Nvidia announced a strategic partnership with Cloverleaf Infrastructure, a data center site-development company founded in 2024 that connects utility companies with data centers, providing power sources and other critical infrastructure. Though deal terms went undisclosed, reports indicated Nvidia’s investment could total several hundred million dollars, giving the chipmaker a minority stake in Cloverleaf. The partnership extends Nvidia’s broader strategy of directing its profits toward financing and developing the data centers that ultimately purchase its AI systems, following an earlier $1.5 billion investment this week in SB Energy, an OpenAI-linked data center project in Ohio.
In other company news, Nvidia published research suggesting that the software harness surrounding an AI model, rather than the underlying model itself, plays the larger role in enabling long-horizon task performance. Using a custom harness called Agentic Variation Operators, which incorporated a supervising agent component to redirect the system when it strayed off course, researchers achieved a perfect score for Claude Opus 5 on the ARC-AGI-3 benchmark, compared to 30 percent without the harness. Adel El Hallak, vice president of product in Nvidia’s AI unit, said an agent should be understood as the combination of a model with its surrounding scaffolding, tools, and runtime, rather than simply an interface to the model itself.
The findings echoed similar conclusions from OpenAI and Databricks, both of which found that harness design significantly affected model accuracy and operating costs. Databricks CEO Ali Ghodsi noted that identical models could produce dramatically different costs depending on harness choice. El Hallak said Nvidia’s broader goal was demonstrating that open, controllable harnesses across infrastructure and runtime layers were essential to advancing the agentic ecosystem securely.
