We move the data that powers AI.
Compute gets the headlines. Data movement decides whether your GPUs ever run. For more than a decade Zettar has done one thing with uncommon focus — get massive data where it needs to be, fast and fully verified. We took that work from a record-setting U.S. DOE transfer to a turnkey appliance — every hardware component selected, matched, and pretuned to our zx data mover — so you get line-rate throughput the day it's racked.
Born at Stanford and SLAC. Proven at petascale.
Zettar was founded in 2008 by Dr. Chin Fang (CEO), with roots at Stanford and SLAC. Since 2014 the focus has never wavered: solve the problem most tools ignore until it stalls everything — moving enormous volumes of data fast enough to keep data-intensive science and infrastructure running.
Since 2015 we've been engaged on SLAC's LCLS-II, a U.S. DOE exascale-class instrument that demanded throughput far beyond what conventional tools could deliver. With SLAC and ESnet, in 2018 our software data mover set the record — 1 PB in 29 hours at 96% link utilization over a 5,000-mile loop, encrypted and checksummed end to end. Zettar also won the Supercomputing Asia 2019 Data Mover Challenge, and we kept proving it in public: SC21 with DDN and NVIDIA, SC22 on the NVIDIA DPU, an AIC/Intel appliance at SC23, and SC24 on the NVIDIA DPU again, with MiTAC's participation.
As the AI era turned data movement into the bottleneck, we packaged everything we'd proven into the turnkey zx Appliance — with the hardware selected, matched, and pretuned to zx as one engineered system. Its proven core platform is the HPE ProLiant DL380 Gen 12. It uses Linux as the OS and moves data roughly 10× faster than typical commercial or freeware movers. We don't sell AI. We move the data that powers it.
Three convictions, earned the hard way
Data movement is its own infrastructure layer
Compute, storage, and network are treated as first-class infrastructure. Moving data between them should be too. When it isn't, GPUs idle and pipelines stall — so we engineer the data-movement layer with the same seriousness as the rest of the stack.
Co-design beats endless tuning
Wire-speed transfer shouldn't be a science project. Rather than hand you knobs to turn across firewalls and mismatched hardware, we co-design hardware and software into one validated appliance — so you get the result, not the homework.
Proof over promises
Our headline claims are backed by numbers you can check. A petabyte in 29 hours. 96% link utilization. A challenge win against the field. Reproducible results at national-lab scale — not marketing math.
The record, year by year
LCLS-II engagement begins
Zettar is engaged on LCLS-II at SLAC — the U.S. Department of Energy's exascale-class instrument — to solve data movement at a scale conventional tools could not reach.
1 PB in 29 hours — a record-setting transfer
With SLAC and ESnet, Zettar moves 1 PB in 29 hours at 96% link utilization over a 5,000-mile loop, encrypted and checksummed end to end. Source →
Data Mover Challenge winner
Zettar wins the Supercomputing Asia 2019 Data Mover Challenge, validating its data mover against the field on a global stage. Source →
ESnet evaluates zx on its DTNs
U.S. DOE's ESnet completes an in-depth evaluation of zx on its data transfer nodes, confirming 2×–7× the performance of Globus GridFTP, among other advantages. Source →
SC21 demonstration with DDN and NVIDIA
Zettar demonstrates at-scale data movement at SC21 in partnership with DDN and NVIDIA. Source →
SC22 NVIDIA DPU demonstration
At SC22, Zettar shows data movement running on the NVIDIA DPU, offloading the work from conventional servers. Source →
SC23 appliance with AIC and Intel
Zettar unveils a data-movement appliance at SC23 in partnership with AIC and Intel. Source →
The turnkey zx Appliance ships
Everything we proved becomes a product you can buy: the zx Appliance — hardware selected, matched, and pretuned to zx, delivered and supported as one unit, and moving data roughly 10× faster than typical commercial or freeware movers. Source →
Published research, and a new national deployment
Dr. Fang and collaborators publish “End-to-End Data Movement: Paradigm Reexamination and Principles for Efficiency” (arXiv, Dec 2025). Source → The same year, Taiwan’s National Center for High-performance Computing (NCHC) deploys two Zettar Mini appliances across its Tainan and Taichung datacenters.
What that work delivers for customers
Move the data that powers AI.
See the turnkey zx Appliance move petabytes at wire speed — line-rate throughput the day it's racked — co-designed and fully supported.
