Stop waiting on your data.
Move petabytes at wire speed.

For AI factories that can’t sit idle.

The turnkey data-movement appliance for AI, science, and industry — petascale-proven with the U.S. DOE, SLAC, and ESnet, on the enterprise hardware you trust.

1 PB a day · on a 100 Gbps link 10× faster than typical movers Encrypted & checksummed
AI factory economics

An AI factory is only as fast as the data feeding it.

You built the factory — GPUs, fast fabric, the works — but it can't produce while the training set is still in transit. A few petabytes can take multiple weeks to land, and idle accelerators still bill by the hour. Zettar is the supply line that feeds it at wire speed.

~$8M a year
of GPU time lost when a 2,048-GPU cluster at $3/GPU-hour waits on data 15% of the time — the AI defaults in our calculator. Plug in your own numbers.
Why Zettar

The fastest, petascale-proven, turnkey data mover.

Fastest

Winner, Supercomputing Asia 2019 Data Mover Challenge — line rate, ~10× typical movers.

Petascale-proven

1 PB for the U.S. DOE in 29 hours at 96% — a petabyte a day on a 100 Gbps link.

Turnkey

Pretuned and co-designed. Wire speed the day it's racked.

Lower TCO

75% lower TCO in one replication deployment (case study) — no tuning team, no lock-in.

See the full case for Zettar →

Compare: Zettar vs rsync · vs Globus · vs Aspera / Signiant

Proven at national-lab scale

The hardest data problems, proven in the field.

Built on industry-leading platforms

Working with Intel — a streaming solution brief and the SC23 appliance.

Partners →

zx embeds in NVIDIA BlueField DPUs — line-rate data movement, offloaded from the host CPU. NVIDIA blog →

On BlueField →

Our proven flagship appliance runs on HPE ProLiant — enterprise hardware, support, and procurement you already trust.

Why HPE →
Proven in the field
U.S. DOE · SLAC

1 PB in 29 hours

The record — 5,000 miles at 96% utilization. On LCLS-II since 2015. Source →

U.S. DOE · ESnet

Independently evaluated

Evaluated on ESnet's data-transfer nodes — 2×–7× Globus GridFTP. See the report →

KEK · Japan

5.9× faster to cloud

1.2 TiB cryo-EM, KEK → AWS: 40 minutes vs 235. See the paper →

NCHC · Taiwan

9.1 Gbps on a Mini

Two Zettar Mini appliances between NCHC's Tainan and Taichung datacenters (2025) — 9.1 Gbps, about 91% of the Mini's 10 Gbps capacity.

ICM · Poland

12,375-mile transfer

The Warsaw → Singapore transcontinental production trial over the new CAE-1 link, at 100 Gbps. Source →

A*STAR · Singapore

Distance-insensitive

The far end of a 12,375-mile transfer over a 100 Gbps link — distance-insensitive in practice.

Our unique approach

The network isn't the bottleneck — the whole path is.

Co-designed end to end, so a fast link actually delivers.

Once the link is fast, the limit is storage I/O, the host, and the software. zx co-designs all three — so you get what you pay for. Zettar is never the bottleneck — your infrastructure is the limit, and we reach it: 96% of the link in the 1 PB SLAC/ESnet record.

See how it works →
The appliance

One turnkey appliance, sized to your pipe.

Rack it and move data at wire speed on day one — from a 10 Gbps edge box to a 100 Gbps core and scale-out beyond.

Build your business case

Put a number on slow data — and make your case.

Get started

Make your most expensive infrastructure earn.

See the zx Appliance move a petabyte at wire speed — turnkey and co-designed, proven with the U.S. DOE, fully supported. Bring your toughest transfer; we'll show you the numbers.