NVIDIA's Earth-2 platform is making significant strides in subseasonal climate forecasting, which involves predicting weather patterns more than two weeks in advance. This capability is crucial for managing risks and making proactive decisions in sectors highly affected by weather variations, such as agriculture, energy, and public health. Subseasonal forecasting helps farmers decide which crops to plant and how to allocate water resources, power companies to balance supply and demand, and governments to prepare for natural disasters and public health issues. For instance, predicting marine heatwaves can aid fisheries in protecting their resources, while foreseeing droughts can guide agricultural planning. AI models, particularly those supported by NVIDIA Earth-2, offer a significant advantage over traditional methods. They can run much larger operational ensembles at a fraction of the computational cost, providing more accurate and reliable probabilistic forecasts. The use of these ensembles allows for a better understanding of the likelihood of seasonal conditions deviating from the norm, expressed in terciles (above normal, near normal, below normal). The University of California, Berkeley, has developed the Bred Vector/Multi Checkpoint (BVMC) methodology, which enables the generation of multi-thousand-member ensembles, known as "Huge Ensembles" (HENS). Enterprises like JBA and AXA are leveraging HENS with the FourCastNet V2 (SFNO) model for insurance applications, demonstrating its practical utility. NVIDIA's latest release of Earth2Studio includes a new subseasonal-to-seasonal (S2S) forecasting capability, built around the Deep Learning Earth System Model (DLESyM). This model couples a multi-layer atmosphere AI model with an ocean AI model to predict sea surface temperature changes. The DLESyM architecture is a U-Net with modified padding operations to support the HEALPix grid, aiming for high-resolution forecasts at about 1-degree resolution. To facilitate the use of these models, Earth2Studio offers a new S2S recipe that simplifies the process of generating large ensemble forecasts. This recipe supports multi-GPU distributed inference and parallel I/O, making it easier to handle large datasets and storage constraints. Users can experiment with different perturbations to initial conditions and model checkpoint weights to generate skillful, calibrated ensembles. One notable example is the 2021 Pacific Northwest heatwave, which was challenging to predict accurately on S2S timescales. Despite the difficulties, all models, including HENS-SFNO, IFS ENS, and DLESyM, began predicting some level of warm anomaly in North America three weeks in advance, though with varying degrees of accuracy. NVIDIA is also working to enhance the adoption of AI for S2S forecasting by collaborating with the European Centre for Medium-Range Weather Forecasting (ECMWF) on the AI Weather Quest competition. This competition aims to accelerate community participation in advancing S2S forecasting. By integrating Earth-2 tools with ECMWF's AI-WQ-package, NVIDIA hopes to reduce the barriers to entry and enable faster iterations in model evaluation. Earth2Studio can now run and score large S2S ensemble forecasts efficiently. For example, DLESyM ensemble forecasts covering an entire year can be processed in less than two hours using eight GPUs. This efficiency is crucial for proper model assessment, which often requires scoring many forecasts to determine their skill. DLESyM has shown competitive results with ECMWF IFS, especially in weeks three through five, though it lags in the earlier weeks due to reduced model spread. Industry insiders view this development positively, noting that NVIDIA's advanced tools and collaborative efforts could significantly enhance the accuracy and accessibility of subseasonal forecasting. The integration of AI with traditional meteorological models promises to provide more robust and timely predictions, which are essential for mitigating the impacts of extreme weather events. NVIDIA Earth-2 is part of a broader ecosystem of tools and resources designed to support both scientific research and enterprise applications. Users interested in exploring these functionalities can find detailed information and tutorials in specific GTC sessions, which highlight how different industries are benefiting from AI-driven ensemble forecasting. These sessions offer valuable insights into the best practices and potential advancements in subseasonal climate prediction.