[ CALCULATOR · BETA · JUNE 2026 ]What would your
What would your
compute cost?
On our side of the meter.
Pick a chip, set your workload, see the real economics — including what grid-tied power would cost you vs. running behind the meter on SmartTec.
5 chips·4 regions·live recalc·no signup
[ LIVE · UPDATES AS YOU TYPE ]
[ INPUTS ]
Hopper. Workhorse. 80GB HBM3.
$0.085/kWh · 18 outages/yr
behind-the-meter
8units
1500
75%
10100
24hrs
124
[ COMPUTE OUTPUT ]
GPU-hours / month4,320hr
Inference throughput (70B model)2,625 tok/s
Total tokens served6.80B tok
[ POWER ]
Average draw7 kW
Peak draw9 kW
z1power battery needed14 kW
Battery autonomy at peak6.0 hrs
[ ECONOMICS ]
Revenue / month$10,368
Power cost / month$384
Margin / month$9,984
Margin %96.3%
[ RESILIENCE & SUSTAINABILITY ]
Outage hours protected / yr21.6 hrs
CO₂ avoided / yr7 tons
Tokens per watt934,615
Carbon avoided (lifetime)22 tons (3-yr)
[ THE SMARTTEC ADVANTAGE ]
What you'd save going behind the meter.
8 × H100 running 24h/day at 75% util, in US average.
SmartTec$384
behind-the-meter
Grid-tied$780
retail electricity
Annual savings$4.7K
by switching to SmartTec
[ ASSUMPTIONS · 2026 RETAIL MARKET RATES · CONFIRM WITH SMARTEC FOR DESIGN-PARTNER PRICING ]
Lock launch pricing →[ ASSUMPTIONS & DISCLOSURES ]
[ RATES ] GPU-hours reflect June 2026 retail market rates (CoreWeave, Lambda, AWS, GCP range). Cerebras CS-3 inference billed per 1K tokens at Cerebras Cloud list rate. SmartTec design-partner pricing typically 20-40% below list — contact us.
[ POWER ] TDP values are spec-sheet, not measured. Server overhead includes cooling + networking at industry typical 30-35%. SmartTec power cost assumes 55% off-peak / 45% on-peak charging arbitrage + ~15% demand-charge savings vs grid baseline.
[ RESILIENCE ] Outage hours/yr per region are estimates from public ISO/RTO reliability reports. "Protected" assumes SmartTec sub-10ms failover prevents any customer-visible downtime.
[ ENVIRONMENTAL ] CO₂ figures use regional grid mix emissions factors. SmartTec charging from grid still has some carbon; we model ~30% reduction via arbitrage away from peak gas peakers. [ RESULTS NOT A QUOTE ]
[ NEXT STEP ]
Economics confirmed. Now: will your model fit?
Use the inference fit-check to pick a model and GPU. You get VRAM fit, throughput, and a ready-to-run vLLM deploy command for SmartTec compute.