● Sovereign Compute · BYOC · Bring Your Own Compute
The Dispatch Optimization Engine that runs the plant.
Inputs dispatched simultaneously
Physical platforms orchestrated
Carbonstamp per MWh
WattTime certified output
The problem
Behind-the-meter AI power infrastructure — generation, storage, interconnection, and cooling — does not self-coordinate. Each system operates on its own control logic, optimizes locally, and produces its own operational record. The result is sub-optimal dispatch, carbon leakage between reporting periods, and a site P&L that cannot be closed without manual reconciliation of four separate data sources.
The carbon accounting problem is structural, not administrative. Without a shared dispatch record, per-MWh carbon attestation is an estimate — not a measurement. And at the permit boundary, uncoordinated engine dispatch can cross CO thresholds that a coordinated dispatch would have avoided entirely.
Without a single dispatch layer aware of all 19 engine permit slots, sites risk crossing minor-source CO thresholds that coordinated dispatch would have avoided.
Engine dispatch decisions made without real-time fuel price and MOER data leave cost-per-MWh on the table every dispatch cycle. At scale, that gap is material.
Manual carbon accounting produces portfolio-average estimates. WattTime NZe attestation requires a per-MWh dispatch record that only a shared orchestration layer can generate.
Customers running eGEN, eSTOR, eGRID, and eCOOL independently see four operational records. No single number closes the site P&L without reconciliation.
The SovAI answer
Scotty — the Dispatch Optimization Engine — is the AI orchestration layer across eGEN, eSTOR, eGRID, and eCOOL. It ingests five continuous data streams and produces four simultaneous output commands on every dispatch cycle. There is no siloed optimization: every decision is made in the context of every other.
Scotty ingests MOER-based carbon intensity signals from WattTime, live fuel prices across each site's active fuel pathway, GPU load telemetry from the compute layer, ambient temperature and site-condition data, and real-time CO permit headroom per engine. It processes these inputs on a continuous dispatch cycle — not a fixed interval — and produces simultaneous output commands to all four physical platforms.
Real-time dispatch — inputs → Scotty → outputs
Scotty's dispatch log feeds WattTime's MOER-based NZe certification pipeline directly. Every MWh the site consumes is tagged with the actual MOER at the moment of consumption and attested by WattTime as an NZe-certified avoided-emissions record. No manual carbon accounting step. No spreadsheet reconciliation. No estimated portfolio average. The number is the number.
Scotty orchestrates the full platform
Engine count, load factor, and fuel selection dispatched per cycle against MOER, permit headroom, and fuel economics.
Supercapacitor and LFP charge/discharge profile coordinated with eGEN load curve and GPU demand spikes.
Grid import and export decisions made in context of full-site generation cost and MOER signal at the meter.
Supply temperature and flow-rate setpoints adjusted in real time against GPU load telemetry and ambient conditions.
All four systems dispatched in real time by Scotty / DOE — against MOER signals, fuel prices, site permit constraints, and GPU load telemetry.
The stack
NVIDIA stack · Inception Program member
Scotty is built on NVIDIA Cosmos Reason for world-model reasoning, Holoscan for real-time sensor and telemetry processing, cuOpt for GPU-accelerated dispatch optimization, and PhysicsNeMo for physics-informed thermal and generation simulation. SovAI is an NVIDIA Inception Program member. NVIDIA DSX AI Factory architects have validated the compute load interface at the eGRID delivery point.
Digital twin · Integration planned
A site-level digital twin built on PhysicsNeMo and fed by eGEN, eSTOR, eGRID, and eCOOL sensor streams is in development. The twin will allow Scotty to simulate dispatch scenarios ahead of real-world execution, validating permit-safe operating envelopes before committing engine dispatch decisions to the physical fleet.
API surface · OMS / EMS integration
Scotty exposes a documented API surface for customer Operations Management System and Energy Management System integration. Customer platforms can read dispatch state, receive carbonstamp data in real time, and post demand-response constraints that Scotty incorporates into its optimization cycle.
What Scotty delivers
Scotty makes a multi-platform behind-the-meter site a single operational entity. Customers see one dashboard, one cost per MWh, and one carbon number — not four separate systems requiring manual reconciliation at period close.
Scotty’s dispatch log is the direct input to WattTime’s NZe certification pipeline. Every MWh consumed carries a third-party-attested avoided-emissions record — verifiable, tied to a specific dispatch event, and issued without manual carbon accounting.
WattTime · NZe partner
Scotty tracks real-time CO permit headroom across the full eGEN fleet. Engine dispatch decisions account for minor-source CO thresholds continuously — not at reporting time. Sites stay within permit boundaries without manual monitoring or operator intervention.
Scotty’s documented API surface gives customer Operations and Energy Management Systems real-time read access to dispatch state and carbonstamp data, and write access to post demand-response constraints into Scotty’s optimization cycle.
NVIDIA Inception Program member
IP posture
PATENT PENDING
Scotty’s multi-platform dispatch optimization architecture — including its MOER-weighted, permit-constrained, fuel-economic engine dispatch model — is the subject of a provisional patent filing with Joe D. Doyle as sole inventor. Assignment to SovAI, Inc. is in progress.
Sovereign Compute · Scotty / DOE
Bring your load, your location, and your timeline. We will configure the dispatch model and run the carbonstamp methodology.