CoreWeave has launched Forge, a unified development environment designed to manage the lifecycle of AI models and agents across multiple clouds and frameworks. The platform integrates tools like Weights & Biases and OpenPipe to streamline model training, evaluation, and continuous improvement for enterprise customers.
This release addresses the operational friction caused by fragmented AI toolchains where training and production data often remain siloed. By centralizing the 'AI loop' of observation and improvement, CoreWeave aims to reduce the time lost during handoffs between separate vendor tools. For the streaming and creative sectors, this infrastructure allows companies like MasterClass to iterate on generative features more rapidly without managing disparate backend services. As enterprise AI moves from experimental to production phases, the focus shifts toward these unified environments that prioritize continuous model evaluation. Watch for whether this integrated approach successfully draws high-compute workloads away from hyperscale competitors like AWS or Azure.
CoreWeave Forge arrives as the company deepens its enterprise AI platform strategy beyond raw GPU compute. The platform integrates Weights & Biases for experiment tracking and OpenPipe for data curation into a single development environment, targeting organizations that run model training and inference across multiple cloud providers. CoreWeave announced Forge alongside partnerships with Canva and MasterClass at its September 2026 launch event, positioning the product as a lifecycle management layer that reduces handoff friction between training, evaluation, and production deployment.
The launch reflects CoreWeave's broader push to move up the stack from infrastructure provider to platform vendor. CoreWeave's IPO in March 2025 valued the company at approximately $23 billion, making it one of the largest AI-focused public offerings and giving it capital to invest in higher-margin software layers. The Forge platform represents a direct attempt to capture that margin by locking in enterprise customers through integrated tooling rather than competing solely on GPU pricing against hyperscalers.
In the same-category competitive landscape, CoreWeave has already begun embedding agent-based tooling into the Weights & Biases platform that Forge orchestrates. CoreWeave ARIA, introduced in July 2026, is a coding agent built directly into Weights & Biases that reads experiments, builds live visualizations, and provides full project context, suggesting that Forge's unified environment is part of a layered product strategy rather than a standalone release. The marimo notebook environment included in Forge signals an intent to capture the developer experience layer as well, competing with Jupyter-based workflows that dominate data science teams at streaming and creative companies.
CoreWeave has launched Forge, a unified development layer designed to manage the full lifecycle of AI models and agents. By integrating tools like Weights & Biases and OpenPipe, the platform centralizes training, inference, and evaluation. This helps companies like Canva and MasterClass reduce operational friction and accelerate generative feature development.
CoreWeave Forge is a unified development layer that manages the full lifecycle of AI models and agents, allowing engineering teams to execute training, inference, and evaluation within a single environment.
Creative platform Canva and streaming educator MasterClass are among the early enterprise adopters of the CoreWeave Forge platform.
Forge reduces friction by centralizing the AI loop of observation and improvement, preventing data from remaining siloed and minimizing time lost during handoffs between separate vendor tools.
Forge integrates Weights & Biases for experiment tracking and OpenPipe for data curation into a single development environment, alongside the CoreWeave Registry for versioning model checkpoints.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source
