Photo: Tima Miroshnichenko / Pexels
CoreWeave CEO unveils Forge, a cloud product designed to boost margins
The AI infrastructure company is betting that software layers on top of its GPU cloud will transform its economics and lock in customers.
CoreWeave just made its clearest play yet to stop being called “just a GPU cloud company.” The firm launched CoreWeave Forge, an integrated AI software platform, at its Fully Connected AI conference in San Francisco, with CEO Michael Intrator framing it as a margin-expanding engine that shifts the business into higher-value territory.
What Forge actually does
Forge unifies the messy, fragmented lifecycle of building AI into what CoreWeave calls an “AI loop.” That loop covers four phases: run, observe, evaluate, and improve. Instead of stitching together a dozen different tools from different vendors, teams can handle training, inference, and reinforcement learning inside one platform that sits on CoreWeave’s bare-metal NVIDIA GPU clusters.
The platform bundles several purpose-built components. CoreWeave Notebooks provides the development environment. Agent Lens handles observability, giving engineers a window into how their models and AI agents are actually behaving in production. Sandboxes offer isolated testing environments. And ARIA, CoreWeave’s own coding agent, is baked in to assist with development tasks.
Forge also plugs into third-party tools that AI teams already rely on, including Weights & Biases for experiment tracking and OpenPipe for fine-tuning pipelines.
Performance claims are aggressive. CoreWeave says certain workloads on Forge achieve up to 96% goodput, a metric that measures the proportion of GPU compute actually doing useful work versus sitting idle. Some training jobs reportedly run 1.8x faster compared to fragmented setups.
The margin math
Reports from the company’s recent financial performance indicate that contribution margins on contracts have already been expanding. Forge is expected to accelerate that trend, with projected margin gains in the range of 5 to 10 percentage points.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The strategy also targets customer stickiness. When a company’s entire AI development workflow runs on Forge, switching to a competitor becomes painful and expensive. That kind of lock-in improves unit economics over time because customer acquisition costs get amortized across longer, more predictable revenue streams.
Who’s already using it
CoreWeave didn’t launch Forge into a vacuum. The company highlighted MasterClass and Canva as early adopters, two brands with very different AI use cases.
Forge is available across three tiers: Free, Pro, and Enterprise. The free tier serves as an on-ramp, letting smaller teams and individual developers experiment without financial commitment.
The competitive landscape gets tighter
CoreWeave’s pivot into platform territory puts it on a collision course with some formidable competitors. AWS, Google Cloud, and Microsoft Azure all offer integrated AI development environments. Specialized players like Lambda Labs and Together AI are also building out software layers on top of GPU clouds.
What CoreWeave brings to the fight is specialization. Unlike the hyperscalers, which offer AI as one of hundreds of cloud services, CoreWeave built its entire operation around GPU workloads from the start. That focus has historically allowed it to offer better performance per dollar on AI-specific tasks, and Forge extends that advantage into the software layer.