The engine
How Pod 1 gets designed for your site.
The Hexi engine is a world model of how a facility gets from an empty plot to working GPUs, with search on top. It is the reason a Pod 1 design fits its site rather than a catalogue.
What it models
- A schedule for grid energisation, transformers, switchgear, generation, permits and construction, as 300 simulated futures
- An hourly year of operation with unit outages, maintenance and grid outages, checked against a critical-availability target
- Cooling energy against the site's own hourly weather
- Ten years of monthly cash flow: cost per useful GPU-hour and NPV
How we keep it honest
- Every cost and lead time comes from 475 table rows, each sourced or labelled as an engineering estimate
- An audit searched for ways a plan could score better than it would in reality; its 21 loopholes are fixed and kept as tests
- Three rounds of independent review followed, with each finding fixed and tested
- GPUs are ordered against the plan's quoted power date, not hindsight
The Texas showcase
One 100 MW campus at three real sites, with nothing confirmed by the utility. Median of 300 futures per plan; 2024 hourly weather.
| Site | Wait for the grid: first GPUs (months) | Engine's plan: first GPUs (months) | Wait for the grid: $/useful GPU-hour | Engine's plan: $/useful GPU-hour |
|---|---|---|---|---|
| Eagle Mountain (retired plant) | 51 | 24 | $4.62 | $4.20 |
| Morgan Creek (West Texas) | 51 | 24 | $4.63 | $4.16 |
| Johnson County (greenfield) | 52 | 25 | $4.64 | $4.16 |
What this is not
A screening exhibit, not a utility offer or an engineering design. Title, gas supply and connection capacity were not verified, and some cost rows are labelled engineering estimates. At $4 per GPU-hour every plan loses money: speed only pays when there is a margin.
Sources: US EIA Form 860M; © OpenStreetMap contributors (ODbL); US EIA pipelines; US EPA eGRID2023; NOAA Integrated Surface Database 2024. No ERCOT large-load material is reproduced.