How Jua is training foundational climate models with Crusoe

About Jua: Building the Future of Weather Prediction

Jua is a seed stage climate tech start-up that is creating foundational weather and climate pattern prediction models to optimize the renewable energy trade and more effectively address the climate risks associated with climate change. Jua is leveraging Crusoe® Cloud to train their large physics model to more accurately predict global weather patterns on a reliable, efficient, and scalable platform.

The Challenge: Scalable AI Compute That Doesn't Undercut the Climate Mission

AI has a crucial role to play in combating climate change. This powerful, emerging technology is already being applied in a myriad of ways to improve predictions of weather patterns, optimize energy use and enhance energy efficiency, monitor and measure changes in carbon levels, forests and icebergs, and identify sources of pollution among many other use cases. Not only can these solutions help to mitigate the effects of climate change, they can also help to save and improve lives.

However, the expanding use of AI has also led to a spike in global energy usage, which threatens to undo the positive climate benefits that AI can enable. The training of AI models and the use of AI tools require extensive computational power. Data centers must run continuously to support AI and machine learning workloads on specialized graphics processing units (GPUs) and computers that consume a significant amount of energy. Forecasts from the IEA estimate that global data centers will require an additional 160 to 590 TWhs of electricity by 2026, a staggering amount equivalent to the electricity usage of Sweden at the low end and Germany at the high end.

The solution: NVIDIA Hopper GPU clusters, Infiniband networking, and energy-first infrastructure

Jua trains on Crusoe Cloud's NVIDIA Hopper GPUs servers and S1A instances.NVIDIA H100s can train models at 5x the speed compared to NVIDIA A100 Tensor Core GPUs, increasing efficiency and reducing energy consumption. They have the ability to control their files and large data sets with high throughput connections with Infiniband networking within their own private cloud, reducing latency and further increasing the efficiency of model training. And because Crusoe is vertically integrated from the energy layer up—powering its data centers from a diversified portfolio of stranded, wasted, and clean energy sources—Jua's training runs on infrastructure that's climate-aligned by design, not by offset.

The Results: Foundational Model Training on a Climate-Aligned Platform

Jua’s mission is to help humanity live in harmony with the climate by creating the first AI ‘Large Physics Model’ to accurately predict global weather patterns. By supporting builders like Jua, Crusoe is helping advance climate-focused AI while reducing the environmental cost of the compute that runs it.