CoreWeave Expanded AI Infrastructure Stack

The company has launched a new software layer for networking, storage, and security to boost efficiency.

Updated on Sept. 30, 2026 in Data Centers

CoreWeave Expanded AI Infrastructure Stack

Live Poll

Do you trust that new AI infrastructure investments will ultimately lower costs for everyday users?

CoreWeave has expanded its infrastructure offerings beyond traditional GPU compute by integrating a new software layer called Forge. This system includes observability and security features designed to improve system-level performance for AI workloads.

Why it matters

The shift addresses the rising demand for inference over training, as customers increasingly prioritize the cost of producing tokens at scale. Organizations are moving toward system-level optimization to ensure efficiency across networking and storage components.

The new Forge system provides a closed-loop environment for evaluation, observation, and runtime curating. This platform supports a shift toward 10/90 training-to-inference workloads, representing a significant move from the current 50/50 ratio.

The players

CoreWeave

CoreWeave is a specialized cloud provider that focuses on GPU-accelerated infrastructure for high-performance computing and AI applications.

The details

CoreWeave has evolved its capabilities to include comprehensive networking and storage options alongside its existing GPU services. The company is responding to customer requests for more flexible service models, such as on-demand access, spot pricing, and shorter contract terms.

CoreWeave announced the Forge software layer on September 30, 2026.
A 10/90 training-to-inference workload ratio is expected by 2027.

The Tech Race

CoreWeave's infrastructure pivot reflects the broader industry transition from training large AI models to the operational demands of high-volume inference. By building out a proprietary software layer, the firm aims to compete against legacy cloud hyperscalers in the race for AI efficiency.

Customers seeking AI compute resources may benefit from more flexible pricing models and shorter contract requirements. These changes allow developers to optimize AI production costs and gain better visibility into the performance of their models through integrated security tools.

The takeaway

As inference becomes the dominant AI workload, businesses must prioritize end-to-end system performance rather than focusing solely on raw GPU power. Companies should evaluate their current infrastructure contracts to determine if flexible, on-demand pricing options can reduce their operational costs.