IntelligentHQ
IntelligentHQ
Skip to content

Energy, Climate & Simulation

AI and simulation for weather, climate, engineering and AI data center energy: Earth-2 open weather models and Earth2Studio, PhysicsNeMo (formerly Modulus) for physics AI, and DSX MaxLPS and DSX Flex for power use in AI factories.

Technology profiles for this category are in research.

Overview

Two audiences meet here. Weather services, energy companies and researchers want faster forecasts and engineering simulations. AI data center operators want to fit more computing into a fixed power supply and respond to the grid.

Earth-2 is NVIDIA's open stack of weather and climate models and tools, including Earth-2 Medium Range (built on the Atlas architecture), Nowcasting, Global Data Assimilation, CorrDiff downscaling and FourCastNet 3. Earth2Studio is the toolkit for running and fine-tuning those models on your own infrastructure. PhysicsNeMo, formerly Modulus, is an open-source framework under Apache 2.0 for physics AI such as surrogate models and neural operators. For AI data centers, DSX MaxLPS manages power at GPU, rack and workload level, and DSX Flex adjusts workloads to grid signals such as demand response.

AI surrogates complement validated numerical models rather than replace them, so check them against physics-based runs. For small sites, standard power monitoring may answer energy questions without DSX.1234

Problems it addresses

  • Short-range hazardous weather1

    Earth-2 Nowcasting forecasts hazardous weather from zero to six hours ahead using predicted satellite and radar imagery.

  • Coarse forecasts for local decisions1

    CorrDiff uses generative AI to downscale weather data to higher resolution.

  • Expensive engineering simulation2

    PhysicsNeMo trains surrogate models and neural operators for CFD, structural and thermal problems.

  • Power-capped AI sites3

    DSX MaxLPS manages power to fit more token output within a fixed power budget.

  • Grid constraints3

    DSX Flex responds to grid signals such as load shedding, demand response and pricing events.

A typical workflow

  1. Choose the model1

    Pick the Earth-2 model that matches the question: Medium Range, Nowcasting or CorrDiff downscaling.

  2. Set up Earth2Studio1

    Install Earth2Studio from GitHub to run and fine-tune models on your own infrastructure.

  3. Prepare initial conditions1

    Generate initial atmospheric conditions with Global Data Assimilation or use existing analysis data.

  4. Add local detail1

    Downscale the forecast with CorrDiff for the area you care about.

  5. Validate

    Compare results against observations and current operational forecasts before relying on them.

AI Factory Efficiency Lab

Model token and infrastructure costs for your own numbers.

Open the lab

Next steps

  1. Run an Earth2Studio example for your region and compare it with your current forecast source.

  2. Engineering teams: train a PhysicsNeMo surrogate on existing simulation results before replacing any solver runs.

  3. AI data center operators: model token output per megawatt in the AI Factory Efficiency Lab.

Sources

  1. NVIDIA Earth-2 (opens in a new tab)NVIDIA · Vendor-reported
  2. NVIDIA PhysicsNeMo (opens in a new tab)NVIDIA · Vendor-reported
  3. NVIDIA DSX AI Factory Platform (opens in a new tab)NVIDIA · Vendor-reported
  4. nvidia-physicsnemo (PyPI project page, maintained by the NVIDIA PhysicsNeMo team) (opens in a new tab)NVIDIA · Vendor-reported

Fill out the form below to request your copy.

Name(Required)