Skip to content
Where it fits Why Nectar The system Deploy How it starts FAQ Start your trial

The on-site AI compute layer for Physical AI.

Robots, autonomous vehicles, industrial vision, drone fleets — they all hit the same wall: on-site compute can't keep up with the models. Nectar closes the gap — shared GPUs on your floor, delivered and managed as a service.

Backed by & building with

WHERE IT FITS

A robot is only as smart as the model behind it.

A new generation of models gives every machine one brain — it sees and acts, coordinates the fleet, and learns from the work.

Across Robot fleets Autonomous vehicles Industrial vision Drone fleets
One brain, three jobs
Sees & acts Vision-Language-Action π0 · GR00T N1 · Gemini Robotics
Coordinates the fleet Fleet foundation models Amazon DeepFleet
Learns & improves World models + fleet learning NVIDIA Cosmos
here's the catch

Too big for the robot. Too far in the cloud.

On the robot Too big The model won't fit on-board.
On your floor On-site AI compute Close enough for the loop, big enough for the model.
In the cloud Too far A 300–500 ms round-trip misses the loop.
Why Nectar

Running on-site compute is a full-time job. It shouldn't be yours.

Unless on-site AI compute is your core business, doing it yourself pulls capital, people, and on-call away from the robots you're actually building.

Who runs the compute where Physical AI works?
Procure

Capex shock. GPU depreciation and a resale cliff — capital tied up in hardware, not your product.

Staffing

Wrong team, wrong work. Your engineers run infrastructure instead of building robots.

Reliability

A 24/7 NOC you have to staff. Alert fatigue and MTTR you own forever.

It shouldn't be your job. So we make it ours — one managed compute layer.

The system

Nectar's Box + Brain = on-site AI compute as-a-service.

~50%Less floor footprintvs air · direct-to-chip · DIY
~30%Less power drawimmersion vs air cooling
15 minTo first workloaddeploy on install
$0Upfront capexOpex · monthly fee
Inside the node on your floor · on your LAN
Hardware

The Box

The immersion-cooled enclosure and the GPUs inside it — capacity that serves your fleet's inference, milliseconds away.

Software · reliability loop

The Brain

The on-box loop that keeps the Box dependable — fuses hardware and workload signal, autoscales within headroom, holds latency inside SLO.

Delivered as a managed service
Install & integration Monitoring 24/7 NOC Upgrades
The architecture
YOUR SITE · LAN Your fleet untouched · on-site data streams actions Nectar Box GPU capacity reserved headroom Brain · reliability loop fuses hardware (thermal · power) + workload (p95 · replicas) autoscale within headroom workload data stays on the LAN runs alongside your Kubernetes — joins as a worker node, or standalone
Deploy

Train anywhere. Deploy on Nectar. Never operate it.

01 · Install

Drops into your cluster, or stands alone.

Join your control plane as a GPU worker, or let the Box bootstrap its own K3s. Same flow everywhere, ~5 minutes.

GKEEKSAKSOKEOpenShiftCoreWeaveRancherbare-metal
02 · Operate

After that, you don't touch it.

Pre-configured on arrival and operated as a managed service — the next five years, not just the install.

driversself-healing24/7 NOCpatches & upgradesthermal & powercapacity
Data sovereignty

Your data stays on your floor.

Your workload data — payloads, prompts, logs — never leaves your network.

Stays on the LAN. Your fleet and the on-site Box exchange inference and workload payloads — never off-site.
Telemetry out. Health and performance telemetry plus model and policy updates leave the boundary, Brain → Nectar.
Managed ops in. Software and policy updates and remote remediation come back from the NOC — never your data, never models learned from your fleet.

Either way, your workload data never crosses the boundary — only operational telemetry out, managed updates in.

How it starts

Two steps. Trust first, scale next.

Start small. A one-week POC on a Nectar Brain-managed Nvidia Spark unit shows you how easy deployment is. Then pilot with a production Box that proves it at scale.

Step 1 about a week

The Spark trial

NVIDIA DGX Spark · GB10

A small, standard stepping-stone unit. It validates that Nectar slots into your stack and that its telemetry reconciles with your ground truth — integration and trust, not performance claims.

  • Self-serve K8s join as a node, with no disruption to running workloads.
  • Runs your workload and reports telemetry that reconciles with your own ground truth.
your workload data never leaves your network.
Step 2 on-ramp to production

The Box pilot

Immersion-cooled · H200/B200-class

Once the Spark proves Nectar installs cleanly and reports honestly, the production Box steps in — a high-VRAM node that proves the operating KPIs at site scale over 60–90 days.

What the pilot proves

  • Tail-latency stability
  • Time-to-capacity
  • Intervention reduction
FAQ

Questions fleet operators ask

The control loop needs 20–100 ms; warehouse-to-region cloud is 300–500 ms. Egress gets expensive at fleet scale. 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.

Two paths. K8s Join — the Box joins your existing control plane as a worker node via kubeadm or k3s-agent in ~5 min. Standalone — it runs alongside your stack with no cluster join. Either way, your workloads aren't re-platformed and your CI/CD runs as it stands.

You choose Brain's access, too: a scoped-RBAC joined-cluster mode, or a read-only shadow mode with no cluster access at all.

Each Box runs high-VRAM, H200/B200-class GPUs — up to 24 per Box — provisioned with reserved headroom so you can scale utilization instantly. We describe the tier, not exact SKUs.

No. Your workload data never leaves your LAN. What crosses the boundary is operational, both ways: health and performance telemetry plus model and policy updates go out to Nectar; managed software and policy updates and remote remediation come in from the NOC.

A two-step on-ramp. First, a one-week POC on a Brain-managed NVIDIA DGX Spark — it proves Nectar joins your stack cleanly and that its telemetry reconciles with your ground truth. Then a production Box pilot that proves the operating KPIs at site scale over 60–90 days. No upfront capex — you pay opex.

Yes. Once you're in production, two levers: each Box ships with reserved GPU headroom, so you scale utilization up instantly with no new procurement — and as the fleet grows, additional Boxes add capacity. Elastic time-to-capacity is the point.

Start your trial

Kick off the Spark trial.

About a week on a pre-installed NVIDIA DGX Spark. The week proves integration — that Nectar slots into your stack and that its telemetry reconciles with your ground truth. Not the operating KPIs; those come with the Box pilot.

  • Proves integration and trust. Telemetry reconciles with your ground truth — not the operating KPIs.
  • Self-serve K8s join. No disruption to running workloads.
  • Your workload data never leaves your network.
  • Non-binding. A clean, reversible test.

Non-binding. We typically reply within one business day. Your workload data never leaves your network.

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