Constrained AI campus
Existing compute and power envelope. Ambient-air, buried-loop, and cold-storage cooling first. Additional behind-the-meter compute next. Optional AI heat recovery after uptime isolation. Qualified generator recovery last.
Platform
Cooling-first infrastructure on land that is already connected.
Komorebi reduces cooling parasitics, recovers rejected heat, and returns electrical headroom to AI compute. Two starting conditions converge on one thermal platform — without requiring an equivalent new interconnect. Start compute first; add each thermal layer only when its incremental compute value exceeds its cost.
Two entry paths. One thermal platform.
Brownfield PV site
Existing land, interconnection, and 100 MWdc+ PV. Modular compute plus the same thermal infrastructure. PV / GCCM repower and recovered aperture. Optional CSP, TES, and ORC. Then qualified heat recovery. No field expansion.
Cooling first
The thermal layer is the compute unlock.
An ambient air and ground-coupled cooling network and cold storage cut the electrical burden of conventional heat rejection. That parasitic reduction is what creates room for additional IT load on an already-interconnected site.
Original dry coolers are retained and later shifted onto ORC condenser duty, so cooling CAPEX is reused rather than stranded.
Commercial sequence
- 01
Cooling first unlocks the most compute, soonest — at the interconnect you already have.
- 02
AI heat recovery comes after uptime isolation. It is real, and it is the smaller step.
- 03
Qualified generator recovery is last. Each asset is integrated only when it pays.
Phase 0
Existing campus
Months, not years
- Fast-deploy generation and modular halls
- Independent rack cooling
- Fail-safe heat-rejection bypass
- Uptime is the first commercial gate
Start workloads while the platform is built
Phase 1
Cooling first
6–12 months
- Ambient air and ground-coupled cooling loops
- Cold storage; conventional cooling retained
- Independent of PV densification on a campus
- On brownfield PV: GCCM reflective foundations and array densification
Full IT load cooled. PUE falls.
Phase 2
AI heat recovery
After uptime isolation
- Heat pumps plus thermal ladder
- TES and right-sized ORC
- Original dry coolers reused for ORC condenser duty
- Failure isolated from rack cooling; no later stage may degrade Stage 1
Useful heat recovered from AI load
Phase 3
Qualified generator recovery
Asset-by-asset, when it pays
- Qualified reciprocating or simple-cycle heat into the existing TES/ORC stack
- Asset-by-asset — only where recoverable heat and OEM/EPC interfaces justify integration
- Not every generator topology qualifies
- No solar-land expansion
Additional hours from qualified generator heat
Pecos Campus Case
300 MW existing IT / PUE 1.25 campus / 1,000-acre envelope, Pecos, Texas. Envelope planning case: ~130.2 MWdc PV and 50 MWth CSP on exactly 1,000 acres. 300 × 1.25 = 375 MW campus, closed envelope. Phase 1: 51 MW freed ÷ 1.08 = +47 MW IT. Grid never exceeds 375 MW.
Added this phaseCumulative additionalY-axis: MW
100 MWdc Brownfield PV Conversion Case
A separate 100 MWdc legacy-PV geometry example — ~390-acre row field. Not the geometry behind the 300 MW campus result. Upgrade in place to ~95 MWdc of higher-density bifacial monocrystalline modules on reflective foundations, and convert recovered aperture to ~94 MWth of Komorebi CSP.
- Legacy plant
- 100 MWdc
- Repowered PV
- ~95 MWdc
- Recovered CSP
- ~94 MWth
- Field expansion
- None
Revive PV
Aging modules become a recoverable asset.
Brownfield repowering should not start with a dumpster. Recover what can run. Recycle what cannot.
01
Remove
On-site take-down and triage. What can run stays in the loop.
02
Repair
Inspect, recoat, repair, and test in a two-container mobile factory.
03
Redeploy
Put recovered modules back to work. Recycle the rest.