Six things people actually ask us to build.

Rack-scale compute build

A rack or a row, delivered as one unit of work: servers, switching, PDUs, cable management and labelling, assembled and tested before it ships.

Typical scope: 1–12 racks · 4–8 weeks from order

You send the workload profile or the bill of materials. We produce a rack elevation, agree power and cooling envelopes against your facility, then build and test the rack on our floor. It ships either fully racked in a shockwatch-monitored crate, or flat-packed with a build book if the destination lift will not take the crate.

• Rack elevation and power budget
• Structured cabling with both ends labelled
• Firmware baseline across the rack
• As-built documentation and asset register

AI and HPC clusters

Accelerated nodes, InfiniBand or RoCE fabric, high-density power and the thermal work that makes it survivable in a Southeast Asian facility.

Typical scope: 4–128 nodes · 8–20 weeks, allocation dependent

Eight-way SXM nodes, PCIe accelerator platforms, or dense mixed clusters. We handle allocation and lead-time management on the accelerators, build the fabric to a topology you sign off, run a cluster-level validation pass, and confirm the facility can actually take 40–130 kW a rack before anything ships.

• Node and fabric topology, signed off before order
• Allocation tracking on constrained accelerators
• NCCL / IB validation run and report
• Liquid-cooling readiness assessment

Storage and backup platforms

Primary arrays, scale-out nodes, object tiers and the nearline capacity underneath them, sized against a retention policy rather than a wishlist.

Typical scope: 100 TB – 20 PB usable

We size from your capacity and change-rate figures, quote the array alongside the raw media so you can see where the money goes, and pre-populate drives with firmware matched to the controller. Where you are extending an existing fleet, we match drive families and firmware so the array does not spend a week rebuilding.

• Usable-capacity model with overhead shown
• Drive firmware matched to the existing fleet
• Array staged, zoned and tested
• Media disposal plan for what it replaces

Network fabric

Top-of-rack, spine and border switching with the optics, breakouts and adapters that make the link come up the first time it is patched.

Typical scope: single ToR to multi-site fabric

Leaf-spine builds at 25G, 100G and 400G, with EVPN-VXLAN where it earns its place. We stage the switches on our floor with your configuration, verify every optic against the port it will sit in, and ship a patching schedule that a contractor can follow without calling you.

• Port map and patching schedule
• Switch configuration staged and saved
• Optic and DAC compatibility verified per port
• Spare optics held against the deployment

Edge and branch

Short-depth and ruggedised systems for cabinets, comms rooms and sites where nobody is coming to reboot anything.

Typical scope: 5–500 sites

Short-depth chassis, DC-powered options, extended-temperature components and out-of-band management that works over a cellular link. Units ship pre-imaged and pre-addressed so the site visit is a mounting job, not a build.

• Pre-imaged and pre-addressed systems
• Out-of-band management verified before despatch
• Site kit: rails, cables, PDU, labels, in one box
• Spares held centrally against the estate

Refresh and decommission

Replacing a generation without a gap in service, and getting rid of the old estate without a data problem.

Typical scope: 50–5 000 units

We stage the incoming fleet, sequence the swap so capacity never drops below your floor, then take the outgoing kit back. Drives are wiped to NIST SP 800-88 purge or physically destroyed, with certificates issued per serial. Residual value is offset against the new order where the platform still has a market.

• Migration sequence with capacity floor held
• Erasure or destruction certificates per serial
• Buy-back valuation on the outgoing estate
• Chain-of-custody records end to end