Home/BYOG
Bring Your Own GPU (BYOG)Buy your GPUs anywhere. We will run them like they are ours.
Purchase your NVIDIA GPU servers from any authorized OEM or distributor. Ship them directly to our facility, and our engineering team will handle deployment, networking, cooling, cluster configuration, monitoring, and ongoing support.
We do not make money on your servers. We make money keeping them running.
That is the whole philosophy behind BYOG. You are free to buy your NVIDIA systems from whichever OEM, distributor or reseller gives you the best price, the best lead time and the best warranty terms — and we will still take care of them as if they were on our own balance sheet. No hardware markup, no vendor lock-in, no pressure.
Five steps from purchase order to first training job
You handle the buying decision. From the moment the freight arrives at our dock, every remaining step is ours — and every step is documented.
What our engineering team handles
BYOG is not "rack and forget". It is the same engineering programme we apply to our own fleet, applied to hardware you happen to own.
- Deployment: receiving, inspection, rack and stack, rail kits, torque-controlled install, coolant plumbing for liquid-cooled systems
- Networking: rail-optimised InfiniBand or Spectrum-X fabric design, cabling, optics, link validation, transit, cloud on-ramps
- Cooling & power: air, rear-door heat exchanger or direct-to-chip liquid up to 160 kW per rack, A/B feeds, phase balancing
- Cluster configuration: firmware baselining, OS images, NVIDIA driver and CUDA stack, Slurm or Kubernetes, parallel storage clients
- Monitoring: DCGM, fabric counters, thermal and power telemetry, per-job attribution, alerting into your channels or ours
- Ongoing support: 24×7 NOC, 15-minute P1 response, node drain and replace, OEM RMA management, spares logistics, quarterly optimisation reviews
Indicative schedule for an eight-node air-cooled deployment into an existing Velyrix hall. Liquid-cooled and rack-scale systems follow the same sequence over a longer window.
Procurement assistance — if you want it, never because you have to
Some customers have a preferred OEM and a signed price book already. Others are buying their first GPU cluster and would like a second opinion before committing seven figures. Both are fine. If you ask, we will help; if you do not, nothing about your service changes.
- Configuration review — we sanity-check the BOM against your workload before you sign: CPU-to-GPU ratio, memory, NVMe, NIC count and rail mapping
- Vendor-neutral comparison — how the equivalent Dell, Supermicro, GIGABYTE, HPE and Lenovo platforms differ in practice for your use case
- Lead-time intelligence — what is genuinely shippable this quarter versus what is a roadmap slide
- Introductions — to OEM and distributor account teams in your region, with you owning the relationship and the negotiation
- Logistics support — freight routing, import documentation, delivery scheduling and site access coordination
- Deployment-readiness check — whether the configuration you are about to buy will actually fit the power, cooling and fabric you are planning
Our procurement promise
You are never required to buy hardware through Velyrix to receive hosting, deployment or managed services. Pricing is identical either way.
If you do ask us to transact, we tell you exactly what our fee is before you agree to anything. Most customers buy direct, and we encourage it.
Your OEM account team, your price book, your warranty and your support contracts remain in your name. We work alongside them, not in front of them.
We will tell you when a cheaper configuration is the right one, and when you should delay a purchase for a better platform. Long-term operations is where our business is.
BYOG, rental, or build it yourself
| BYOG at Velyrix | Rent from Velyrix | Build your own data centre | |
|---|---|---|---|
| Hardware ownership | You own it — asset, depreciation, residual value | Velyrix owns it | You own it |
| Capital required | Hardware capex only | None — pure opex | Hardware plus facility capex |
| Vendor choice | Any authorized OEM or distributor | Velyrix fleet | Any |
| Time to production | 2–6 weeks after delivery | Minutes to days | 12–36 months |
| Power & cooling risk | Velyrix | Velyrix | You |
| Who racks and cables it | Velyrix engineers | Velyrix engineers | Your team or a contractor |
| Who answers at 3 a.m. | Velyrix NOC | Velyrix NOC | Your team |
| OEM warranty & RMA | Yours, managed by Velyrix on your behalf | Velyrix | Yours |
| Scaling increment | Your next purchase order | Per GPU-hour | Next building phase |
| Exit | Ship the hardware out, or sell it | End the term | Sell the facility |
Many customers run a hybrid: owned hardware for steady-state load, Velyrix rental capacity for bursts and for evaluating a new GPU generation before buying it.
If it is an authorized NVIDIA platform, we will host it
Velyrix hosts customer-owned equipment sourced through authorized channels. We complete export-control and restricted-party screening before accepting delivery of controlled hardware, and we install to the manufacturer's published procedures so your warranty and support entitlements are preserved. Product names are trademarks of their respective owners.
Three line items, all of them optional except the first
| Component | What it covers | Price | Required? |
|---|---|---|---|
| Colocation | Space, power, cooling, security, cross-connects | $110 – $215 / kW | Yes |
| Deployment & commissioning | Receiving, rack & stack, cabling, firmware, burn-in, acceptance report | from $2,300 | Optional — you may install it yourself |
| Fabric build | InfiniBand or Spectrum-X design, switches, optics, cabling, validation | from $66,000 / 64 | Optional — or bring your own switches |
| Managed operations | 24×7 NOC, monitoring, incident response, RMA handling, firmware lifecycle | from $500 | Optional — remote hands available à la carte |
| Storage & network add-ons | Parallel FS, object storage, transit, cloud interconnect | See pricing page | Optional |
Indicative USD list pricing, exclusive of tax and metered power. Every BYOG engagement is quoted after an engineering review of your specific hardware. See the full price list.
Own the GPUs. Let us run them.
Buy your NVIDIA servers from any OEM or distributor you like, ship them to a Velyrix hall, and we handle the rest — deployment, fabric, cooling, monitoring and support. Or rent ours. Either way, you get a plan in one business day.
Frequently asked questions
What is BYOG (Bring Your Own GPU)?
BYOG means you purchase your NVIDIA GPU servers from any authorized OEM or distributor and ship them directly to a Velyrix data centre. Our engineering team then handles deployment, networking, cooling, cluster configuration, monitoring and ongoing support. You own the hardware; we operate it.
Do I have to buy my servers through Velyrix?
No, and most customers do not. Procurement assistance is available if you want a second opinion on configuration, vendor comparison or lead times, but it is entirely optional and your hosting, deployment and managed service pricing is identical either way. We would rather earn your business operating the hardware for years than take a margin on it once.
Which OEM hardware can I bring?
Any authorized NVIDIA platform: Dell PowerEdge XE9680, XE9712 and XE7745, Supermicro SYS-821GE and A21GE and GB200/GB300 NVL72, GIGABYTE G593, G894 and XL44, HPE Cray XD670 and ProLiant XD685, Lenovo ThinkSystem SR675 V3 and SR685a V3, NVIDIA DGX systems, plus ASUS, QCT, Wiwynn and Foxconn designs.
Does third-party deployment void my OEM warranty?
No, when installation follows the manufacturer's published procedures. Velyrix engineers work to OEM installation guidance with ESD-controlled handling, torque specifications and serial-level documentation, so your warranty and support entitlements remain intact. We also manage the RMA process with your OEM on your behalf.
Who handles a hardware failure at 3 a.m.?
The Velyrix NOC. We detect the fault, drain the node from your scheduler, open the RMA with your OEM under your support contract, fit the replacement part from the on-site spares depot where one is held, and return the node to service - then report what happened.
How long does a BYOG deployment take?
For an air-cooled deployment into an existing Velyrix hall, typically two to six weeks from delivery to signed acceptance, depending on node count and fabric scope. Liquid-cooled and rack-scale systems such as GB300 NVL72 take longer because of coolant commissioning.
Can I move my existing cluster to Velyrix?
Yes. Migration and relocation is a standard service: staged cutover planning, de-install, freight, re-install, re-cabling, asset reconciliation and full re-validation, designed to minimise the time your training runs are offline.