● Solar-powered Ubiquitous Neural-mesh

The rooftop was always
a bigger data center.

Project SUN turns stranded residential solar capacity — capped by regulation, not physics — into a decentralized, carbon-negative AI compute mesh. No transmission queue. No new substation. Just the 65–80% of your roof that utilities won't let you use.

The Problem

AI's growth curve is about to hit a grid wall

Centralized data centers are colliding with transmission queues, water limits, and thermal physics — right as inference, not training, becomes the dominant and permanent energy load.

~950 TWh
Global data center electricity demand projected by 2030 — roughly Japan's entire annual consumption
4–7 yrs
Typical transmission interconnection queue for a new hyperscale campus in major U.S. ISOs
80–90%+
Share of a model's lifetime energy spent on inference, not training — and it never stops
40–100 kW
Power draw per AI rack today, versus 7–10 kW for a legacy server rack — past the limit of air cooling
The Stranded Asset

Regulation caps your roof at 20–35% of its physical capacity

California's Net Billing Tariff (NEM 3.0) sizes solar interconnects to your home's historical consumption, not your roof's physical collection area — leaving most of a typical rooftop's solar potential unbuilt for its entire 25–30 year lifespan.

4.8x
Nameplate capacity unlocked by building to physical roof limits (30.5 kW) instead of the consumption-capped allowance (6.4 kW)
46.7x
Value multiplier when surplus solar is converted to behind-the-meter AI compute instead of exported at $0.05/kWh
MetricRegulatory-capped (NEM 3.0)Max. physical build-out
DC nameplate capacity6.36 kW-DC30.53 kW-DC
Annual generation10,800 kWh51,840 kWh
Mid-day export value$0.04 – $0.08/kWh$2.15/kWh (as compute)
Simple payback period12.4 years2.78 years
Internal rate of return32.8%
The Hardware

Edge Pods: confidential inference behind the meter

A wall-mounted, 1.0–2.5 kW appliance built on four pillars — engineered so a property owner hosts real revenue-generating compute without ever being able to see what runs on it.

01

Confidential compute, zero-trust by default

TPM 2.0 / HSM hardware root of trust, plus AMD SEV-SNP, Intel TDX, or ARM CCA trusted execution environments. Model weights, prompts, and outputs stay encrypted end-to-end — the host cannot inspect or tamper with a workload.

02

Physical tamper response

Chassis micro-switches, conductive mesh wiring, and MEMS accelerometers trigger instant cryptographic key zeroization on intrusion — drilling, cutting, or unauthorized access wipes the enclave before it can be read.

03

Total network isolation

No physical or logical link to the resident's home LAN. Traffic moves exclusively over an encrypted WireGuard mesh with 5G/LTE failover — a dedicated, air-gapped path in and out.

04

Direct hydronic heat reuse

Direct-to-chip liquid cooling recaptures 85–92% of electrical input as usable thermal heat for domestic hot water — targeting a net-negative PUE and zero evaporative water consumption.

SpecValue
Power budget1.0 – 2.5 kW continuous, 80 Plus Titanium PSU
Acoustic output< 30 dBA at 1 meter, peak load
Compute siliconDual/quad FP8 · INT4 accelerator blades, HBM3e / LPDDR5X
ComplianceUL 62368-1 · NFPA 70 · SOC 2 Type II · ISO 27001 · HIPAA · GDPR
The Business

Energy and compute, cleared as one dual-commodity market

kWh consumed are pegged directly to tokens generated. Enterprises buy carbon-neutral inference at a steep discount; hosts get paid in cash and free hot water; the grid gets a shock absorber.

40–60%
Compute discount vs. hyperscale cloud providers (AWS / Azure) for buyers
14–18 mo
Simple CapEx payback period per 2 kW Edge Pod node
15–25%
Net compute margin share paid to the hosting property owner
100%
Grid energy cost reimbursed to hosts, plus free domestic hot water
StakeholderEconomic benefit
Compute buyerDynamic rates ~$0.001–$0.002 / 1k tokens — a 40–60% discount vs. hyperscalers
Host property owner100% grid-energy reimbursement + 15–25% net margin share + free hot water
Node operator14–18 month simple hardware payback period
Protocol & DAO5–10% transaction fee funding insurance, development, and governance
The Settlement Layer

The Balancing Ledger: a blockchain clearinghouse for kWh and tokens

An Ethereum L2 settles multi-party payments trustlessly; off-chain state channels stream per-token, per-kWh micro-payments with sub-cent fees, batched into 15-minute settlement epochs.

Mode 1 — Surplus behind-the-meter solar
Midday solar peak. Token costs fall to a floor of $0.0004–$0.0008 per 1k tokens.
$0.00 – $0.03/kWh
Mode 2 — Battery storage discharge
Stored clean energy extends low-cost compute into the evening.
$0.05 – $0.08/kWh
Mode 3 — Off-peak grid ingestion
Standard off-peak utility draw once solar and battery reserves are depleted.
$0.12 – $0.18/kWh
Mode 4 — Peak grid draw / stress event
4–9 PM peak. Engine surges spot pricing or triggers VPP curtailment for demand-response rewards.
$0.40 – $0.60/kWh
Execution

Three phases from pilot to virtual power plant

Phase 1 · Months 1–12

Alpha pilot & proof of concept

  • 100 Edge Pods across solar homes + a 150-unit flagship multi-family site in Southern California
  • Validate thermal integration, enclave security, and dual-ledger accuracy
Phase 2 · Months 13–24

Commercial multi-family expansion

  • Scale to 2,500 units across CA, TX, and the Southwest
  • Launch self-service developer inference API; sign national property-management partners
Phase 3 · Months 25–36+

Utility VPP integration & national scale

  • 25,000+ nodes delivering 50+ MW of distributed, carbon-negative compute
  • Certified Virtual Power Plant status with CAISO, ERCOT, and PJM
Source Research

The six founding documents

Everything on this page is synthesized from Project SUN's master documentation set.

Project SUN: Master Index & Architecture Blueprint
The connective document — links and key metrics across all five artifacts below
Open →
Green AI: Energy Sustainability & the Compute Transition
The macro energy crisis facing centralized AI infrastructure
Open →
Decentralized AI Compute & Energy Mesh: Business Plan
Commercial model, GTM strategy, and unit economics
Open →
Decentralized AI Compute Boxes: Hardware & Security
Edge Pod chassis, silicon, TEEs, and thermal design
Open →
The Mathematical Potential of Unused Residential Solar
NEM 3.0 regulatory analysis and rooftop capacity modeling
Open →
Balancing Ledger: Decentralized Architecture & Settlement
Blockchain L2 design, oracles, and the dynamic pricing engine
Open →