|  | |  |  | QHT, re: Inference Data Center Spare-  Nvda BOM and Energy (TCO) vs QCOM solution- 
 This is what is so exciting about entering into this space. Nvidia racks are so wildly expensive that when QC come in with mostly competitive inference racks, but at a significant discount its a win-win all around.
 
 DCs can get inference capabilities, QC gets revenue that was never forecasted -- so QC can undercut.
 
 >>>>>>>>>>>>>>>
 Per Copilot- thought this was an interesting comparison.
 
 Can  you verify the reasonableness ???
 
 >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
 
 — Total Cost of Ownership (TCO) &  Energy Cost Comparison: Qualcomm vs. Nvidia vs. AMD
 
 Qualcomm’s AI250  racks deliver up to 60% lower energy cost and 45–65% lower TCO per rack compared  to Nvidia Rubin and AMD MI300X, driven by LPDDR5X memory, near-memory compute,  and 160 kW rack power vs. 500–600+ kW for competitors.
 
 
 
 Total Cost of Ownership (TCO) — 3-Year Estimate  per Rack
 
 
        Assumes 24/7  operation, $0.04/kWh industrial rate, 90% utilization, and 3-year amortization.     | Vendor 
 
 | Platform 
 
 | CapEx (USD) 
 
 | Energy Cost (3 yrs) 
 
 | TCO (3 yrs) 
 
 | vs. QCOM AI250 
 
 |      | Qualcomm 
 
 | AI250 
 
 | $332,000 
 
 | ~$168,000 (160kW @ $0.04/kWh) 
 
 | $500,000 
 
 | — (baseline) 
 
 |      | Qualcomm 
 
 | AI200 
 
 | $211,000 
 
 | ~$168,000 
 
 | $379,000 
 
 | 24% lower TCO 
 
 |      | Nvidia 
 
 | Rubin NVL576 
 
 | $741,000 
 
 | ~$630,000 (600kW) 
 
 | $1.37M 
 
 | 63% higher 
 
 |      | Nvidia 
 
 | GH200 Grace Hopper 
 
 | $700,000 
 
 | ~$420,000 (400kW) 
 
 | $1.12M 
 
 | 55% higher 
 
 |      | AMD 
 
 | MI300X 
 
 | $303,000 
 
 | ~$525,000 (500kW) 
 
 | $828,000 
 
 | 39% higher 
 
 |      | AMD 
 
 | MI300A 
 
 | $330,000 
 
 | ~$570,000 (540kW) 
 
 | $900,000 
 
 | 45% higher 
 
 |  
 
 
 Energy Cost Breakdown (3-Year, per Rack)
 
 
             | Platform 
 
 | Power Draw 
 
 | Annual Energy (kWh) 
 
 | 3-Year Energy Cost 
 
 |      | QCOM AI250 
 
 | 160 kW 
 
 | ~1.4M kWh 
 
 | ~$168,000 
 
 |      | Nvidia Rubin 
 
 | 600 kW 
 
 | ~5.3M kWh 
 
 | ~$630,000 
 
 |      | AMD MI300X 
 
 | 500 kW 
 
 | ~4.4M kWh 
 
 | ~$525,000 
 
 |  Key Efficiency Drivers for Qualcomm
 
 
    LPDDR5X vs. HBM3/3e: Lower voltage, lower cooling loadNear-memory compute: Reduces data movement energyDirect Liquid Cooling: Cuts HVAC overhead by up to 40%ARM-based SoC: Lower idle and orchestration power | 
 |