AI & ML
{{ topic.name }}
No items found.
Plan how you'll scale inference in 2027
Models and agents are moving into production, and 2027 budgets will follow. The decisions you make now determine how inference scales, where it runs, and whether cost, latency, and reliability hold steady as demand grows.
Plan, compare, decide
Join Naveen Chhabra, Principal Analyst at Forrester, and Shadi Saba, VP of AI/ML Engineering at CoreWeave, for a year-end look at where inference is headed in 2027. Naveen breaks down the infrastructure trends shaping AI leaders' plans. Shadi shares the trends CoreWeave sees running inference in production. You'll leave with a framework for comparing deployment options and deciding what to commit to now. We'll close with how CoreWeave Forge connects inference to the rest of the AI loop, so production results feed your next model and agent improvement.
In this webinar, we’ll cover:
- Which inference trends are likely to shape infrastructure planning in 2027
- How production inference requirements differ from experimentation, and what that changes for capacity and budget
- How to compare deployment options on cost, latency, reliability, and control
- Which decisions to make now and which to defer
- How CoreWeave Forge connects inference, observability, and training so you can turn production results into compounding improvement
- What to ask vendors and internal teams during 2027 planning
Register to take away a decision framework for your 2027 inference roadmap and start building on CoreWeave Forge.
The spice must flow: train your Mentat quartermaster
The harvest is behind schedule and the Guild shipment is due. In this hands-on virtual lab, you’ll follow a model-improvement loop in CoreWeave Forge: watch a base model fumble a desert logistics job in Agent Lens, then fix it by fine-tuning a LoRA adapter and tracking training in Weights & Biases Models. You’ll serve it on CoreWeave Inference with no GPU servers to manage, rerun the same shifts, and compare traces and scores to see what changed. You’ll also learn when fine-tuning is the right tool and when it isn’t.
What to expect
- Run - run the task with CoreWeave Inference
- Observe - inspect the base model's behavior in Agent Lens
- Curate - identify failure cases to guide the next iteration
- Improve - fine-tune a LoRA adapter; track training in Weights & Biases Models
- Evaluate - rerun shifts and compare traces and scores
Who should attend
- AI Researchers - specialists in advancing algorithms, neural network architectures, and statistical models, enabling systems to learn from data
- AI Engineers - software professionals building, deploying, and maintaining AI systems, solving real-world business problems
- AI Practitioners - professionals focused on applying AI / ML concepts to real-world business scenarios, bridging the gap between technical teams and business strategy
Win prizes for top-scoring agents
Fine-tune it, rerun it, and climb the scoreboard. The top three finishers win prizes, and first place walks away with Ray-Ban Meta glasses.
What is What’s New Wednesdays?
The CoreWeave Cloud Platform is one unified platform for AI: Forge to accelerate the loop of building and improving models and agents, and Infrastructure to power it at scale. Join us one Wednesday per month for live demos, fun conversations, and real-world advice for building models and agents using the latest product capabilities. Each session is interactive by design, so attendees can ask questions live about the latest product enhancements and get practical guidance directly from our panel of speakers.
Upcoming sessions
November 18, 2026 @ 10am PT: What’s new from GTC Berlin
GTC Berlin brought new launches across the CoreWeave platform, from running Physical AI workloads, to securing your models and data, to the next generation of NVIDIA compute. Join CoreWeave for a walkthrough of what was announced, what it means for AI teams, and where to start:
- Run Physical AI workloads: See how CoreWeave Physical AI Launchables Templates bring supported robotics and autonomous-systems workloads, such as driving scenario evaluation and synthetic scene generation, together with tested configurations, a common CLI, and inspectable results.
- Zero Trust AI in action: Explore how CoreWeave is putting its Zero Trust AI strategy into practice, with a closer look at Self-Managed CrowdStrike and how customer-controlled security capabilities help enterprises protect sensitive AI workloads.
- Plan for NVIDIA HGX Rubin NVL8: Preview the eight-GPU, HBM4-based node coming to CoreWeave Cloud this quarter, where it fits across agentic AI, data analytics, and HPC, and how to test your workload early.
December 16, 2026 @ 10am PT: What’s new in AI development
Make every experiment more informative and every iteration count. Join CoreWeave for live demos and technical insights into the latest capabilities and workflows for AI researchers and developers. More details to follow.
- Build more reliable AI: Learn techniques for using evaluation and observability to diagnose failures and guide improvements in models and agents
- Put new capabilities to work: Get implementation tips and practical examples you can adapt to your own research and development workflows
- See what’s new in CoreWeave Forge: Get a live look at the latest Forge releases for running, observing, curating, improving, and evaluating models and agents in one environment, open to any model, framework, or cloud
We’ll close each session with open Q&A, giving you direct access to our panel of AI experts to workshop your real-world challenges.
Turn experiment data into your next research decision
Every training run generates valuable data. Keeping research moving means turning those results into confident decisions about what to try next. But when the evidence is scattered across experiments and tools, understanding what drove performance and which findings to act on takes a lot of work.
Join CoreWeave for a technical deep dive into AI-assisted model development. See how CoreWeave Forge streamlines experiment tracking and supports deeper analysis and more efficient research workflows, so you can spend less time piecing together results and more time putting insights to work.
You’ll learn how CoreWeave Forge helps you:
- Compare training runs and identify promising model configurations using Weights & Biases Models
- Investigate training behavior through custom analysis, visualizations, and debugging in CoreWeave Notebooks
- Connect AI coding agents through MCP to examine results and troubleshoot issues
- Use CoreWeave ARIA to analyze experiments and automate repetitive research tasks that slow iteration
See live demos of how Forge workflows help you identify high-performing checkpoints and diagnose training anomalies. Explore how AI agents can help you analyze experiment results and decide what to try next.
How You.com, NVIDIA, and CoreWeave Improved Nemotron 3.5 Lightning with RL Rollouts
Reinforcement learning moves quickly, but training and rollout serving don’t always move at the same cadence. Policy updates may land every few hours. Getting new checkpoints onto inference workers while keeping serving available adds another infrastructure problem to solve.
Join engineers from You.com, NVIDIA, and CoreWeave for a technical walkthrough of how they customize NVIDIA Nenmotron 3.5 Lightning for complex web research using You.com Search, supervised fine-tuning (SFT), reinforcement learning (RL), and RL Rollouts powered by NVIDIA Dynamo on CoreWeave infrastructure. Then see the mechanisms come together in a worked example with Nemotron 3.5 Lightning on NVIDIA HGX B200. NVIDIA Nemotron 3.5 Lightning is a customizable open model for always-on agents to complete specialized tasks faster with leading domain-specific accuracy across local, on-premises, and cloud deployments. The walkthrough will cover before-and-after traces on the same task, the compute cost and timeline, and the reported results: a 22.94% accuracy gain and 30.24% fewer tool calls. Leave with a blueprint for applying the approach to your own model.
In this webinar, we’ll cover:
- How post-training fits into the model-capability stack, and why search infrastructure matters for RL
- How You.com builds training questions from the web graph, how that differs from open datasets, and why decontamination matters
- How filtering training tasks by difficulty can improve accuracy and training efficiency
- How RL Rollouts handles delta weights, checkpoint hot-loading, and on-policy weight transitions to keep inference synchronized with policy updates
- Where per-token training metadata, including logprobs, expert routing, and token IDs, enters the response contract
- A worked Nemotron 3.5 Lightning example on NVIDIA HGX B200, with before-and-after traces on the same task, compute cost, timeline, and results
Register to see how the pieces fit together and take away a blueprint for your own post-training workflow.
Give agents room to work. Set boundaries on what they can reach.
As agent, evaluation, and reinforcement learning workloads scale, so does the challenge of controlling what their code can access. How do you give each task the environment it needs while keeping execution within your platform’s policies?
Watch this 40-minute on-demand walkthrough of CoreWeave Sandboxes. See how one API connects execution across your CoreWeave Kubernetes Service (CKS) clusters and serverless, building on the infrastructure and governance controls you already use. Learn how to isolate workloads, control network access, and scale execution next to your models and GPUs
In this session, we’ll cover:
- What a sandbox is, and how one API places CPU and GPU sandboxes across CKS and serverless
- How to set guardrails once to keep agent requests within policy
- How opt-in spillover sends new sandboxes to serverless when your CKS clusters are full
- How to control network access and preserve the state your agents need between runs
- How to run agents, fan out evaluations, and verify RL rewards at scale
Walk away with a clearer plan for where your agent or evaluation workflow should run and how to launch your first sandbox.
Bringing NVIDIA Vera Rubin to production for agentic AI
Agentic AI needs infrastructure that keeps up with longer reasoning chains and nonstop inference. In January, CoreWeave announced we would be among the first cloud providers to deploy the NVIDIA Vera Rubin platform. In June, we were the first cloud provider to bring up and validate the NVIDIA Vera Rubin NVL72. In July, we delivered its first measured silicon performance numbers. In September, we provided the first customers with production NVIDIA Vera Rubin NVL72 instances. And next, we’re bringing the Vera CPU and HGX Rubin NVL8 to CoreWeave Cloud.
In this webinar, we’ll walk through the Vera Rubin family, what it took to bring Vera Rubin NVL72 to production, and the engineering behind CoreWeave Cloud that helps teams run agentic AI at scale.
In this webinar, we’ll cover:
- The difference in the NVIDIA Vera Rubin platform, and why it matters for agentic AI
- How AI infrastructure needs are changing as AI moves from training toward reasoning and agentic workloads
- How CoreWeave engineers its rack-scale cloud, including Valvey for liquid-cooling management and Racky for rack management
- CoreWeave’s ability to bring new NVIDIA platforms to customers across generations
- What’s next for Vera Rubin on CoreWeave and how to get involved
Ready to explore what’s next for agentic AI infrastructure? Register now to join us live.
As AI workloads move from the lab to production, the economics of intelligence has become one of the most consequential factors platform leaders face.
Watch this session for a discussion of Signal65’s in-depth economic study of training and inference costs across CoreWeave and leading hyperscalers. The analysis examines cloud economics, performance benchmarks, and multi-year total cost of ownership to show how infrastructure choices affect the economics of AI over time.
Going beyond simple compute pricing, this session explores the hidden costs that compound at scale and how infrastructure economics translate directly into token costs at production volume. You’ll leave with a practical framework for evaluating fully loaded economics across the complete training-to-inference lifecycle.
In this webinar, we covered:
- How training and inference costs compare across CoreWeave and leading hyperscalers
- What performance benchmarks reveal about the relationship between infrastructure and AI economics
- Which hidden costs can compound as workloads scale from development to production
- How infrastructure decisions translate into token costs at production volume
- How to evaluate fully loaded TCO across the training-to-inference lifecycle
- What platform leaders should consider when comparing infrastructure options for long-term AI workloads
AI is rapidly entering the world of engineering, but how capable are today's large language models when faced with real physics, real data, and real design challenges?
In this one-demand webinar, the CoreWeave Physical AI team reveals insights from its LLM Benchmarking Study, in which we tested models across real engineering workflows on the CoreWeave platform, including Claude 4, Grok 4, Gemini 2.5 Pro, and ChatGPT o3.
Uncover which models show genuine engineering reasoning, which struggle with complex analysis, and what these results tell us about current and future capabilities of AI in engineering.
Webinar highlights
- How we benchmarked today's top LLMs, tested on realistic engineering scenarios inside CoreWeave's MCP framework
- How effectively LLMs understand engineering problems, from feature selection and model training to reasoning through physics-based constraints
- Where current models fall short and where they excel, drawing insights into performance, repeatability, and engineering intuition
- How AI is lowering the barrier to engineering analysis, making advanced problem-solving faster, easier, and accessible to more engineers
Who should watch?
This session is for engineers and technical leaders who want to stay ahead as AI reshapes engineering workflows:
- Engineers who are curious about advanced AI in engineering and how it could be applied in real R&D and testing contexts
- Professionals who want to stay ahead of the curve and see how new AI standards like MCPs are shaping engineering
- Early adopters in R&D and testing who are eager to explore practical, cutting-edge applications before they become mainstream