Written by: Olivier Lam, Physical AI Team, Jua.ai AG | Last updated: June 25, 2026
Key Takeaways for Energy Trading Teams
- Energy traders lose the critical 7–9 a.m. window stitching forecasts from raw GRIB files and scattered dashboards before the market moves.
- Traditional NWP models update only 2–4 times per day, which leaves traders exposed to stale data and silent mid-cycle revisions that others may already have traded on.
- Athena turns natural-language queries into written briefings, benchmarks, and alerts in about 90 seconds by orchestrating a 25-model ensemble that includes EPT-2 and leading NWP and AI sources.
- EPT-2 RR delivers up to 24 daily updates at a fraction of traditional NWP compute cost, which enables intraday refresh rates that were previously uneconomical.
- Book a demo with Jua to compress your morning routine into a single, auto-refreshed workspace.
The Structural Problem Behind Manual Forecast Stitching
In Europe's weather-driven energy markets, traders are turning to AI and machine-learning tools designed not to predict temperatures and precipitation, but to forecast the forecast, specifically to anticipate revisions in the ECMWF two-week outlook before those revisions reprice heating demand, renewable output, and system tightness. The underlying problem is structural. The two supercomputers that run global numerical weather prediction (NWP) produce two to four global forecasts per 24-hour period. Between runs, traders look at stale numbers.
Six pain points define the manual routine:
- Manual morning prep. Downloading GRIB files, running brittle in-house pipelines, and stitching spreadsheets consumes the 7–9 a.m. window before the market opens.
- Stale forecasts. With only two to four NWP runs per day, positions are built on data that is hours old.
- Silent model revisions. When ECMWF or GFS revises an output mid-cycle, the trader often notices because someone else has already traded on it.
- Unscalable meteorology. Internal meteorology teams produce daily briefings by hand. The work is high-quality and irreplaceable, yet impossible to scale across more desks, regions, or asset classes.
- Raw AI subscriptions without workflow. Quant teams that subscribe to AI weather research outputs receive raw model files and must build the ingestion pipeline, ensemble logic, and benchmarking harness themselves.
- The compute ceiling on forecast frequency. A single traditional NWP simulation consumes approximately 8,400 kWh and costs €1,000–€20,000 to run. That cost profile has capped update frequency for forty years.
Book a demo to see how Jua for Energy compresses the 7–9 a.m. routine into a single workspace.
Agent-Assisted Forecasting as the New Workflow Layer
An agent-assisted forecasting platform turns a natural-language objective, typed by a trader rather than coded by an engineer, into a written briefing, a model benchmark, a backtest, or a divergence alert. The platform handles model ingestion, consensus logic, refresh scheduling, and output formatting. The trader receives a finished deliverable, not a raw data file.
Jua is a foundation model and agent company. EPT (Earth Physics Transformer) is a general physics foundation model, a spatiotemporal transformer that learns the governing conservation laws of any continuous physical system directly from observational data. It is not limited to weather. Athena is an AI agent, currently instrumented with the Jua for Energy tool surface: forecast queries, model benchmarks, backtests, and widget generation. Jua for Energy is the first applied product built on both.
The positioning is precise: Jua does not replace ECMWF, it displaces the plumbing around it. The ECMWF two-week outlook remains the definitive reference point for traders repricing risk around heating demand, renewable output, and system tightness. What changes is everything between that raw signal and the trading decision, and that is where Athena operates.
How Athena Produces a Natural-Language Briefing in About 90 Seconds
Picture a power trader at a European utility at 6:45 a.m. She types into Athena: “What is the 100 m wind forecast spread across models for northern Germany tonight, and how has it shifted since yesterday's run?” Athena plans the query, calls the forecast and benchmarking tools, evaluates intermediate outputs, and returns a written briefing in approximately 90 seconds.
The briefing covers model consensus, model delta since the previous run, convergence tracking, and price implications. It draws on a 25-model benchmarking engine that includes 10 proprietary AI models from the EPT family alongside 15 third-party NWP and AI models: ECMWF HRES, ECMWF ENS, ECMWF AIFS, NOAA GFS, GFS GraphCast, Microsoft Aurora, DWD ICON Global, ICON-EU, and others. EPT-2 outperforms leading AI weather models and traditional numerical baselines across all forecast horizons on energy-relevant variables, documented in the peer-reviewed technical report arXiv:2507.09703. EPT-2e, the ensemble variant, beats the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time, documented in arXiv:2410.15076.
EPT-2 RR, Jua's rapid-refresh model, updates up to 24 times per day, compared with the two to four runs per day that define traditional NWP. As Jua's team frames it: “We realized early that this is about human nature more than mother nature. Traders need a system that understands how the physical world moves markets.” That cadence lets you act before the market does.
How Jua for Energy Resolves Each Trading Pain Point
Manual Morning Prep → Auto-Refresh Briefings
Day-Ahead and Intraday briefings on the Jua platform auto-refresh on every new model run. Each briefing covers model consensus across more than 25 models, model delta since the previous run, convergence tracking, market spread, and price implications, already written in plain language. Power forecasts for solar, wind onshore, wind offshore, total wind, total renewables, load, and residual load are live in five countries (Germany, Great Britain, France, the Netherlands, and Belgium). The Actual Generation model refreshes every 15 minutes.
Stale Forecasts → High-Cadence EPT-2 RR Updates
EPT-2 RR, Jua's rapid-refresh model, updates up to 24 times per day. EPT-2 HRRR delivers the same high-cadence refresh at up to 5 km spatial resolution over Europe. A single EPT-2 inference runs on a single GPU at approximately 0.25 kWh and $0.20–$15, which is roughly four orders of magnitude cheaper than a traditional NWP simulation. That cost profile makes up to 24 daily runs economically viable without an HPC cluster.
Silent Model Revisions → Divergence and Correction Alerts
Divergence alerts fire the moment two or more models disagree on a key variable. Correction alerts fire the moment a model revises its own output between runs. Both alert types are filterable by zone and PSR (Production Source Resource) type. The trade window opens with a notification instead of a missed move.
Unscalable Meteorology → Athena as a Digital Analyst
Athena turns a natural-language question into a briefing, a benchmark, a backtest, or a custom widget. A typical query resolves with the same sub-two-minute response time described earlier, and a backtest completes in approximately 5 minutes. Internal meteorologists can focus on deeper forecast research instead of manual briefing production. Trading houses and quant desks describe Athena as “another headcount, for free.”
Raw AI Subscriptions Without Workflow → Unified SDK and API
pip install jua installs the Python SDK. The REST API exposes more than 25 models through a single schema, with Apache Arrow support for large payloads. Hindcast data is available across multiple Jua and third-party models for backtesting. Integration work that typically takes a quant team a quarter to build elsewhere stands up in days.
Compute Ceiling → Single-GPU Inference Economics
EPT-2 was trained on 8 × H100 GPUs over 10 days. Microsoft Aurora required 32 × A100 GPUs over 18 days. At inference, EPT-2 runs approximately 25% faster than Aurora on a single GPU. The economics of HPC infrastructure no longer cap update frequency, so Jua refreshes up to 24 times per day where traditional providers refresh two to four times.
Book a demo to run a live benchmark on your own region and variables against more than 25 models in under 5 minutes.
Athena vs Aviation Briefing Tools and Generic Dashboards
The SERP for “automated weather briefing tools” is dominated by aviation products such as ForeFlight and 1800WXBRIEF. These tools are built for pilots filing flight plans, not traders positioning around renewable generation. They deliver point-in-time weather observations and TAF or METAR formats. They do not produce model consensus briefings, do not benchmark competing NWP outputs, and carry no concept of power-market price implications. Generic NWP dashboards process raw model outputs but lack natural-language delivery, ensemble benchmarking, and intraday alert logic.
| Capability | Athena / Jua for Energy | ForeFlight | 1800WXBRIEF | Generic NWP Dashboards |
|---|---|---|---|---|
| Update frequency | up to 24x/day (EPT-2 RR); 15-min actual generation | On-demand aviation obs, no NWP refresh schedule | On-demand preflight briefing, no NWP refresh schedule | Typically 2–4×/day, matching NWP run cadence |
| Natural-language delivery | Full: Athena returns written briefings, benchmarks, backtests, and widgets from plain-text queries at the speed described earlier | None: structured aviation formats (TAF, METAR, SIGMET) | None: structured preflight briefing format | None: chart and map outputs only |
| Energy-specific power forecasts | Solar, wind on/offshore, load, residual load; 5 countries; 20-day horizon | Not applicable | Not applicable | Not native; requires custom post-processing |
| Benchmark transparency | 25+ models on one platform; head-to-head on any region and variable; results in seconds | None | None | Single-model or limited multi-model; no cross-vendor benchmarking |
Aviation briefing tools are purpose-built for regulatory compliance in flight operations. Their output format, such as TAF, METAR, and SIGMET, is meaningless to a power trader positioning around a wind ramp in northern Germany. Generic NWP dashboards process the same ECMWF and GFS runs the trader already has access to, without adding ensemble benchmarking, divergence logic, or a natural-language analyst layer. Neither category was designed for the intraday cadence, multi-model consensus, and price-implication framing that power markets require.
FAQ: Performance, Trust, and Integration
How accurate is EPT-2 compared to ECMWF HRES?
EPT-2 outperforms ECMWF HRES on every lead time across the full 0–240 hour range on the four variables that drive energy P&L: 10 m wind, 100 m wind, 2 m temperature, and surface solar radiation. EPT-2e, the ensemble variant, beats the 50-member ECMWF ENS mean on both RMSE (root mean square error) and CRPS (continuous ranked probability score) at virtually every lead time. Both results are documented in peer-reviewed technical reports on arXiv (2507.09703 for EPT-2; 2410.15076 for EPT-1.5) and validated against more than 10,000 real ground stations on open-source StationBench, with no post-processing or station fine-tuning.
Does Jua for Energy replace our ECMWF subscription?
No. Jua for Energy runs alongside the ECMWF feed, not instead of it. ECMWF AIFS, ECMWF's own AI model, runs natively on the Jua platform alongside EPT-2, EPT-2e, ECMWF HRES, ECMWF ENS, and more than 20 other models. Jua for Energy handles the integration work around the ECMWF feed: the in-house GRIB pipeline, the manual benchmarking, the morning-briefing analyst, and the dashboard stitching. The 7–9 a.m. routine compresses into a single workspace, refreshed up to 24 times a day, where every model is on the same screen with one schema and one API.
Can AI weather models be trusted, or do they hallucinate?
Large language models hallucinate because they are unconstrained on the symbolic surface, so token sequences that look plausible can be physically nonsensical. EPT is constrained at the representation. It is a general physics foundation model trained on observational data, and its outputs respect the conservation laws of mass, momentum, and energy that govern the real atmosphere. The architecture cannot produce outputs that violate those laws in the way a generic transformer applied naively to physics would. Validation is external and concrete. EPT-2 is benchmarked against more than 10,000 real ground stations on open-source StationBench, with results published in peer-reviewed technical reports on arXiv.
How quickly can we integrate Jua for Energy into our existing pipelines?
pip install jua installs the Python SDK from PyPI. The REST API exposes more than 25 models through a single schema with Apache Arrow support for large payloads, documented at query.jua.ai/docs. Hindcast data is available across multiple Jua and third-party models for backtesting. ENTSO-E grid data integrates directly for European power-market data. Quant teams that typically spend a quarter building equivalent ingestion infrastructure stand up the integration in days. A live benchmark on the prospect's own region and variable returns results in seconds, and a full backtest via Athena runs in approximately 5 minutes.
What is on the roadmap beyond atmospheric forecasting?
The atmosphere is the first physical system EPT has been fine-tuned for, and energy trading is the first market Athena has been instrumented for. Neither represents the endpoint. Jua is a foundation model and agent company, and the architecture learns physics while the domain remains a variable. The same EPT model that learns atmospheric dynamics already predicts plasma behaviour inside a tokamak. The roadmap extends to other physical-economy domains such as plasma fusion, aerospace, materials, and fluids, each shipped as a new vertical product on the same horizontal platform. Customers buying Jua for Energy today are buying the first surface of a foundation-model and general-agent platform that will expand outward from there.
Conclusion: Keep the ECMWF Signal, Replace the Plumbing
The 7–9 a.m. manual prep routine reflects infrastructure limits, not trader preference. It is a structural consequence of building an energy trading operation on top of systems designed for two to four forecast updates per day, with no agent layer to translate physics into market language. A 1 GW wind portfolio that gains four percentage points of forecast accuracy saves about €1.5 M per year, and that figure scales linearly across multi-GW portfolios.
Jua for Energy turns the manual routine into a single workspace with auto-refreshed briefings, 25-model benchmarking, divergence and correction alerts, and Athena answering natural-language questions at the speed demonstrated earlier. EPT-2's documented performance advantage over ECMWF HRES gives traders a measurable edge. EPT-2 RR's 24-times-daily refresh rate removes the stale-data problem. The ECMWF subscription stays. The plumbing around it does not.
Jua is a foundation model and agent company, and Jua for Energy is the first applied product. The relationship mirrors Anthropic and Claude Code, a horizontal AI platform with a flagship vertical product. The physical economy is larger than the digital economy, and LLMs cannot touch it because physics is not language. EPT can.
Book a demo and see EPT-2 head-to-head against your current forecast provider on your region, your variables, in under 5 minutes.