November 2026 Summaries
6 posts from Coreweave
Filter
Month:
Year:
Post Summaries
Back to Blog
No summary generated yet.
Nov 16, 2026
1,191 words in the original blog post.
No summary generated yet.
Nov 16, 2026
1,220 words in the original blog post.
CoreWeave Database is now generally available as a managed, PostgreSQL-compatible service that runs on CPU nodes within CoreWeave Kubernetes Service clusters alongside AI training and inference workloads. It supports migration from existing PostgreSQL deployments using native tools and uses a primary-and-replica architecture, CoreWeave Distributed File Storage, automatic failover, continuous backups with point-in-time recovery, encryption, audit logging, and PostgreSQL-based access controls. By keeping application state, experiment metadata, agent workflows, and RAG vector queries within the same cluster, the service aims to reduce cross-cloud networking, latency, operational complexity, and security boundaries. CoreWeave manages database replication, patching, TLS certificates, backups, and monitoring while customers retain control of their data, schemas, queries, and users. The offering is built on the open-source CloudNativePG operator and includes CoreWeave contributions to related PostgreSQL ecosystem tools such as Barman and pgexporter.
Nov 11, 2026
641 words in the original blog post.
CoreWeave presents itself as a purpose-built AI cloud provider that has grown through customer-contracted demand, disciplined capital investment, and rapid infrastructure deployment. The company reported quarterly revenue of $3.7 billion, up 168% year over year, adjusted operating income of $379 million, low-double-digit operating margins ahead of schedule, and approximately $133 billion in revenue backlog. Its active power capacity exceeded 1.7 gigawatts across global data centers, including expansion into Indonesia, India, Denmark, and Spain, and it launched CoreWeave Forge to unify AI development, deployment, and improvement tools. CoreWeave argues that its advantage stems from operating large-scale integrated systems of compute, storage, networking, and software rather than simply supplying GPUs, while financing expansion through customer prepayments, contracted cash flows, loans, and operating cash flow rather than unsecured debt. It says its platform serves nine of the top 10 AI foundation model providers, AI-native companies, and enterprises such as Caterpillar, Nissan, and Jane Street, and frames its longer-term role as enabling broader innovation by making advanced AI infrastructure accessible to researchers, businesses, and individuals.
Nov 10, 2026
1,354 words in the original blog post.
CoreWeave announced its Platinum membership in the Cloud Native Computing Foundation at KubeCon + CloudNativeCon North America, expanding its support for the Kubernetes ecosystem that underpins its AI cloud platform. The announcement coincided with Principal Software Engineer Rita Zhang receiving a Lifetime Achievement award for a decade of Kubernetes and CNCF community contributions. CoreWeave uses Kubernetes across its infrastructure, including bare-metal GPU nodes, managed CoreWeave Kubernetes Service environments, and shared GPU pools for Slurm and Kubernetes workloads, while contributing to projects such as Kubernetes, Kueue, CloudNativePG, and llm-d. At the event, its engineers are presenting sessions on Kubernetes authorization, topology-aware scheduling for GPU clusters, batch-workload developments, and scalable inference pipelines, and the company is demonstrating AI training, inference, agent development, and Forge at booth 147. CoreWeave also highlighted its managed Kubernetes, inference, sandbox, and development offerings for organizations building AI applications.
Nov 10, 2026
834 words in the original blog post.
Turbulence remains a difficult scientific and engineering problem with major implications for transportation drag, energy consumption, emissions, cooling, combustion, and industrial design, driving demand for increasingly detailed computational fluid dynamics simulations. Convergent Science and CoreWeave evaluated the CONVERGE CFD GPU solver using direct numerical simulation, which computes all turbulence scales without modeling approximations but requires exceptionally high memory capacity, compute power, and rapid communication between GPUs. Their study assessed weak scaling, where both case size and GPU count grow together, and strong scaling, where more GPUs reduce the runtime of a fixed case, across NVIDIA RTX PRO 6000 Blackwell Server Edition systems and NVIDIA HGX B200 infrastructure. The RTX PRO configuration was positioned for everyday simulations that fit within one server, while HGX B200 systems, with larger memory, higher bandwidth, NVLink, and 400 Gb/s InfiniBand networking, were designed for multi-node workloads; the team completed a 536-million-cell DNS case on 16 B200 GPUs in 3.62 hours. The results emphasize that CFD job feasibility is determined first by memory requirements, which vary with modeled physics, numerical precision, solver storage, and adaptive meshing, while scaling efficiency depends heavily on minimizing GPU synchronization and network overhead through direct GPU communication, non-blocking networking, and topology-aware workload placement.
Nov 02, 2026
2,417 words in the original blog post.