CoreWeave launches Interconnect, SUNK Anywhere and LOTA Cross-Cloud

A direct link between CoreWeave Cloud and Google Cloud

CoreWeave Interconnect is a private, fiber-based network interconnect that directly connects CoreWeave Cloud to Google Cloud. CoreWeave built it in collaboration with Google Cloud and says it removes the need to manage additional equipment and providers.

According to CoreWeave, the traditional approach to linking two clouds relies on a third-party provider. That leaves organizations operating across three separate environments, with added work in circuit provisioning, routing policies, security validation, performance testing and observability, plus extra contracts to negotiate. The company says such deployments often take months, typically increase latency, create new points of failure and may lack features such as link-layer encryption.

CoreWeave says the Interconnect minimizes latency, helps ensure maximum reliability and provides built-in security through MACsec encryption at line rate. It uses Google Cloud's Partner Cross-Cloud Interconnect to form a unified cross-cloud network fabric. The service will be available in private preview soon, in select regions, and customers will be able to choose the bandwidth they need.

Chen Goldberg, EVP of Product and Engineering at CoreWeave, said: "By removing the friction of cross-cloud, we are making it simpler and faster for organizations to run workloads anywhere." Rob Enns, VP/GM of Cloud Networking at Google Cloud, said the collaboration gives customers "the secure, low-latency connectivity needed to simplify complex cross-cloud workloads and accelerate AI outcomes."

CoreWeave also points to its own network. It says it operates 44 data centers worldwide, connected by a private, carrier-grade backbone secured with MACsec encryption over terrestrial and subsea fiber. The company quotes capacity from 400Gbps to terabits and says there are no data transfer fees across its network or at egress.

SUNK Anywhere extends the training system beyond CoreWeave

SUNK is CoreWeave's system that combines Kubernetes with the Slurm job scheduler. SUNK Anywhere extends it to GPU environments across clouds or on-premises. According to CoreWeave, platform and research teams can add capacity on Google Cloud, AWS or Azure without rebuilding the workflows and operational model they already use on CoreWeave.

The company describes SUNK as a unified control plane for scheduling, scaling and running distributed workloads across environments. It lists use cases including large-scale training that bursts across regions, hybrid setups spanning on-premises clusters and cloud GPUs, multi-tenant environments that need strict isolation and prioritization, and iterative experimentation where researchers spin up, pause and scale jobs. SUNK Anywhere offers a managed Slurm on Kubernetes environment with automated provisioning and visibility into job performance and resource utilization.

Xander Dunn, Member of Technical Staff at Periodic Labs, is quoted in the announcement. He says his team runs large distributed jobs across thousands of GPUs on both CoreWeave and non-CoreWeave providers, and that deploying SUNK on non-CoreWeave clusters "requires very few configuration changes."

LOTA Cross-Cloud and Weights & Biases on Google Cloud

The third capability, LOTA Cross-Cloud, concerns data access. CoreWeave says a training job is only as strong as the data it can reach, and ties the feature to CoreWeave AI Object Storage. The source text provided for this article cuts off before the details of LOTA Cross-Cloud, so its specifics are not covered here.

CoreWeave also announced new integrations between its Weights & Biases tools and Google Cloud. It says customers including Noble Machines and PathAI use the tools on Google Cloud to accelerate AI workflows, and that the integrations let customers observe agents and continuously improve their reliability.

What it means for AI teams

CoreWeave frames the shift as a change in motivation: many organizations first adopted multi-cloud for resiliency, while teams now want speed, agility and throughput from their accelerated platforms. The three tools aim at the network, compute and storage layers of that problem.

For infrastructure engineers, the main point is fewer moving parts, with one provider-to-provider link instead of a third-party network and one scheduling system across several clouds. Availability is still limited, since Interconnect is only entering private preview in select regions, and the announcement gives no pricing for the new capabilities.