| GPU | Tier | Price (eStore B2B net for Supermicro, list in brackets) | CPUs | Memory | Storage | Network | kW nameplate → provisioned | Part | Vendor | Lead time |
|---|---|---|---|---|---|---|---|---|---|---|
| NVIDIA B300 (HGX NVL8) | Tier 1 OEM build | $537,196.40 (list $610,450.46) | 2× 48C Intel | 2TB | 1× 480GB M.2 | 8× 800G | 13.6 → 16.96 | SYS-822GS-NB3RT | Supermicro | 30+ wk; confirm at order |
| NVIDIA B300 (HGX NVL8) | Tier 2 system builder | $537,196.40 (list $610,450.46) | 2× 48C Intel | 3TB | 2× 1.92TB M.2 + 4× 3.84TB | 1× 100G | 13.6 → 16.96 | KR9288X3 | Aivres | 11-20 wk |
| NVIDIA B300 (HGX NVL8) | Tier 2 system builder | $537,196.40 (list $610,450.46) | 2× 64C AMD | 2.3TB | 2× 1.92TB M.2 + 4× 3.84TB | 1× 100G | 13.6 → 16.96 | KR9288E3 | Aivres | 16-20 wk on a November order |
| NVIDIA H200 (HGX) | Tier 1 OEM build | $318,117.39 (list $361,497.03) | 2× 48C Intel | 1TB | 1× 240GB SATA | 2× 25G | 9.1 → 11.34 | SYS-821GE-TNHR | Supermicro | not recommended for new designs (reference only); 6-10 wk |
| NVIDIA RTX PRO 6000 Server Edition | Tier 1 OEM build | $177,430.31 (list $201,625.35) | 2× 72C Intel | 1TB | 2× 480GB + 1× 3.84TB | 2× 10G | 6.8 → 8.47 | SYS-422GA-NRT | Supermicro | 3-5 wk |
| NVIDIA RTX PRO 6000 Server Edition | Tier 2 system builder | $163,909.60 | 2× 96C AMD | 768GB | 1× 3.84TB + 2× 1TB | 2× 25G | 6.7 → 8.37 | 4U-E10G-GENOAX2 | Octoserver | 3-5 wk |
| NVIDIA RTX 5090 (passive server edition) | Tier 2 system builder | $76,679.60 | 2× 96C AMD | 512GB | 1× 3.84TB + 1× 1TB | 2× 25G | 6.4 → 7.99 | 4U-E10G-GENOAX2 | Octoserver | 5-6 wk |
| Host | Price (eStore B2B net for Supermicro, list in brackets) | Slots | CPUs | Memory | Storage | Network | kW nameplate → provisioned | Part | Vendor | Lead time |
|---|---|---|---|---|---|---|---|---|---|---|
| 8-GPU host, Tier 1 (Supermicro), PCIe 5.0, Intel Xeon 6 (TDX) | $47,935.10 (list $54,471.71) | no GPUs, 8 slots | 2× 72C Intel | 512GB | 2× 960GB | 2× 25G | 1.1 → 1.38 | SYS-422GA-NRT | Supermicro | 3-5 wk |
| 8-GPU host, Tier 1 (Supermicro), PCIe 5.0, Intel 5th Gen Xeon (TDX) | $41,472.73 (list $47,128.10) | no GPUs, 8 slots | 2× 28C Intel | 512GB | 2× 960GB | 2× 25G | 1.7 → 2.19 | SYS-521GE-TNRT | Supermicro | 3-5 wk |
| 8-GPU host, Tier 1 (Supermicro), PCIe 5.0, AMD | $40,244.08 (list $45,731.91) | no GPUs, 8 slots | 2× 32C AMD | 512GB | 1× 960GB | 2× 25G | 1.0 → 1.25 | AS-5126GS-TNRT | Supermicro | 3-5 wk |
| 8-GPU host, Tier 2 (Octoserver), PCIe 5.0, AMD | $32,919.60 | no GPUs, 8 slots | 2× 32C AMD | 512GB | 2× 960GB | 2× 25G | 1.1 → 1.31 | 4U-E10G-GENOAX2 | Octoserver | 2-3 wk |
| 8-GPU host, Tier 2 (Octoserver), PCIe 4.0 budget, AMD | $15,146.20 | no GPUs, 8 slots, PCIe 4.0 | 2× 32C AMD | 512GB DDR4 | 2× 960GB | 2× 25G | 1.1 → 1.34 | 4U-E10G-ROMEX2 | Octoserver | 2-3 wk |
| 8-GPU barebone, OneChassis 7U, PCIe 4.0, DDR4, AMD (retail RTX 5090) | $4,585 | no GPUs, 8 slots (10 dual-slot), PCIe 4.0 | no CPUs (2× SP3, EPYC 7002/7003) | no memory (32× DDR4-3200) | 12× 3.5" bays, no drives | none (NIC risers; board 2× 10GbE) | — | 7U GPU chassis, PCIe 4.0 risers + ROME2D32GM-2T | OneChassis | 4-6 wk plus freight from Shenzhen |
| 8-GPU barebone, OneChassis 7U, PCIe 5.0, DDR5, AMD (retail RTX 5090) | $5,285 | no GPUs, 8 slots (10 dual-slot), PCIe 5.0 | no CPUs (2× SP5, EPYC 9004/9005) | no memory (24× DDR5) | 12× 3.5" bays, no drives | none (NIC risers; board 2× 1GbE) | — | 7U GPU chassis, PCIe 5.0 risers + TURIN2D24G-2L+ | OneChassis | 3-4 wk plus freight from Shenzhen |
| 8-GPU barebone, OneChassis 7U, PCIe 4.0, DDR4, AMD (RTX PRO 6000 fan kit) | $4,785 | no GPUs, 8 slots (10 dual-slot), PCIe 4.0 | no CPUs (2× SP3, EPYC 7002/7003) | no memory (32× DDR4-3200) | 12× 3.5" bays, no drives | none (NIC risers; board 2× 10GbE) | — | 7U GPU chassis, PCIe 4.0 risers, Pro 6000 fan kit + ROME2D32GM-2T | OneChassis | 4-6 wk plus freight from Shenzhen |
| 8-GPU barebone, OneChassis 7U, PCIe 5.0, DDR5, AMD (RTX PRO 6000 fan kit) | $5,485 | no GPUs, 8 slots (10 dual-slot), PCIe 5.0 | no CPUs (2× SP5, EPYC 9004/9005) | no memory (24× DDR5) | 12× 3.5" bays, no drives | none (NIC risers; board 2× 1GbE) | — | 7U GPU chassis, PCIe 5.0 risers, Pro 6000 fan kit + TURIN2D24G-2L+ | OneChassis | 3-4 wk plus freight from Shenzhen |
| 8-GPU barebone, OneChassis 4U OCG7466M6, PCIe 5.0, DDR5, AMD (RTX PRO 6000) | $6,100 | no GPUs, 10 dual-slot slots, PCIe 5.0 | no CPUs (2× SP5, EPYC 9004/9005) | no memory (24× DDR5-6400) | 12× 3.5" bays + 2× M.2, no drives | none (OCP 3.0 slot; BMC port) | — | OCG7466M6 + TURIN2D24G-2L+ | OneChassis | 4-6 wk plus freight from Shenzhen |
| Ports | Price (FS public) | $/port | W nameplate | Provision kW | Part | Lead time | Spec |
|---|---|---|---|---|---|---|---|
| 64× 800G OSFP | $44,620 | $697 | 1,800 | 2.25 | N8650-64OD, PicOS | ~1 week | 64x 800G OSFP (or 128x 400G breakout); Broadcom Tomahawk 5, 51.2T; PicOS: RoCEv2, EVPN-VXLAN, MLAG; 5-yr warranty; options: Same spec on Dell Z9864F-ON / Enterprise SONiC quote; Same spec on Arista 7060X6-64PE / EOS quote; NVIDIA Spectrum-4 SN5600 / Cumulus quote |
| 32× 800G OSFP | $37,500 | $1,172 | 1,000 | 1.25 | N8650-32OD, PicOS | ~1 week | 32x 800G OSFP (or 64x 400G breakout); Broadcom Tomahawk 5, 25.6T; PicOS; options: Dell / Arista / NVIDIA equivalent quote |
| 64× 400G QSFP-DD | $24,000 | $375 | 1,000 | 1.25 | N8610-64DB, PicOS | ~1 week | 64x 400G QSFP-DD; Broadcom Tomahawk 4, 25.6T; PicOS; options: Dell Z9664F-ON / Enterprise SONiC quote; Arista 7060DX5-64S / EOS quote |
| 32× 400G QSFP-DD | $20,000 | $625 | 600 | 0.75 | N8610-32D, PicOS | ~1 week | 32x 400G QSFP-DD; Broadcom Tomahawk 4, 12.8T; PicOS; options: FS N5680-32D (Tomahawk 3) budget variant −$700; Dell Z9432F-ON / Enterprise SONiC quote; Arista 7060DX5-32 / EOS quote |
| 64× 100G QSFP28 | $11,930 | $186 | 550 | 0.69 | N5630-64C, PicOS | ~1 week | 64x 100G QSFP28; Broadcom Tomahawk 2, 6.4T; PicOS; options: Dell Z9264F-ON / Enterprise SONiC quote; Arista 7060CX2-32S class quote |
| 32× 100G QSFP28 | $7,919 | $247 | 250 | 0.31 | N8560-32C, PicOS | 1 day to ~1 week | 32x 100G QSFP28; Broadcom Trident 3-X7, 3.2T; PicOS; options: Dell S5232F-ON / Enterprise SONiC quote; Arista 7050CX3-32S / EOS quote |
| 48× 25G + 8× 100G | $6,049 | $126 | 150 | 0.19 | S5470-48Y, PicOS | ~1 week | 48x 25G SFP28 + 8x 100G QSFP28; PicOS, MLAG; options: FS S5580-48Y (Trident 3, EVPN-VXLAN) +$1,700; Dell S5248F-ON / Enterprise SONiC quote; Arista 7050SX3-48YC8 / EOS quote |
| 48× 10G + 8× 100G | $4,569 | $95 | 120 | 0.15 | S5860-48SC | 1 day to ~10 days | 48x 10G SFP+ + 8x 100G QSFP28; FS OS rather than PicOS (PicOS 10G options: S5860-20SQ, S5440-12S) |
| 48× 10GBASE-T + 4× 25G + 2× 40G | $5,659 | $118 | 200 | 0.25 | S5860-48XMG | 1 day to ~10 days | 48x 1/2.5/5/10GBASE-T RJ45 + 4x 25G SFP28 + 2x 40G QSFP+; Cat6a to 100 m, no transceivers or DACs; L3 managed, FS OS; the PicOS build of the same box is $5,749 (4 in US stock 2026-10-05); For hosts with on-board 10GBASE-T (the ROME2D32GM-2T in the PCIe 4.0 DIY node has 2x): the simplest front end for an independent-GPU farm; options: PicOS build (S5860-48XMG PicOS, 1-yr licence included) +$90 |
| 48× 1G RJ45 + 4× 10G | $989 | $21 | 60 | 0.07 | S3410-48TS, PicOS | ~1 week | 48x 1GbE RJ45 + 4x 10G SFP+ uplinks; PicOS L2+ managed (VLAN, LACP, MLAG); out-of-band / BMC / IPMI layer; Fanless-class 1U, redundant PSU not standard; 5-yr warranty; options: Budget: S3540L-48T, 48x 1G + 4x 10G, PicOS L2+ (369 in stock) −$480; L3: S5810-48TS, 48x 1G + 4x 10G, PicOS L3 (US warehouse) +$490; 24-port: S3410-24TS, 24x 1G + 4x 10G, PicOS L2+ −$290; PoE+: S3410-48TS-P, 48x 1G PoE+ + 2x 10G + 2x 1G +$530; 48x 1G SFP fiber: S5810-48FS, PicOS L3 +$1,440; Non-PicOS (FS OS) same hardware: S3910-48TS L2+ $1,049 / S5810-48TS L3 $1,509 quote |
| Speed | 2 m passive DAC | Passive reach | Active DAC / AEC | AOC 10 m (max length) | Short-reach optic, each (reach) | Long-reach optic, each (reach) | Breakout cable |
|---|---|---|---|---|---|---|---|
| 800G | $239 | 3 m | AEC 5 m $1,424 (max 5 m) | $3,011 (50 m) | SR8 $879 (50 m) | DR8 $1,319 (500 m) · 2×FR4 $1,869 (2 km) | 800G→2×400G 2 m $299 |
| 400G | $174 | 3 m | Active DAC 5 m $611 (max 5 m) | $1,055 (30 m) | SR8 $219 (100 m) | DR4 $549 (500 m) · FR4 $719 (2 km) · LR4 $999 (10 km) | 400G→4×100G 2 m $199 |
| 200G | $120 | 3 m | Active DAC 7 m $325 (max 7 m) | $1,034 (50 m) | SR4 $299 (100 m) | FR4 $1,249 (2 km) · LR4 $2,624 (10 km) | 200G→2×100G 2 m $119 |
| 100G | $54 | 5 m | Active DAC 9 m $300 (max 9 m) | $221 (30 m) | SR4 $99 (100 m) | CWDM4 $209 (2 km) · LR4 $399 (10 km) | 100G→4×25G 3 m $136 |
| 25G | $39 | 5 m | Active DAC 9 m $73 (max 9 m) | $73 (30 m) | SR $49 (100 m) | LR $74 (10 km) | — |
| 10G | $18 | 7 m | Active DAC 10 m $105 (max 10 m) | $39 (30 m) | SR $25 (300 m) | LR $34 (10 km) | — |
| 1G | RJ45, Cat6 patch | — | — | — | 10G SFP+ uplinks: see 10G row | — |
10G over copper (RJ45, Cat6a to 100 m): the 48× 10GBASE-T leaf is listed (S5860-48XMG); for an SFP+ switch, FS's 10GBASE-T SFP+ module is $86 (30 m) or $149 (80 m), $169 for 100 m (Cisco-coded shown; FS/generic coding at the same list), so 48 copper ports on the SFP+ leaf cost $4,128 in modules on top of the $4,569 switch. Hosts with on-board 10GBASE-T (ASRock Rack ROME2D32GM-2T, 2 ports) need no NIC. Second-hand: Arista 7050TX-64 at about $175 (rate-card page, market parts).
| Cable | 1 m | 3 m | 10 m | 30 m | 50 m | Used with |
|---|---|---|---|---|---|---|
| LC-LC duplex OM4 | $8.40 | $10 | $20 | $47 | — | SR (10G, 25G) |
| MTP-12 OM4 | $59 | $89 | $150 | $301 | $452 | SR4, SR8 (800G SR8 is dual MPO-12/APC: 2 per link) |
| LC-LC duplex OS2 | $8.40 | — | — | — | — | FR4, CWDM4, LR, LR4 |
| Distance | Use | Cost per link |
|---|---|---|
| Inside passive reach | 2 m passive DAC | 1 cable |
| To active max | Active DAC or AEC | 1 cable |
| To AOC max, or to short-reach optic reach | AOC, or 2× short-reach optic + multimode fibre | 1 cable, or 2 optics + fibre |
| Beyond | 2× long-reach optic + single-mode fibre | 2 optics + fibre |
| Any | Connected cost per port = switch $/port + link |
Choosing a link by distance: 2 m passive DAC inside passive reach; active DAC or AEC up to its maximum; beyond that an AOC (10 m priced; longer lengths cost more) or two short-reach optics plus multimode fibre out to the optic's reach; past that, two long-reach optics over single-mode. At 100G and up the optics, not the fibre, set the cost. Connected cost per port = switch $/port + one cable, or + two optics plus fibre. GPU-fabric links are estimated all-optical: GPU node density (1–3 HGX nodes per 40 kW rack) puts most nodes past passive-copper reach of their leaf, since a 3 m DAC spans at most the adjacent rack. Mixing copper in is realistic — expect 25% or more of fabric cables to land as passive DAC once installation is engineered (peer links, co-racked leaf↔spine, nodes in or beside the leaf's rack) — but which links qualify depends on rack elevations, so estimates stay optical and copper is taken as a savings at engineering time.
| PicOS | 100G | 400G | 800G |
|---|---|---|---|
| Licence | included | included | included |
| 1-yr support | $369 | $439 | $659 |
| Dell/SONiC or Arista, same ports | quote | quote | quote |
| Product | Price (eStore B2B net for Supermicro, list in brackets) | Form | CPUs | Memory | Storage | Network | kW nameplate → provisioned | Part | Vendor | Lead time |
|---|---|---|---|---|---|---|---|---|---|---|
| 1U compute node | $34,523.31 (list $39,231.03) | 1U | 2× 32C Intel | 512GB | diskless, 12 NVMe bays | 2× 25G | 1.0 → 1.2 | SYS-122C-TN | Supermicro | 3-5 wk |
| Product | Price (eStore B2B net for Supermicro, list in brackets) | Form | CPUs | Memory | Storage | Network | kW nameplate → provisioned | Part | Vendor | Lead time |
|---|---|---|---|---|---|---|---|---|---|---|
| 2U all-NVMe storage node | $177,958.91 (list $202,226.03) | 2U | 2× 32C Intel | 512GB | 24× 7.68TB NVMe, 184TB | 2× 100G | 1.3 → 1.65 | SYS-222C-TN | Supermicro | 3-5 wk |
| 2U HDD storage node | $56,013.71 (list $63,651.94) | 2U | 1× 16C AMD | 128GB | 24× 24TB HDD, 576TB | 2× 100G | 0.6 → 0.78 | ASG-2015S-E1CR24H | Supermicro | 3-5 wk |
| Item | Rate | Typical / basis (list evidence) |
|---|---|---|
| DDR5-6400 RDIMM 32GB (Supermicro) | $50.6/GB | $1,840 per DIMM (now resolves to MEM-DR532L-CL02-ER64), identical on every chassis probed 2026-11-03; was $1,635 in October and $1,474 on 09-09 |
| DDR5-6400 RDIMM 64GB (Supermicro) | $42.31/GB | $3,077 per DIMM on 2026-11-03 (was $2,791 in October, $2,530 in September); cheapest per GB |
| DDR5-6400 RDIMM 96GB (Supermicro) | $48.06/GB | $5,243 per DIMM (MEM-DR596NC-ER64) on 2026-11-03; was $4,901 in October |
| DDR5-6400 RDIMM 128GB (Supermicro) | $49.26/GB | $7,166 per DIMM (MEM-DR512PC-ER64) on 2026-11-03 (was $6,697); 256GB $16,574 ($64.74/GB); 128GB/256GB still unavailable on AS-5126GS-TNRT |
| DDR5-4800 RDIMM 32GB (Octoserver) | $39.53/GB | $1,265 per DIMM |
| DDR4-3200 RDIMM 32GB (Octoserver Rome host) | $13.44/GB | $430 per DIMM |
| AMD EPYC 9005/9004, mainstream 16-96C (Supermicro) | $27.28–83.6/core | $58.08/core |
| Intel Xeon 6700P, mainstream 32-64C (Supermicro) | $88.88–146.08/core | $109.12/core |
| Intel Xeon 6900P, 72-128C (Supermicro) | $51.04–72.16/core | $57.2/core |
| AMD EPYC 9004/9005 at Octoserver | $29–51/core | $42/core |
| AMD EPYC 7003 at Octoserver (PCIe 4.0 host) | $13–29/core | $20/core |
| Nearline HDD 16-30TB SAS/SATA (Supermicro) | $46.64–71.28/TB | $68.64/TB |
| NVMe Gen4 U.2 TLC 1.9-15TB (Supermicro) | $693.44–1261.92/TB | $862.4/TB |
| NVMe Gen5 U.2/E1.S/E3.S (Supermicro) | $683.76–983.84/TB | $753.28/TB |
| SATA/SAS SSD (Supermicro) | $877.36–1795.2/TB | $880.0/TB |
| Exotic parts: Xeon 6768P/6788P/6748P, 3DS DIMMs, dual-port drives | quote |
Example: 1TB → 2TB with 64GB DIMMs = 1,024 GB × $42.31 = $43,325.44 (the H200 line lists this option at +$43,324.16). Rates marked (Supermicro) are shown at eStore B2B net (12% off the eStore list prices quoted in each basis), the same basis as the Supermicro base builds. Marginal rates from non-exotic SKUs, taken from the eStore configurator (the price its cart charges). Within a family the price rises almost linearly with cores and capacity, so alternative configurations are not priced separately: apply these rates to the base build. Supermicro's drive pricing is high: buyers who source NVMe or HDD elsewhere should price drives from the market and the chassis from here. Exotic parts (Xeon 6768P/6788P/6748P at $250-320/core, 3DS 256GB DIMMs, dual-port or FIPS drives) sit outside these bands. October 2026 reset (re-probed 2026-10-05): DIMMs +10.9%, M.2 boot drives +20%, NVMe +8-20%, nearline HDDs +8.7%; CPUs, NICs, TPM, DCMS, GPU cards and chassis remainders unchanged. November 2026 reset (re-probed 2026-11-03): 32GB DIMM +12.5%, 64GB +10.2%, 960GB M.2 +9.9%, 24TB HDD +16%, 7.68TB PM9D3a unchanged; 96GB +7.0% ($5,243), 128GB +7.0% ($7,166), 22TB HDD +25% ($1,725); 20TB HDD unchanged; CPUs and NICs unchanged.
| Part | Median | Range | Fits | Note |
|---|---|---|---|---|
| AMD EPYC 7543 32C 2.8 GHz (Milan, SP3, PCIe 4.0, 225 W) | $595 | $367–$730 | ROME2D32GM-2T | The volume 32-core Milan part; 7543P (single-socket) excluded |
| AMD EPYC 7532 32C 2.4 GHz (Rome, SP3, 200 W) | $318 | $243–$560 | ROME2D32GM-2T | Cheapest 32-core for the DDR4 board; 256 MB L3 |
| AMD EPYC 7742 64C 2.25 GHz (Rome, SP3, 225 W) | $680 | $475–$710 | ROME2D32GM-2T | 64 cores for the price of a 7543; the high-volume Rome part |
| AMD EPYC 7713 64C 2.0 GHz (Milan, SP3, 225 W) | $1,175 | $900–$1,499 | ROME2D32GM-2T | |
| AMD EPYC 9354 32C 3.25 GHz (Genoa, SP5, PCIe 5.0, 280 W) | $1,695 | $1,695–$1,926 | TURIN2D24G-2L+/500W | Thin market; Supermicro configurator lists the same part at $1,739 new, so there is no second-hand discount on Genoa yet |
| AMD EPYC 9654 96C 2.4 GHz (Genoa, SP5, 360 W) | $2,600 | $2,400–$2,850 | TURIN2D24G-2L+/500W | Octoserver lists it at $2,780 |
| 32GB DDR4-3200 ECC RDIMM (PC4-25600R, Samsung/Hynix/Micron) | $240 | $198–$281 | ROME2D32GM-2T | $7.50/GB; Octoserver $430 ($13.44/GB) |
| 32GB DDR5-4800 ECC RDIMM (PC5-4800B) | $800 | $700–$950 | TURIN2D24G-2L+/500W | $25/GB second-hand; Octoserver $1,265, Supermicro DDR5-6400 $1,635. DDR5-5600 32GB sold at a $1,000 median |
| Samsung PM9A3 3.84TB U.2 PCIe 4.0 NVMe (MZQL23T8HCLS) | $533 | $375–$900 | any U.2 bay | Wide range: used drives with 70-80% health at the bottom, new-in-box around $1,000-1,100; other Gen 4 3.84TB U.2 (Kioxia CD6/CM7, Intel P5510) sold at ~$950 median. Octoserver $2,050 |
| NVIDIA GeForce RTX 5090 32GB, retail board-partner card (3.5-4-slot, 575 W) | $5,999 | $5,200–$7,450 | OneChassis 7U | MSRP $1,999 (NVIDIA, Jan 2025); first-party stock has largely evaporated and the purchasable price is 2.5-4x MSRP. The low end of the range is in-store pickup when a shop has stock; the high end is online marketplace stock that ships today |
| NVIDIA RTX PRO 6000 Blackwell 96GB (Server Edition, passive 600 W; Workstation and Max-Q within ~$500) | $15,000 | $13,000–$18,000 | OneChassis 4U, OneChassis 7U | Octoserver sells the passive card at $14,800 in stock, inside this range; Supermicro's accessory card is out of stock. Data-center driver licence included, unlike GeForce. Unsolicited broker allocation offers above list appeared in October 2026; the p75 stays $18,000 |
| Arista DCS-7050TX-64 (used), 48× 1/10GBASE-T RJ45 + 4× 40G QSFP+, 1U, EOS | $175 | $129–$259 | front-end for PCIe farms | The cheapest 10GBASE-T leaf there is: a 2016-era Arista, end-of-sale, no support contract or EOS image downloads without an Arista account, ~150 W, loud; two of them front a 64-node farm for under $500 against $11,318 for two new FS S5860-48XMG. Check PSU count and airflow direction per listing |
Second-hand and street prices for the parts a build-your-own barebone still needs (CPUs, memory, boot NVMe, GPU cards). eBay figures are US sold listings read 2026-10-05 (newest 120 per search), filtered to single units, excluding vendor-locked (Dell/Lenovo/HPE), QS/ES engineering samples, lots and bundles; the median of the sold prices with the interquartile range and the count. These are planning figures for the DIY worked designs only: the catalog's product lines carry vendor prices, and a buyer's own purchase will land somewhere in the range. Used (pulled) server parts carry no warranty; new-in-box listings sit at the top of each range. GPU card prices come from retailers, price trackers and eBay together (sources in each row); eBay alone is unreliable for GPUs. Worked example using them: the DIY 8× RTX 5090 build on the designs page.
| Rule | Value |
|---|---|
| Provisioned kW | nameplate W ÷ 0.8 (no circuit above 80%) |
| Feeds | A + B, each sized for the whole box |
| Colo | billed on provisioned kW at the location's rate (table below); Montana is the default for every computed figure |
| Circuit | Rated kW |
|---|---|
| 208V 1-ph 20A | 4.2 |
| 208V 1-ph 30A | 6.2 |
| 208V 3-ph 30A | 10.8 |
| 208V 3-ph 50A | 18.0 |
| 208V 3-ph 60A | 21.6 |
| 415V 3-ph 30A | 21.6 |
| 415V 3-ph 60A | 43.1 |
| 415V 3-ph 100A | 71.9 |
| Location | $/kW/month | Term | What UpStation has seen | State sales tax on the hardware | State property tax on the hardware | Facility |
|---|---|---|---|---|---|---|
| Montana | $260 | 36 months | $260 quoted for a 36-month term, October 2026 ($220 once, on released space); 30 kW air-cooled racks; 3% off for 12+ months prepaid | none | none on the first $1M of equipment, about 0.8%/yr above | Tier III-class telco colo, 30 kW air-cooled racks: room confirmed for small deployments (4 × 30 kW); about 480 kW offered for a January 2027 start (October 2026, offered to several buyers), so larger designs are subject to confirmation on the introduction |
| Texas | $225 | 36 months | $225 all-in (space and power) at a 60 kW commit, quoted 2026 | 8.25% (6.25% state + local); the data-center exemption needs $200M and 20 jobs | about 2.2%/yr of market value | Tier III, blended IP on site |
| Oregon | $175 | 36 months | $175 at a Tier II facility (power, cooling, UPS, no battery); other outreach came back without capacity, 2026 | none | about 1.5-2%/yr of market value (ORS 308.290); abatements closed to data-center property since 2026 | Tier II (no battery); Tier III capacity scarce into 2027 |
| Pennsylvania | $250 | 36 months | $250 at a certified data center, reported second-hand, October 2026 | 6% (8% Philadelphia, 7% Allegheny), or none if the facility is state-certified and the tenant holds its own exemption certificate (100 kW+, 2-year contract) before buying | none | Certified data center assumed |
| Canada (Ontario, Quebec) | $225 | 36 months | $225 on 3 years and $175 on 5 years from several operators; short-term available | HST 13% (Ontario) or GST+QST (Quebec) on hardware and colo, recovered as an input tax credit by a registered business | none | Tier III; cheapest power in Quebec (hydro) |
| Self-hosted / DIY / custom facility (PCIe-class servers: RTX PRO 6000, RTX 5090) | $50–$100 | n/a | Not a price an operator can buy: the running cost, per provisioned kW per month, of a facility the operator builds and operates themselves; where it lands in the range depends on how far they go | by state (rows above) | by state (rows above) | Low end: free cooling, basic UPS, one or two network providers. High end: commercial or data-center air conditioning, redundant UPS, two or more network providers, physical security, backup generators |
USD per provisioned kW per month, billed on nameplate ÷ 0.8. Rates are what UpStation has seen in each market (a quote, a second-hand report, or several operators); the site introduction gets the current quote. Virginia was quoted at $382 (Tier IV) and is not modelled.
| Service | Monthly | Seen |
|---|---|---|
| 1 Gbps blended IP transit | $1,000 | Montana, 2026; in line with other markets |
| 10 Gbps blended IP transit | $2,500 | Montana, 2026; in line with other markets |
| Blended IP per Mbps on a 1 Gbps commit, 10G port | $0.65 | Texas quote, 2026 ($650/month for 1,000 Mbps; port and cross-connect included) |
| IP block /28 (16 addresses), per address | $4 | Texas quote, 2026 |
| Carrier cross-connect for a 100GE circuit, each | $250 | Texas quote, 2026 |
| Location | Colo/month | Colo over 36 months | State sales tax on the hardware (one-time) | State property tax on the hardware, 3 years | Hardware + colo + taxes, 36 months |
|---|---|---|---|---|---|
| Montana | $4,410 | $158,760 | $0 | $0 | $695,956 |
| Texas | $3,816 | $137,376 | $44,319 | $35,455 | $754,346 |
| Oregon | $2,968 | $106,848 | $0 | $28,203 | $672,247 |
| Pennsylvania | $4,240 | $152,640 | $0 | $0 | $689,836 |
| Canada (Ontario, Quebec) | $3,816 | $137,376 | $0 | $0 | $674,572 |
Sales tax is one-time on the hardware; property tax is per year on market value (taken flat for three years, before depreciation); income and margin taxes depend on profit and are not in the total. Pennsylvania assumes the certified-data-center exemption; Canada's HST is recovered by a registered business and is shown as $0.
Nameplate max for PCIe GPU servers = (GPU TDP x count + CPU TDP x 2 + 8 W per DIMM + 15 W per drive + 400 W fans/board/NICs) / PSU efficiency (Titanium 0.94, Platinum 0.92); HGX servers use the vendor configurator total, which matches NVIDIA system guidance. Vendor configurator totals for PCIe boxes are kept in each product's basis as a cross-check only (Supermicro pads chassis overhead by 1.5-2.4 kW; Octoserver omits fans and PSU loss). Minimum circuit kW = nameplate / 0.8 so the circuit never runs above 80% (NEC continuous). Pick the smallest standard circuit whose rated kW (V x A x sqrt3 for 3-phase) >= that minimum; provision it twice (A+B) so either feed carries the box alone. Colo kW to buy = nameplate / 0.8; monthly colo = that kW x the location's rate (colo_locations; Montana $260 is the default, 36-month term).
| GPU server | CapEx | OpEx per month | 3-year total | 5-year total | $/GPU-hour, 3 years | $/GPU-hour, 5 years |
|---|---|---|---|---|---|---|
| 8× NVIDIA B300 (HGX NVL8) (colo) | $537,196–$581,515 | $3,751–$4,801 | $672,247–$754,346 | $762,281–$869,567 | $3.20–$3.59 | $2.18–$2.48 |
| 8× NVIDIA H200 (HGX) (colo) | $318,117–$344,362 | $2,448–$3,135 | $406,261–$457,212 | $465,023–$532,445 | $1.93–$2.17 | $1.33–$1.52 |
| 8× NVIDIA RTX PRO 6000 Server Edition (colo or self-hosted) | $163,910–$192,068 | $418–$2,231 | $178,976–$272,386 | $189,020–$325,931 | $0.85–$1.30 | $0.54–$0.93 |
| 8× NVIDIA RTX 5090 (passive server edition) (colo or self-hosted) | $76,680–$83,006 | $400–$2,077 | $91,062–$152,786 | $100,650–$201,324 | $0.43–$0.73 | $0.29–$0.57 |
| 8× RTX 5090, OneChassis 7U barebone (PCIe 5.0, DDR5) (DIY, parts at market; colo or self-hosted) | $66,800–$72,311 | $386–$2,005 | $80,678–$139,171 | $89,930–$187,076 | $0.38–$0.66 | $0.26–$0.53 |
| 8× RTX 5090, OneChassis 7U barebone (PCIe 4.0, DDR4) (DIY, parts at market; colo or self-hosted) | $58,140–$62,937 | $380–$1,976 | $71,820–$129,276 | $80,940–$176,700 | $0.34–$0.61 | $0.23–$0.50 |
| 8× RTX PRO 6000 Server Edition, OneChassis 4U barebone (DIY, parts at market; colo or self-hosted) | $139,623–$151,142 | $399–$2,075 | $153,987–$224,995 | $163,563–$274,230 | $0.73–$1.07 | $0.47–$0.78 |
Low = the cheapest listed build of that GPU at the cheapest location; high = the dearest build at the dearest location. For PCIe-class servers (RTX PRO 6000, RTX 5090) the range also includes the self-hosted class at $50–$100/kW/month, a cost the operator builds rather than buys; HGX servers are colo only. CapEx = hardware price plus the location's sales tax. OpEx = colo at the location's all-in $/kW rate on provisioned kW, plus the location's property tax on the hardware spread monthly (flat, before depreciation). Totals = CapEx + months × OpEx. $/GPU-hour = total ÷ (GPUs × 8,760 hours per year × years) at 100% utilisation, so it is the floor a GPU has to earn per hour; internet, financing, software, staff and uplift are not included.
| GPU | Term | $/GPU-hour | What UpStation has seen |
|---|---|---|---|
| B300 (HGX NVL8) | 5-year, large scale (2,000-10,000+ GPUs) | $3.60–$3.70 | Top of the market in scale and bottom in price: one offtaker signed about 10,000 B300 GPUs at this level in early October 2026, 20% down, with late-delivery penalties flowing back to back to the supplier; it wants a signed DC agreement and an OEM ship-date letter before it contracts, and says NVIDIA is quoting about six months on any new order |
| B300 (HGX NVL8) | 5-year, mid and small scale (8-120 nodes) | $3.90–$4.25 | Operators signing at about $4.20 with 20% down in October 2026; 5-year is the term operators prefer |
| B300 (HGX NVL8) | 3-year | $4.15–$5.00 | A wide range across operators and offtakers; operators dislike 3-year and price it up (one targets $4.70), 20-25% down, strong SLA and 24/7 operations expected; July 2026 demand was $3.80-4.60 on 1-3 year terms |
| B300 (HGX NVL8) | Short-term / on-demand | $8.00 | About $8 in October 2026; on-demand capacity is sold out on the marketplaces, which were quoting $5+ in July |
| B200 (HGX) | 1-3 year | $3.90–$4.50 | July 2026 demand with zero B200 nodes available in market; a $7 short-term B200 contract was signed the same month. Priced close to B300 because of scarcity |
| GB300 NVL72 | 5-year | $4.50 | One operator's 5-year ask, 20% down, October 2026 (rack-scale; not in this catalog) |
| Vera Rubin NVL (forward) | 5-year, forward | $7.60 | One operator's forward ask for 2027 capacity, October 2026 |
| RTX 5090 (8-GPU PCIe nodes) | 1-2 year, 25% prepaid | $0.50–$0.63 | US data center, October 2026; marketplaces report high 5090 utilisation and no spare supply; enterprise renters want identical node specs across the lot and a 24-hour proof-of-concept node first |
| RTX 5090 (8-GPU PCIe nodes) | Short-term | $0.80 | Short-term workload, October 2026 |
Rates are per GPU-hour on the whole node at the stated term, for capacity delivered December 2026 to January 2027 unless dated otherwise. Compare with the $/GPU-hour floor in the total-cost table above: that floor is hardware plus colo at 100% utilisation and excludes internet, financing, software, staff and margin. UpStation does not sell or broker rental; the table is here so a buyer's payback math uses current numbers.
Source: Operator, marketplace and offtaker conversations, July-October 2026 (UpStation's own supply-side threads); not a published price list.
| GPU server | RU | Provisioned kW | Per 40 kW rack | Rack kW | Binds on |
|---|---|---|---|---|---|
| 8x NVIDIA B300 (HGX B300 NVL8), Intel | 8 | 16.96 | 2 | 33.92 | power |
| 8x NVIDIA B300 (HGX B300 NVL8), Intel, 3TB (Aivres) | 10 | 16.96 | 2 | 33.92 | power |
| 8x NVIDIA B300 (HGX B300 NVL8), AMD, 2.3TB (Aivres) | 10 | 16.96 | 2 | 33.92 | power |
| 8x NVIDIA H200 (HGX H200), Intel | 8 | 11.34 | 3 | 34.02 | power |
| 8x NVIDIA RTX PRO 6000 Blackwell Server Edition, Intel | 4 | 8.47 | 4 | 33.88 | power |
| 8x NVIDIA RTX PRO 6000 Blackwell Server Edition, AMD | 4 | 8.37 | 4 | 33.48 | power |
| 8x NVIDIA RTX 5090 (passive server edition), AMD | 4 | 7.99 | 5 | 39.95 | power |
| 8-GPU host, Tier 1 (Supermicro), PCIe 5.0, Intel Xeon 6 (TDX) | 4 | 1.38 | 10 | 13.8 | space |
| 8-GPU host, Tier 1 (Supermicro), PCIe 5.0, Intel 5th Gen Xeon (TDX) | 5 | 2.19 | 8 | 17.52 | space |
| 8-GPU host, Tier 1 (Supermicro), PCIe 5.0, AMD | 5 | 1.25 | 8 | 10.0 | space |
| 8-GPU host, Tier 2 (Octoserver), PCIe 5.0, AMD | 4 | 1.31 | 10 | 13.1 | space |
| 8-GPU host, Tier 2 (Octoserver), PCIe 4.0 budget, AMD | 4 | 1.34 | 10 | 13.4 | space |
Full rules, feed-rate tables, protection math and the questions to settle first: How solutions are sized. Worked designs computed with them (a 4× B300 cluster with storage; 2 PB usable on HDD nodes): Reference builds and worked designs.
UpStation is paid by suppliers through referral and partner programs where they exist; it adds no markup and takes no part in the transaction. The prices shown are the suppliers' own. Book a meeting when a design is ready.
| Supplier | Kind | Pricing on this site | What | How to buy / terms | Lead time | Notes |
|---|---|---|---|---|---|---|
| Supermicro eStore | OEM web store (Tier 1) | Public: listed exactly (eStore price book) | Gold Series ready-to-ship systems and configure-to-order chassis; the price basis for every Supermicro line here | Buy online with a B2B account: 12% off list observed (free application; Supermicro publishes 'start with 3%'). Credit card or Supermicro Business Pay Net-30 (needs 2 years of operating history); no purchase orders | Gold Series RTX PRO 6000, compute and storage systems ship within 24 hours (November 2026); configure-to-order 3-5 weeks non-HGX; HGX B300 on NVIDIA allocation, 30+ weeks (the B300 Gold SKUs show no price on 2026-11-03 and the configurable chassis carries a "longer than usual lead times" banner); HGX H200 configure-to-order, contact Supermicro before purchasing; on-site service required at checkout on HGX chassis | Ships to the continental US only. GPU orders pass NVIDIA export review before ship; 8U HGX boxes (300 lb) need freight the store does not quote. Gold prices reset on the 1st of each month |
| Supermicro through distribution (TD SYNNEX US and Canada) | Distributor | Private: distributor quotes are not published; introduction | The same Supermicro SKUs on a distributor quote: purchase-order terms, reseller or end-user paper, Canadian delivery | UpStation introduces the buyer to the Supermicro desk; a quote is issued only with the end user's name, address and website. NCNR; wire or net terms. Quotes carry a memory and storage volatility clause (price adjusted if component costs rise before production, buyer may confirm or cancel) and expire in two to three weeks | Drop-shipped from Supermicro San Jose; same stock position as the eStore. About a week for the Supermicro desk to issue a quote (October 2026) | Canada: quoted in USD, freight covered on prepaid-wire orders, a 25% reciprocal tariff was flagged as possible and unresolved (September 2026). Distributor pricing on Gold SKUs landed close to eStore B2B net plus a 3-year on-site NBD service |
| Aivres | System builder (Tier 2), OEM direct | Listed as the Supermicro HGX B300 equivalent (same public price basis; interchangeable); Aivres confirms its own price on the quote | HGX B300 8-GPU 10U air-cooled: KR9288X3 (Intel Xeon 6) or KR9288E3 (AMD EPYC); also RTX PRO 6000 platforms (KR6268) | New-account paperwork (account form, end-user statement, W-9, resale certificate), then a customer-specific quote. 25% deposit within days of an NCNR PO, 75% before ship; DAP; 3-year warranty; on-site NBD available as an option | B300 on Aivres allocation (October 2026 schedule): new Intel orders from mid-to-late January 2027 in limited quantities, larger lots January-March; AMD builds from February 2027; GPU supply confirmed per project; nothing is reserved until a deposit is paid. Ready-stock RTX PRO 6000 8-GPU systems about 8 weeks (October 2026). Quotes valid one week | Aivres quotes introduced buyers directly; sales director ex-Supermicro; prioritises confidential-computing (TDX) builds. Delivery outside North America goes through Aivres's regional desks |
| Dell, HPE, Lenovo | OEM (Tier 1) via partner programs | Partner pricing: on introduction (deal registration) | PowerEdge XE, HPE ProLiant / Cray and ThinkSystem GPU servers | UpStation is a Dell Technologies Authorized Partner (Canada), an HPE Business Partner (Partner Ready Vantage) and has a Lenovo Canada channel account manager; priced builds on request through deal registration and distribution | By OEM and configuration; not stocked | Not priced on this site yet |
| Octoserver | System builder (Tier 2) | Public: listed exactly (online configurator) | 4U dual-socket AMD chassis with 8 usable double-width slots; the price basis for the Tier 2 lines here | Online configurator with per-part prices; 3-year parts warranty, rails and cables included; order direct | Chassis in stock (93 and 280 on 2026-11-03); builds 2-4 weeks after order confirmation (Octoserver), 3-5 weeks planned here; custom passive RTX 5090 cards 30 days | Custom passive dual-slot RTX 5090 (and 4090) cards that fit a server chassis; RTX PRO 6000 passive card $14,800 in stock (Supermicro's accessory card is out of stock) |
| OneChassis | Chassis maker (build your own) | Public: OneChassis's standard customer price, quoted to every buyer; shown exactly | 7U GPU chassis that takes retail 3.5-4-slot RTX 5090 cards (front and rear trays, 5× 2000 W) with an ASRock Rack board, and the 4U OCG7466M6 for RTX PRO 6000 Server Edition; the basis for the OneChassis barebone lines; also standard 1U-4U rackmount cases for E-ATX, EEB and ATX boards, shipped empty with no risers or board (product range sheet, 2026-10-10) | Order direct from Shenzhen (WhatsApp/email, proforma invoice, bank transfer); ships chassis + risers + PSUs + motherboard, PCIe links tested before shipping; no CPUs, memory or drives | Chassis in stock (both 7U riser versions sold in late October 2026; rails restocked end of October); board stock last confirmed September; plus freight from Shenzhen, about 50 days by sea or faster by air | The only listed host for retail RTX 5090s (every other 5090 line needs a custom passive card). Shipping and duties are extra; confirm Incoterms on the proforma. OneChassis pays UpStation a referral on introduced orders |
| Thinkmate | System builder | Private: quote on introduction | Intel and AMD GPU servers built to order | Quote on request; the quote states motherboard and BIOS version so TDX / confidential computing can be validated | By build | |
| ASRock Rack | Chassis OEM | Public barebone prices at integrators; no system quotes | MGX and conventional GPU barebones (4UXGM-GNR2: 8x RTX PRO 6000 SE, 16x E1.S, 3+1 3200 W) | No direct sales to small buyers; barebones through Hypertec (Canada), TD SYNNEX, Ma Labs, ASI, or integrators such as ASA Computers and ICC | About 8 weeks for an HGX barebone (September 2026) | 4UXGM-GNR2 barebone configurations start around $17k at integrators (no GPUs) |
| QCT (Quanta) | ODM | Private: quote on introduction (50-unit minimum) | Hyperscale-style GPU servers built to customer specification | Minimum order 50 servers, NCNR purchase order with payment in full, through TD SYNNEX, Arrow or Cambridge | About 38+ weeks; secure NVIDIA allocation first | Not a fit below 50 units |
| Surplus and secondary market (brokers) | Broker / B2B surplus marketplace | Asks quoted per lot on request; not published here | Ready-to-ship HGX lots when they appear; serials, warehouse proof and warranty transfer to be confirmed per lot | Lot BOM and ask on request; the seller asks for the end user and deployment site before releasing stock; wire. Confirm new versus used, warranty transfer and remaining OEM support; the NVIDIA end-user statement still applies | In stock: days to ship plus export review | Asks sit above OEM build-to-order and change daily; brokers prefer lots of 8-16 nodes and do not always engage for 2-3 |
| Colocation (Montana, Texas, Oregon, Pennsylvania, Canada) | Colocation | Listed: per-location rates on the power page; the site and quote come with the introduction | Powered racks for the servers above | UpStation introduces the site; service order, MSA and colo schedule with the operator | Montana has about 480 kW available for a January 2027 start (October 2026) and confirmed room for small deployments; Ontario has air or liquid capacity; the Pacific Northwest is tight | Modelled per location on the power page with the tax folded in: Montana $260, Texas $225, Oregon $175 (Tier II), Pennsylvania $250 with the certified-data-center exemption, Canada $225 (3-year, Ontario air or liquid) or $175 (5-year). Internet rates seen: 1G $1,000, 10G $2,500 per month. |
NVIDIA Inception members (guidance 2026-07-27 to 2026-10-25; confirm current amounts): $60,000 off an HGX B300 node, $50,000 off HGX B200, $44,000 off H200 x8, $1,400 per RTX PRO 6000 card, with lifetime caps and internal-use terms: table.
All 16 current market notes (and 5 superseded ones), with the compliance, Inception, confidential-computing and colocation items: Suppliers and how buying works.