade.

Data sovereignty

Your data stays on your floor.

Workload data - payloads, prompts, logs - never leaves your network.

Only operational telemetry goes out; only managed updates come in. Never your data, never models learned from your fleet.

How it starts

Start in the sandbox.

Twenty minutes in the operator console, the one that runs on the box, on seeded telemetry. Set up a fleet, run the dry-run, break it.

01

The sandbox

One link, your name on every screen. Pick the picking-cell template or describe your own cell in a sentence, run the dry-run against your declared fleet, and take the session record with you.

by invitethe console that runs the boxnumbers from real GPU runs
Ask for a link
FAQ

Questions fleet operators ask

Shared, on-site AI compute that serves a whole fleet's real-time inference from one pool - close enough for the control loop, big enough for the model. Instead of a GPU sized for each robot's peak, one pooled Box on your floor serves every agent and trains between shifts. Nectar delivers and operates it as a service.

Tight control loops run 30-200 Hz - a 5-33 ms budget. Practical cloud round-trips measure 50-150 ms: outside the loop, before egress cost. The Box keeps inference and data on-site.

Neither. The Box serves the inference your stack calls; your agent control and orchestration stay yours, with a safe fallback if the Box is unavailable. Reflexes stay on the robot - ~100 Hz local control and safety fallback - while the fleet's heavy models run from the pool.

Exactly - the serving pattern is public, and we embed it rather than reinvent it. What you pay for is what doesn't commoditize: the Box on your floor, the reliability loop underneath it - thermal, power, network, failure, everything the pattern assumes stable - and the operation: install, monitoring, 24/7 NOC, and hardware refresh on our clock.

More questions - deployment, hardware, data, the sandbox - are answered on the long page. Still weighing it? The sandbox is non-binding - start there.

Ask for a link

Ask for a sandbox link.

Twenty minutes, by invite. We read every request and reply within a business day.

  • The console that runs the box.
  • Seeded telemetry, numbers from real GPU runs.
  • Your name on every screen.
  • Non-binding.
  • Your workload data never leaves your network.

Non-binding. We reply within one business day.

We use your details only to respond and coordinate - we don't sell or share them. Privacy.