Great Energy is a pure trading company. Every morning our AI ranks some 24,000 node-hours on the Texas grid by where day-ahead and real-time prices will part, and trades the ones it is most sure about.
Wind and sun decide what the grid produces, and the grid decides the price. Weather is where the uncertainty comes from, so it is where the trade begins.
Every morning we read several independent forecast families side by side, from neural weather ensembles to the grid operator's own, and look closest at where they disagree.
From the turbine, power travels the transmission lines to the nearest substation: the node where every megawatt-hour is metered and priced.
One of roughly 1,000 settlement points on the Texas grid, each priced twice: the day before, and on the day itself. About 24,000 node-hours, every single day.
The pipeline runs unattended every morning. Data is frozen, every node-hour is scored, and positions are recorded before ERCOT's 10:00 deadline.
Weather ensembles, load forecasts, gas prices, historical LMPs, transmission and outages, and 170 more features. Each exactly as it was published before the deadline.
One ranking a day. Positions on the node-hours the model is most sure about, settled against real-time prices the next morning.
Once the day before it is used, in ERCOT's day-ahead market, and again on the day itself, in real time. The two prices are almost never the same: usually a few dollars apart, on a hot afternoon or a cold night, hundreds. Our models predict which way the gap falls, node by node and hour by hour.
Illustrative · one node, one day · $/MWh
| Hour | Day-ahead | Real-time | Gap |
|---|
Sell day-ahead, buy back in real time. Pays when the day-ahead price is higher.
Buy day-ahead, sell back in real time. Pays when the real-time price is higher.
No power plants, no batteries, no customers. A prediction, and the positions that follow from it.
Weather is where the grid's uncertainty comes from, so it is where the trade begins. Every morning we read several independent forecast families side by side, and look closest at where they disagree.
Neural network weather models from the world's leading forecasting labs, at roughly 10 km resolution, so every node gets its own weather cell.
Global numerical weather prediction sampled across the grid: the classic baseline every new model has to beat.
ERCOT's own load, wind and solar forecasts, including its competing internal wind models, each stored with the minute it was published.
When the models part ways, the official forecast is the one most likely to miss. That disagreement goes into the ranking as a signal of its own.
We build our own models end to end. 175 signals go in for every node and hour on the Texas grid; a ranked list of positions comes out, sized by rule, and every settled day is measured against a timestamped record. The weather comes from neural networks. The ranking does not: it is a gradient-boosted ensemble, the model class that still beats deep learning on structured data, and the one that won on ours.
Several weather forecast families, ERCOT's own load, wind and solar forecasts, outages, congestion shadow prices and the full nodal price history: 175 signals, each stored exactly as it stood before the deadline. No model ever sees what was not known at the time.
An ensemble of gradient-boosted models scores around 24,000 node-hours every morning and returns a score for the spread at each one. Where the weather sources disagree with ERCOT's own forecast is part of the signal.
Every node-hour is ranked by that score and the book is built from the top of the list. Size follows written rules set on the worst days on record. Automatic brakes and a weather stand-down sit above the models.
Positions are recorded before the 10:00 deadline and settled against real-time prices the next morning. Every settled day joins a timestamped record and is measured against the research window the models were built on. Models earn their place by measurement, not by date.
Before training a single model we searched the entire Texas grid for simple rules: a node that always pays, an hour that always pays, a weather pattern that always pays. Against what luck alone produces, none held up. The machine-learned ranking did.
Combinations tested against a null built to fool the same procedure.
Data points in the model, each exactly as it was published before the deadline.
Node-hours scored and ranked every single morning.
Positions placed by hand. The system runs from data to settlement.
The pipeline runs on its own every morning. People set the rules and can stop it at any time. Nobody can add a position by hand.
One decision a day, made before the deadline, settled against real time the next morning.
Every node-hour scored by probability; the book is built from the top of the list.
Every claim is tested against what luck alone would produce. Results that can't beat it don't ship.
Model-free rules sit around the models and act on their own. Size is set by the worst days on record.
Since August 2026 the full pipeline has run unattended every morning. Data is frozen, models are scored, and positions are recorded before ERCOT's deadline, then settled against real-time prices the next day. After estimated fees, it is in profit.
We start where the data is deepest and the gap between day-ahead and real time moves the most: the Texas grid. New markets open one at a time, each once the last runs at full size.
An energy-only market with scarcity pricing and frequent, sharp gaps between day-ahead and real time.
The largest wholesale electricity market in North America.
A wind-heavy footprint from the Canadian border to the Gulf.
Among the highest wind shares of any US grid.
Deep solar penetration and steep evening ramps.
Zonal congestion between upstate supply and city load.
Winter gas constraints and sharp price spikes.
Different market rules, one research method. Each market is earned on its own data before capital follows.
Bo and Anders co-founded and ran their previous venture together for ten years. Great Energy pairs the people who built the system with the people who connect it to the market.
Designed and built Great Energy's entire system: the data infrastructure, the models and the automated pipeline. Two decades of building data-driven, high-frequency businesses, including bidding infrastructure that managed over $400M in annual ad spend. Before the US focus he built a complete, fully automated trading system for EPEX SPOT, the European power exchange, from predictive models to execution, audited and certified by the exchange. Great Energy traded it live, and profitably, through 2025 and into 2026. Certified energy trader.
Runs the market side: everything between a working model and a licensed counterparty at ERCOT, from registration, credit and banking to controlling and reporting. A background in computer science and business administration, and more than two decades as an entrepreneur, with board seats at Twill.net (Maersk Go) and Clever, Denmark's largest EV charging network.
We're glad to hear from market counterparties, engineers and researchers, and professional investors.