AI Server System BOM & Rack Cost
Newest print Mar 2026Reviewed semi-annuallyCSV — newest observations; full series with a Pro keyAPI
Networking: $80K (GB200 NVL72 (2024)) — SemiAnalysis rack cost analysis, Dec 2024 (estimate). Basis: BOM cost.
AI Server System BOM & Rack Cost — latest readings
8 observations across 4 series, 2023–2024.
| Series | Latest ($K) | Source |
|---|---|---|
| GPU (COGS)GB200 NVL72 (2024) | 486 | Our model estimateSilicon Analysts modelMar 2026 |
| NetworkingGB200 NVL72 (2024) | 80 | SemiAnalysis rack cost analysisDec 2024 |
| Power & CoolingGB200 NVL72 (2024) | 120 | NVIDIA NVL72 spec / Silicon AnalystsMar 2026 |
| Memory, Storage & ChassisGB200 NVL72 (2024) | 65 | Our model estimateSilicon Analysts modelMar 2026 |
Full system-level bill of materials for AI training servers — from DGX H100 (8-GPU) through GB200 NVL72 (72-GPU rack). Tracks how rack economics evolved as HBM and networking costs exploded.
Power & Cooling: $120K (GB200 NVL72 (2024)), NVIDIA NVL72 spec / Silicon Analysts, Mar 2026
Source: Silicon Analysts model, Epoch.ai / NVIDIA DGX spec, SemiAnalysis, Data center analyst estimates, SemiAnalysis rack cost analysis, NVIDIA NVL72 spec / Silicon Analysts · Basis: BOM cost
Prints ledger
| Note and copy | |||||||
|---|---|---|---|---|---|---|---|
| GB200 NVL72 (2024) | GPU (COGS)Our model estimate | 486 | |||||
Source: Silicon Analysts / chipSpecs.ts, Mar 2026 · Basis: BOM cost · Confirmed 72x B200 x $6,750 COGS = $486.0K. Re-based 2026-08-16 with chipSpecs.ts; the prior $6,400 anchor predated the HBM3e re-base. | |||||||
| GB200 NVL72 (2024) | Power & Cooling | 120 | |||||
Source: NVIDIA NVL72 spec / Silicon Analysts, Mar 2026 · Basis: BOM cost · Estimate 120kW rack; liquid cooling mandatory; cold plates + CDU | |||||||
| GB200 NVL72 (2024) | Memory, Storage & ChassisOur model estimate | 65 | |||||
Source: Silicon Analysts estimate, Mar 2026 · Basis: BOM cost · Estimate 36× Grace CPU (LPDDR5X), NVMe, chassis, power distribution, optics | |||||||
| GB200 NVL72 (2024) | Networking | 80 | |||||
Source: SemiAnalysis rack cost analysis, Dec 2024 · Basis: BOM cost · Estimate NVLink5 switch, InfiniBand NDR400, PCIe retimers | |||||||
| DGX H100 (2023) | Memory, Storage & ChassisOur model estimate | 6.4 | |||||
Source: Silicon Analysts estimate, Mar 2026 · Basis: BOM cost · Estimate System DDR5, NVMe SSDs, chassis, optics | |||||||
| DGX H100 (2023) | GPU (COGS) | 26.6 | |||||
Source: Epoch.ai / NVIDIA DGX spec, Dec 2023 · Basis: BOM cost · Confirmed 8× H100 SXM5 × $3,320 COGS = $26.6K | |||||||
| DGX H100 (2023) | Networking | 8 | |||||
Source: SemiAnalysis, Dec 2023 · Basis: BOM cost · Estimate NVLink switch + InfiniBand HDR NICs | |||||||
| DGX H100 (2023) | Power & Cooling | 4 | |||||
Source: Data center analyst estimates, Dec 2023 · Basis: BOM cost · Estimate 700W TDP per H100 × 8 = 5.6kW; power delivery + cooling | |||||||
Provenance
Synthesized from Epoch.ai hardware cost database, SemiAnalysis system architecture deep dives, and analyst estimates. All costs are approximate manufacture/procurement costs, not list prices.
Basis: BOM cost
System-level manufacture/procurement cost stack. EXCLUDES vendor gross margin, NRE, software and support: a comparable GB200 NVL72 system sells for roughly 4x this figure (~$3.1M street vs ~$726K BOM). Never benchmark this column against a rack sell price.
Sources
- Epoch.ai / NVIDIA DGX spec (Dec 2023)
- SemiAnalysis (Dec 2023)
- Data center analyst estimates (Dec 2023)
- SemiAnalysis rack cost analysis (Dec 2024)
- NVIDIA NVL72 spec / Silicon Analysts (Mar 2026)
Our derivations
- Silicon Analysts estimate (Mar 2026)
- Silicon Analysts / chipSpecs.ts (Mar 2026)
Suggested citation
Silicon Analysts, "AI Server System BOM & Rack Cost", as of Mar 2026, https://siliconanalysts.com/market-data/ai-server-system-bom, retrieved October 19, 2026.
Related Tools
Related Analysis
Market intelligence on this data
All market intelligence →Samsung to Use TSMC 3nm for Base Dies of Nvidia Custom HBM
N/A (no unit pricing disclosed); TSMC 2026 CapEx maintained at $60–64B, new Tainan 3nm fab production 1H 2028, Arizona Fab 2 mass production Q3 2027
Micron Opens Second Taiwan HBM Packaging Fab; All 2027 Output Already Sold Out
Capacity: new AATT2 HBM assembly/test line activated, with 100% of planned 2027 output contracted; Micron HBM capacity cited at ~60,000 wafers/month, roughly one-third of SK Hynix or Samsung
GlobalFoundries Lands $2B TSMC Deal to Add First U.S. CoWoS Interposer Source
Capacity: ~$2B multi-year interposer supply agreement (five-year initial term), volume production from H1 2028; CoWoS capacity reported sold out through 2026; ~$400M/year, ~6% of GF 2025 revenue ($6.79B)
Samsung HBM4 2027 Pricing Set to More Than Double
Samsung HBM blended ASP: +>100% YoY in 2027 (KB Securities); reported HBM4 asking price mid-to-high $4 per gigabit for 2027 vs ~$1.50 per gigabit for HBM3E (asking prices, not confirmed contracts); TrendForce: industry HBM blended ASP +121% YoY in 2027; HBM4 to be ~80% of Samsung HBM revenue in 2027 vs ~40% in 2026