CoreWeave, a company that once made money mining Ethereum, just unveiled a platform that trains AI models from real-world usage in a safe and isolated environment. The product is called CoreWeave Sandboxes and targets three specific use cases: reinforcement learning, using AI agent tools, and large-scale model evaluation.
Most current AI training is static. When you feed a dataset into a model, it calculates numbers and you can expect it to work well in a real-world environment. CoreWeave Sandboxes are built for a different paradigm, where models continually adapt through real-world interactions, rather than sitting in a training lab waiting for the next batch of cherry-picked data.
What the CoreWeave sandbox actually does
The platform provides what CoreWeave calls an “execution layer” for dynamic AI workloads. This is a set of secure walled environments where AI agents can operate, learn, and evaluate without contaminating production systems or each other.
Businesses can run the sandbox on CoreWeave’s own infrastructure or choose a serverless model through a partnership with Weights & Biases, a machine learning operations platform commonly known as W&B.
This proposal focuses on two selling points. First, continuous online learning. This means you don’t just train and deploy your model once. We continue to improve it based on actual usage. The second is improved GPU utilization. According to CoreWeave, this directly translates into reduced training costs.
From crypto mining to AI cloud infrastructure
The story of CoreWeave’s creation is one of the most dramatic turning points in technology. The company was founded in 2017 as Atlantic Crypto, a GPU mining operation focused on Ethereum. When the economics of mining started to become less attractive, the team realized that the same GPU hardware powering mining rigs could serve a much larger market: AI computing.
Rebranded to CoreWeave in 2019, the company has since transformed into a specialized cloud provider that competes with the likes of AWS, Google Cloud, and Microsoft Azure for GPU-intensive workloads. It currently operates data centers in both the United States and Europe, and trades on Nasdaq under the ticker CRWV.
The sandbox launch builds on existing relationships that show where CoreWeave is headed. The company has a deep partnership with NVIDIA, including the integration of advanced hardware like the HGX B300 specifically for agent inference workloads. The collaboration with W&B added a layer of software to refine the workflow for online assessment and continuous learning.
Notably, none of the presentation materials include cryptocurrencies. There are no tokens, no references to blockchain, and no homage to the company’s mining past.
