Neocloud market leader CoreWeave wants customers to pull their AI-relevant unstructured data out of the hyperscalers and into its own facilities, and is mounting an attack on public cloud egress fees to make it happen.
Its Zero Egress Migration (0EM, not ZEM) program has CoreWeave paying the data exit egress fees for moving a customer’s dataset in AWS, Azure, Google Cloud, or others to CoreWeave’s AI Cloud, its GPU server farm datacenters. It is offering a fully managed service providing secure, end-to-end validated and high-speed data migrations. Specifically, it will cover egress fees for typical AI workloads from third-party cloud networks for the initial migration without additional egress fees, meaning no Coreweave egress fees.
Peter Salanki, CoreWeave co-founder and CTO, said: “Today’s industry standard for data storage is not designed for the era of AI, where massive datasets need to move fast and freely within and between clouds. Penny pinching egress and transfer fees do little but slow down innovation.”
CoreWeave’s AI Cloud features a single global dataset that, it says, prevents additional capital, time, and resource waste from data replication and data sprawl. Its Local Object Transport Accelerator (LOTA) technology delivers 7 GB/s per GPU throughput, regardless of where in CoreWeave’s distributed datacenter estate the data is located.
The 0EM program applies to AWS, Azure, Google Cloud, IBM, Alibaba, and others. Customers can retain active accounts with third-party cloud providers during and after the migration and will not incur any exit penalties from CoreWeave after.
A migration dashboard lets customers monitor data migration in real time, showing migration progress, performance metrics, and data integrity checks.
As we understand it, customers will indirectly pay the migration costs through their CoreWeave GPU rental bills and CoreWeave can afford this because it will get new GPU rental business. Moving 500 TB out from Amazon’s S3 storage in a month could cost between $25,000 and $45,000. The cost varies by S3 tier and AWS region.
Initial customers such as Hologen and Periodic Labs praised the program. Hologen co-founder and CTO Chris Goddard said: “Without any egress fees, we’re free to move data in and out of CoreWeave and innovate without limits.”
Xander Dunn, Member of Technical Staff at Periodic Labs, said: “CoreWeave AI Object Storage delivers the throughput and reliability our research pipelines depend on, balancing speed and efficiency across active and inactive data. It’s allowed us to experiment faster while keeping costs under control.”
CoreWeave’s AI Object Storage is separate from its VAST Data infrastructure.
Salanki reckons: “With AI Object Storage and 0EM, CoreWeave is creating the most supportive and performant cloud ecosystem for AI data management, eliminating physical, digital, and financial barriers to progress.”
CoreWeave says it is uniting the essential tools for AI on one cloud platform, bringing together high-performance computing, multi-cloud-compatible storage, a public cloud data exit strategy, and the software layer builders rely on to develop, test, and deploy AI at scale.
The CoreWave 0EM program is generally available now. Read more about in a technical blog.
