CoreWeave Launches Forge Platform for Full-Stack AI Cloud

Specialized cloud provider CoreWeave unveiled CoreWeave Forge, an integrated AI development platform designed to connect training, inference, and evaluation into a continuous improvement loop. Announced at the Fully Connected 2026 conference, the platform marks a deliberate evolution for the company, expanding from high-performance hardware leasing into an open, full-stack artificial intelligence cloud. Rather than isolating infrastructure tiers, Forge combines serverless reinforcement learning sandboxes, observability, and fine-tuning tools with advanced hardware, directly targeting enterprise platform teams that require specialized alternatives to legacy hyperscalers.

As enterprise spending pivots from exploratory prototyping to production inference and agentic systems, infrastructure buyers face persistent integration bottlenecks across fragmented tooling. CoreWeave is positioning Forge to capture this demand by bridging operational telemetry back into model and agent optimization across multi-cloud environments.

From GPU Rentals to an Integrated Developer Platform

Historically recognized for securing massive allocations of high-demand accelerators, CoreWeave is moving beyond raw capacity. As analyzed in CoreWeave’s transition from GPU scarcity to a durable enterprise cloud, specialized providers must deliver durable software value to avoid commoditization as hardware supplies stabilize.

Speaking at the Fully Connected 2026 conference, CoreWeave Chief Marketing Officer Jean English outlined the strategic imperative behind the announcement. In a discussion with theCUBE Research broadcast by SiliconANGLE, English emphasized that modern AI workflows demand an open, full-stack approach rather than siloed infrastructure. “We believe that the loop should be connected. It should be open to different models, different frameworks, different clouds,” English explained, contrasting CoreWeave’s modular approach with closed proprietary ecosystems.

According to reporting from SiliconANGLE, CoreWeave Forge establishes a unified developer environment where engineering teams can run, observe, curate data, improve, and evaluate models in production. By capturing runtime telemetry and live inference edge cases, teams can directly trigger targeted fine-tuning and reinforcement learning cycles.

Connecting the Loop Across Silicon and Software

The technical foundation of Forge relies on tight integration between purpose-built physical infrastructure and upper-layer developer tooling. CoreWeave was among the first cloud operators to bring up and validate NVIDIA’s Vera Rubin NVL72 architecture alongside Vera CPUs, according to technical details shared on the NVIDIA Blog. This silicon deployment is tailored for multi-agent reasoning, high-throughput inference serving, and massive inter-GPU bandwidth. The company also secured SemiAnalysis’ Platinum ClusterMAX rating for a third consecutive evaluation cycle, affirming its low-latency networking fabric.

Above the hardware substrate, Forge integrates both proprietary runtime services and established third-party developer frameworks. As documented in CoreWeave’s platform announcement, Forge features native interoperability with partner tools including Weights & Biases for experiment tracking and OpenPipe for automated fine-tuning. Enterprise teams can also spin up serverless reinforcement learning sandboxes to test reasoning agents without managing dedicated cluster state.

Enterprise Adoption and Hyperscaler Competition

CoreWeave’s full-stack pivot directly challenges traditional hyperscalers, which have historically treated AI as an ancillary compute workload layered atop web-oriented virtualization stacks. As industry analysis from Constellation Research highlights, neoclouds are attempting to compete on workload specialization, raw execution velocity, and purpose-built networking rather than expansive enterprise feature sprawl.

To substantiate these performance claims, early enterprise adopters have begun deploying Forge across complex agent pipelines. According to analysis published by the Futurum Group, AI software vendor Cognition reported performance gains of up to 4.8x running production workloads on CoreWeave’s Vera Rubin NVL72 infrastructure. The broader CoreWeave Partner Network further aggregates vetted software vendors across data governance, vector storage, and security to ease enterprise procurement.

However, enterprise buyers navigating this shift must evaluate substantial architectural trade-offs. While CoreWeave stresses support for multi-cloud deployments, migrating mission-critical data pipelines out of established environments like AWS, Google Cloud, or Microsoft Azure introduces egress overhead and operational dual-stack management.

What happens next

Following the Fully Connected unveiling, CoreWeave is gradually expanding Forge access across its enterprise customer base. Engineering leaders and cloud architects evaluating the platform should assess whether their immediate development bottlenecks stem from hardware scarcity or fragmented evaluation workflows. For teams primarily running straightforward inference endpoints, conventional managed container services may remain sufficient without introducing another cloud vendor.

In contrast, organizations deploying multi-step autonomous agents, complex fine-tuning pipelines, or active reinforcement learning loops stand to gain measurable velocity from unified observability and specialized silicon. Over the coming quarters, industry observers will watch whether CoreWeave’s partner-centric strategy can retain enterprise loyalty against aggressive price cuts and bundled ecosystem incentives from legacy hyperscalers.