CoreWeave growth playbook

Generate and edit transcripts for the 75 empty pages, starting with the 15 customer talks. For each customer talk, publish a one-page written summary (the problem, the stack on CoreWeave, the numbers the speaker gave, a quote) and link it from the Agentic AI page, the case-study hub and the matching product pages. Ask each speaker's company to approve the summary so it can count as a reference story.

Buyers, search engines and AI assistants read text, not video. Transcripts make the talks findable for the questions agent builders ask, and the summaries give sales and the Agentic AI page named proof within weeks instead of waiting for new case studies.

[ ] Export captions for the 15 customer talks
[ ] Correct names (models, frameworks, products)
[ ] Fill the Transcript tab and render it as page text
[ ] One-page summary: problem, stack, numbers, quote
[ ] Customer approval as a reference story
[ ] Link from /solutions/agentic-ai and /case-studies
[ ] Add VideoObject data with the transcript
[ ] Repeat for the remaining 60 pages

• Talks contain claims a customer would not put in writing

  Publish only the approved summary; keep the raw transcript as given on stage.

• Machine transcripts mangle model and product names

  Edit against a glossary of CoreWeave, NVIDIA and model names.

• Your websiteVideo library (all 115 /resources/videos pages checked October 21, 2026): every page has a Transcript tab, but on 75 of them it is empty. Empty ones include customer talks such as "How Grammarly Scales Real-Time Inference with vLLM on CoreWeave", "How MasterClass Builds Multi-Agent AI That Learners Can Trust", "Training Cognition SWE-2: Asynchronous RL at Global Scale", "Building the Document Context Layer for AI Agents at LlamaIndex", "Beyond the Benchmark: How Parallel Web Systems Builds Reliable Web Agents", "Building the Voice OS at Willow", "Inside Pinterest's Machine Learning Platform", Nasdaq, IBM Granite 4.0 and You.com's Nemotron agentic search; the 40 filled ones include product demos such as the ARENA demo.Source
• Your websiteCustomer stories: the sitemap lists 17 written case studies. The hub features Cohere ("North, Agentic AI for Enterprise"), OpenAI, IBM ("80%") and Mistral AI ("2.5x Faster Training"); Cohere is the only one about agents, and none is about robotics or autonomous vehicles. Several are older stories (NightCafe, Zeet, Odyssey's McDonald's metaverse experience, Bit192 GPT-NeoX-20B).Source
• Your websiteAgentic AI solution page: H1 "Agentic AI that paves the path to superintelligence", sections on training, inference and observing agents, "Exemplary agent inference. Validated by NVIDIA." and 11 FAQs, but no customer story or named agent builder.Source
• HiringDirector of Product Marketing, Agentic AI (open, posted June 17, 2026): position CoreWeave "as the premier infrastructure platform for agentic AI"; build "use-case playbooks" for "autonomous coding agents, multi-agent reasoning pipelines"; and "Build and scale a customer proof program featuring case studies and reference customers across AI labs, enterprises, and high-growth AI startups deploying agents in production"; plus thought leadership "white papers, technical blogs, benchmark reports, webinars".Source
• HiringSenior Manager, Marketing Research & Intelligence (posted September 10, 2026): "Develop research and measurement frameworks for new / emerging audience and platform discovery channels, including Generative AI / LLMs"; size TAM and SAM; "Track market share against key competitors"; identify buyer groups by use case; and "push account targeting signals into marketing automation and CRM platforms"; success includes "identifying accounts and prospects that are ready to buy".Source

Write CoreWeave case studies for the five named customers, covering the full stack (compute, storage, W&B tracking, Launchables where used), publish them under /resources/case-studies with a Physical AI filter, and keep the W&B pages as secondary links. Use them as the first stories in the new communications program and in the launch of Physical AI Launchables.

Proof on your own domain builds your own search and AI-assistant visibility for "GPU cloud for robotics" questions, and keeps a visitor who is evaluating CoreWeave on CoreWeave.

• Customers already approved W&B wording only

  Start from the approved text and add only new, approved facts.

• Your websitePhysical AI page: five customer cards (Decart, NEURA Robotics, PhysicsX, Woven by Toyota, Nissan "reduces physical vehicle tests by 17%"). Decart, NEURA and Toyota link to customer pages on wandb.ai, Nissan to a news post; none links to a CoreWeave case study.Source
• HiringPhysical AI hiring: a Senior Manager, Physical AI Communications, the "first dedicated communications owner for CoreWeave's Physical AI business", to "Create a repeatable pipeline of customer and partner stories across robotics, autonomous systems, industrial applications, and scientific AI" and measure "share of voice"; and a Business Development Manager - Physical AI to own partnerships with "simulation ISVs, robotics platform companies, OEMs" and the NVIDIA Physical AI ecosystem (Isaac Sim, Isaac Lab, Cosmos, Omniverse, DRIVE), with "joint reference architectures".Source
• Your websiteCustomer stories: the sitemap lists 17 written case studies. The hub features Cohere ("North, Agentic AI for Enterprise"), OpenAI, IBM ("80%") and Mistral AI ("2.5x Faster Training"); Cohere is the only one about agents, and none is about robotics or autonomous vehicles. Several are older stories (NightCafe, Zeet, Odyssey's McDonald's metaverse experience, Bit192 GPT-NeoX-20B).Source
• Your websiteLaunches this week: Physical AI Launchables (ready-to-run Templates for Isaac Sim, CARLA, MuJoCo, AlpaSim and Cosmos, with agent skills) and the Zero Trust AI Framework; Fully Connected 2026 sessions are on the site.Source

Add a "Who runs agents on CoreWeave" strip near the top with five logos and one-line outcomes, each linking to a case study or the talk summary from idea 1, plus a short quote. Repeat the strip on the inference and RL pages.

Agent builders choose infrastructure from peers; named companies doing the same work answer "who else does this here?" in seconds.

• Logo permission delays

  Launch with the customers already shown elsewhere on the site.

• #1 Fill the 75 empty transcript tabs, starting with the customer agent talks, and turn them into written agent proof

• Your websiteAgentic AI solution page: H1 "Agentic AI that paves the path to superintelligence", sections on training, inference and observing agents, "Exemplary agent inference. Validated by NVIDIA." and 11 FAQs, but no customer story or named agent builder.Source
• Your websiteVideo library (all 115 /resources/videos pages checked October 21, 2026): every page has a Transcript tab, but on 75 of them it is empty. Empty ones include customer talks such as "How Grammarly Scales Real-Time Inference with vLLM on CoreWeave", "How MasterClass Builds Multi-Agent AI That Learners Can Trust", "Training Cognition SWE-2: Asynchronous RL at Global Scale", "Building the Document Context Layer for AI Agents at LlamaIndex", "Beyond the Benchmark: How Parallel Web Systems Builds Reliable Web Agents", "Building the Voice OS at Willow", "Inside Pinterest's Machine Learning Platform", Nasdaq, IBM Granite 4.0 and You.com's Nemotron agentic search; the 40 filled ones include product demos such as the ARENA demo.Source
• Your websiteCustomer stories: the sitemap lists 17 written case studies. The hub features Cohere ("North, Agentic AI for Enterprise"), OpenAI, IBM ("80%") and Mistral AI ("2.5x Faster Training"); Cohere is the only one about agents, and none is about robotics or autonomous vehicles. Several are older stories (NightCafe, Zeet, Odyssey's McDonald's metaverse experience, Bit192 GPT-NeoX-20B).Source
• HiringDirector of Product Marketing, Agentic AI (open, posted June 17, 2026): position CoreWeave "as the premier infrastructure platform for agentic AI"; build "use-case playbooks" for "autonomous coding agents, multi-agent reasoning pipelines"; and "Build and scale a customer proof program featuring case studies and reference customers across AI labs, enterprises, and high-growth AI startups deploying agents in production"; plus thought leadership "white papers, technical blogs, benchmark reports, webinars".Source

Correct the mobile label, unpublish /topics/test and remove it from the sitemap, redirect /pricing/classic to /pricing (or mark it clearly as legacy with a link), and redirect coreweave.com to www.coreweave.com. Add a monthly check for empty titles and test pages.

Engineers comparing GPU prices read labels literally, and a test page or two pricing pages in search results looks careless for a company selling reliability.

• Old links to classic pricing

  Redirect so they still land on current prices.

• Your websitePricing page: on-demand NVIDIA GB200 NVL72 at $42.00 an hour, HGX B200 at $68.80 (spot $34.11), RTX PRO 6000 at $20.00; GB300 NVL72 and HGX B300 on-demand are "Contact sales". The table header says "Inference Single GPU Price", but the 12 mobile cards label the same figure "Inference Single CPU Price".Source
• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source

Render the H1 and hero copy in the first HTML response without waiting for animation, preload the headline font with a fallback, reserve space for the announcement bar and logo strips, and lazy-load the mega-menu and below-fold media. Re-test the homepage, pricing and product pages.

A faster first view keeps more paid and event traffic, and the improvement already under way (24% faster LCP) shows it is fixable.

• Design team wants the animation

  Keep it, but start it after the text has painted.

• Google PageSpeedHomepage on mobile: performance 29 (desktop 33), lab Largest Contentful Paint 13.9 s with the H1 text as the LCP element. Real Chrome visitors (75th percentile, whole site): LCP 2.76 s and layout shift 0.24, both above Google's "good" limits (2.5 s, 0.1); interaction 95 ms is good. LCP has improved 24% over the period.Source
• Your websiteHomepage: "The essential cloud for AI"; "The only AI cloud to earn SemiAnalysis' Platinum ClusterMAX rating three times", "9 of 10 leading foundation model providers rely on CoreWeave", "First NVIDIA Exemplar Cloud for training on NVIDIA GB200 NVL72", "Up to 98% effective training time", "Minutes to a human solving your issue". Forge is "free to build with any cloud, model, or framework". Menu programs: ARENA, Direct-to-Expert, Ventures, Academic Research, Zero Egress Migration.Source

Write 60 buyer questions across the four use cases and three personas (research lead, platform engineer, CTO), run them monthly in the main assistants, and record which providers are named and which pages are cited. Report share of voice by use case and the pages that win, and feed gaps into ideas 1, 2 and 12.

A baseline turns "AI visibility" into a number the team can move, and the cited pages show exactly what content to build.

• Answers vary run to run

  Run each question several times and use averages.

• HiringSenior Manager, Marketing Research & Intelligence (posted September 10, 2026): "Develop research and measurement frameworks for new / emerging audience and platform discovery channels, including Generative AI / LLMs"; size TAM and SAM; "Track market share against key competitors"; identify buyer groups by use case; and "push account targeting signals into marketing automation and CRM platforms"; success includes "identifying accounts and prospects that are ready to buy".Source
• Competitor researchRivals' own homepages: Lambda, "Supercomputers for training and inference"; Nebius, "The Ultimate AI Cloud"; Crusoe, "The AI factory company"; Together AI, "the AI Native Cloud"; Runpod, "The AI Developer Cloud". Hyperscalers (AWS, Google Cloud, Azure) sell GPU capacity alongside everything else.See the research Source
• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source

Publish a decision guide: which scheduler pattern fits which team (research cluster, mixed training and inference, multi-cloud), what each costs to operate, and where SUNK's published numbers (96% goodput, 97-98% ETTR, 10x MTTF) come from. Add an honest "when you do not need SUNK" section and a path to an ARENA test.

Platform engineers trust guides that admit trade-offs, and this is the page sellers can send in training deals.

• Open-source communities see it as marketing

  Have engineers write the open-source sections and link to the projects.

• HiringSenior Product Marketing Manager, AI Training Infrastructure (posted August 25, 2026): owns the narrative tying "CoreWeave Kubernetes Service, SUNK, and SUNK Anywhere into a coherent market story", the competitive framework "against alternative infrastructure approaches", where "Ray, SkyPilot, and Kueue fit", how the Weights & Biases integration "shapes researcher perception", and a customer evidence strategy that is "not a library of case studies"; measured on "deal velocity, contract value, competitive win rates".Source
• Your websiteSUNK page proof points: "up to 96% training goodput, 97-98% effective training time (ETTR), 10x longer mean time to failure (MTTF), and demonstrated >50% model FLOPs utilization (MFU)". The blog celebrates "One Billion Runs" on Weights & Biases.Source
• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source
• Your websiteARENA: "Prove production readiness", a pass with "14 days free" and "Expanded inference credits"; founder stories from Axiom Math, Rime and Anam; "Not ready for ARENA? Start building free".Source

Add an opt-in "Run this on CoreWeave" panel in W&B for long runs: estimated time and cost on comparable CoreWeave GPUs using public prices, plus an ARENA pass offer. Keep it informational, respect workspace privacy settings, and never use run data for sales without consent.

It reaches researchers at the exact moment a run is slow or fails, inside a tool they already trust.

• Users see it as an upsell

  Make it opt-in and useful even if they never switch.

• Your websiteSUNK page proof points: "up to 96% training goodput, 97-98% effective training time (ETTR), 10x longer mean time to failure (MTTF), and demonstrated >50% model FLOPs utilization (MFU)". The blog celebrates "One Billion Runs" on Weights & Biases.Source
• HiringSenior Product Marketing Manager, AI Training Infrastructure (posted August 25, 2026): owns the narrative tying "CoreWeave Kubernetes Service, SUNK, and SUNK Anywhere into a coherent market story", the competitive framework "against alternative infrastructure approaches", where "Ray, SkyPilot, and Kueue fit", how the Weights & Biases integration "shapes researcher perception", and a customer evidence strategy that is "not a library of case studies"; measured on "deal velocity, contract value, competitive win rates".Source
• HiringStaff Product Manager, AI Inference (posted October 12, 2026): own the inference roadmap and go-to-market "including pricing strategies, positioning, messaging, and competitive differentiation", with inference engines "vllm, SGLang, TensorRT-llm, and llm-d". A Senior Product Manager, W&B Models (posted August 28) works on the experiment workflow researchers "live in" daily, as "autoresearch and agentic workflows are reshaping" it.Source
• Your websitePricing page: on-demand NVIDIA GB200 NVL72 at $42.00 an hour, HGX B200 at $68.80 (spot $34.11), RTX PRO 6000 at $20.00; GB300 NVL72 and HGX B300 on-demand are "Contact sales". The table header says "Inference Single GPU Price", but the 12 mobile cards label the same figure "Inference Single CPU Price".Source
• Your websiteARENA: "Prove production readiness", a pass with "14 days free" and "Expanded inference credits"; founder stories from Axiom Math, Rime and Anam; "Not ready for ARENA? Start building free".Source

Map the path as one funnel: Forge signup, first logged run or inference call, ARENA pass, production contract. Publish a "first hour" tutorial per use case, add in-product prompts at each step, and report conversion between steps every week.

Self-serve builders become the startup and enterprise pipeline when each step leads clearly to the next.

• Sales and self-serve compete

  Agree hand-off rules by account size.

• HiringDirector, Developer Relations (open since March 11, 2026, updated October 12): grow a community around "CoreWeave SaaS offerings, including tools from Weights & Biases, OpenPipe, and Marimo"; produce "onboarding guides, tutorials, demos, reference architectures" that "accelerate time-to-first-value"; "Own and continuously improve the developer onboarding experience"; track "onboarding conversion, product adoption, engagement, and retention".Source
• Your websiteForge pricing: Free at $0 a month, Teams "Starts at $60/month" with a 30-day free trial, Enterprise, and "Free forever for academic research".Source
• Your websiteARENA: "Prove production readiness", a pass with "14 days free" and "Expanded inference credits"; founder stories from Axiom Math, Rime and Anam; "Not ready for ARENA? Start building free".Source
• Your websiteHomepage: "The essential cloud for AI"; "The only AI cloud to earn SemiAnalysis' Platinum ClusterMAX rating three times", "9 of 10 leading foundation model providers rely on CoreWeave", "First NVIDIA Exemplar Cloud for training on NVIDIA GB200 NVL72", "Up to 98% effective training time", "Minutes to a human solving your issue". Forge is "free to build with any cloud, model, or framework". Menu programs: ARENA, Direct-to-Expert, Ventures, Academic Research, Zero Egress Migration.Source

Build a page per top partner (NVIDIA, VAST Data, CrowdStrike, top agent frameworks and model providers): the joint architecture, a co-branded customer story, a joint webinar and a partner-specific offer with a tracked link. Give partners the link for their own sellers.

Partner pages turn partner traffic into measurable pipeline and give both sales teams one shared asset.

• Partners slow to approve

  Start with partners already featured on the site.

• HiringDirector, Partner Marketing (posted July 31, 2026): joint go-to-market with "NVIDIA, leading model and agent framework providers, and Weights & Biases"; "better together" narratives; "co-branded case studies"; "Track and report on partner-sourced and partner-influenced pipeline"; lead partner presence at Fully Connected, NVIDIA GTC, NeurIPS and AI Engineer World's Fair.Source
• Your websiteCustomer stories: the sitemap lists 17 written case studies. The hub features Cohere ("North, Agentic AI for Enterprise"), OpenAI, IBM ("80%") and Mistral AI ("2.5x Faster Training"); Cohere is the only one about agents, and none is about robotics or autonomous vehicles. Several are older stories (NightCafe, Zeet, Odyssey's McDonald's metaverse experience, Bit192 GPT-NeoX-20B).Source
• Your websiteHomepage: "The essential cloud for AI"; "The only AI cloud to earn SemiAnalysis' Platinum ClusterMAX rating three times", "9 of 10 leading foundation model providers rely on CoreWeave", "First NVIDIA Exemplar Cloud for training on NVIDIA GB200 NVL72", "Up to 98% effective training time", "Minutes to a human solving your issue". Forge is "free to build with any cloud, model, or framework". Menu programs: ARENA, Direct-to-Expert, Ventures, Academic Research, Zero Egress Migration.Source

Publish a written recap with key numbers for every session, group them into tracks (training, inference, agents, physical AI), and send attendees and registrants a track-based follow-up with the related tutorial and an ARENA invitation.

Recaps keep the content earning search traffic, and track-based follow-up turns attendance into evaluations.

• Recaps misstate speakers

  Send recaps to speakers before publishing.

• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source
• HiringDirector, Partner Marketing (posted July 31, 2026): joint go-to-market with "NVIDIA, leading model and agent framework providers, and Weights & Biases"; "better together" narratives; "co-branded case studies"; "Track and report on partner-sourced and partner-influenced pipeline"; lead partner presence at Fully Connected, NVIDIA GTC, NeurIPS and AI Engineer World's Fair.Source
• Your websiteARENA: "Prove production readiness", a pass with "14 days free" and "Expanded inference credits"; founder stories from Axiom Math, Rime and Anam; "Not ready for ARENA? Start building free".Source

Use the question set from idea 6 to choose 40 new explainers, write each with a clear answer in the first paragraph, a diagram and a link to the matching product, and refresh the featured list.

Explainers are what assistants and search engines quote, and they are cheap to produce from existing engineering knowledge.

• Thin content

  One clear, expert answer per page.

• #6 Measure what AI assistants recommend for "best GPU cloud" questions, by use case

• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source
• HiringSenior Manager, Marketing Research & Intelligence (posted September 10, 2026): "Develop research and measurement frameworks for new / emerging audience and platform discovery channels, including Generative AI / LLMs"; size TAM and SAM; "Track market share against key competitors"; identify buyer groups by use case; and "push account targeting signals into marketing automation and CRM platforms"; success includes "identifying accounts and prospects that are ready to buy".Source

Build a calculator: pick a model, traffic and latency target, and see cost per million tokens on serverless, dedicated and reserved options, with the benchmark source for each number. Offer to save the result and send it to sales.

Inference buyers compare on cost per token; showing the math wins the shortlist and creates qualified leads.

• Benchmarks age quickly

  Date every figure and update monthly.

• HiringStaff Product Manager, AI Inference (posted October 12, 2026): own the inference roadmap and go-to-market "including pricing strategies, positioning, messaging, and competitive differentiation", with inference engines "vllm, SGLang, TensorRT-llm, and llm-d". A Senior Product Manager, W&B Models (posted August 28) works on the experiment workflow researchers "live in" daily, as "autoresearch and agentic workflows are reshaping" it.Source
• Your websitePricing page: on-demand NVIDIA GB200 NVL72 at $42.00 an hour, HGX B200 at $68.80 (spot $34.11), RTX PRO 6000 at $20.00; GB300 NVL72 and HGX B300 on-demand are "Contact sales". The table header says "Inference Single GPU Price", but the 12 mobile cards label the same figure "Inference Single CPU Price".Source
• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source

Join these signals by company, score each account weekly (team size, usage growth, use case, funding), and route the top accounts to sales with the reason. Use W&B data only where terms and consent allow.

First-party intent is more accurate than bought intent data, and the reasons give sellers a relevant first message.

• Privacy concerns

  Use only consented, company-level signals.

• HiringSenior Manager, Marketing Research & Intelligence (posted September 10, 2026): "Develop research and measurement frameworks for new / emerging audience and platform discovery channels, including Generative AI / LLMs"; size TAM and SAM; "Track market share against key competitors"; identify buyer groups by use case; and "push account targeting signals into marketing automation and CRM platforms"; success includes "identifying accounts and prospects that are ready to buy".Source
• Your websiteForge pricing: Free at $0 a month, Teams "Starts at $60/month" with a 30-day free trial, Enterprise, and "Free forever for academic research".Source
• Your websiteARENA: "Prove production readiness", a pass with "14 days free" and "Expanded inference credits"; founder stories from Axiom Math, Rime and Anam; "Not ready for ARENA? Start building free".Source
• Your websiteSUNK page proof points: "up to 96% training goodput, 97-98% effective training time (ETTR), 10x longer mean time to failure (MTTF), and demonstrated >50% model FLOPs utilization (MFU)". The blog celebrates "One Billion Runs" on Weights & Biases.Source

Publish a page per Template (what it runs, GPU used, time to first result, sample output), co-launch each with its tool's community or vendor, and add a "run this Template" link to partner docs.

Robotics engineers search by tool name; Template pages meet them there and give partners something concrete to share.

• Open-source projects decline vendor links

  Contribute examples to their repos instead.

• Your websiteLaunches this week: Physical AI Launchables (ready-to-run Templates for Isaac Sim, CARLA, MuJoCo, AlpaSim and Cosmos, with agent skills) and the Zero Trust AI Framework; Fully Connected 2026 sessions are on the site.Source
• HiringPhysical AI hiring: a Senior Manager, Physical AI Communications, the "first dedicated communications owner for CoreWeave's Physical AI business", to "Create a repeatable pipeline of customer and partner stories across robotics, autonomous systems, industrial applications, and scientific AI" and measure "share of voice"; and a Business Development Manager - Physical AI to own partnerships with "simulation ISVs, robotics platform companies, OEMs" and the NVIDIA Physical AI ecosystem (Isaac Sim, Isaac Lab, Cosmos, Omniverse, DRIVE), with "joint reference architectures".Source
• Your websitePhysical AI page: five customer cards (Decart, NEURA Robotics, PhysicsX, Woven by Toyota, Nissan "reduces physical vehicle tests by 17%"). Decart, NEURA and Toyota link to customer pages on wandb.ai, Nissan to a news post; none links to a CoreWeave case study.Source

Run a 15-minute monthly form for solutions architects: workload, before and after numbers, quote, and whether the customer might go public. Turn the best into anonymized proof points within a week and named stories when approved.

Field engineers see the results first; a light routine turns them into assets while they are fresh.

• Busy SAs skip it

  Keep it to five minutes and share how their input was used.

• HiringSenior Product Marketing Manager, AI Training Infrastructure (posted August 25, 2026): owns the narrative tying "CoreWeave Kubernetes Service, SUNK, and SUNK Anywhere into a coherent market story", the competitive framework "against alternative infrastructure approaches", where "Ray, SkyPilot, and Kueue fit", how the Weights & Biases integration "shapes researcher perception", and a customer evidence strategy that is "not a library of case studies"; measured on "deal velocity, contract value, competitive win rates".Source
• Your websiteCustomer stories: the sitemap lists 17 written case studies. The hub features Cohere ("North, Agentic AI for Enterprise"), OpenAI, IBM ("80%") and Mistral AI ("2.5x Faster Training"); Cohere is the only one about agents, and none is about robotics or autonomous vehicles. Several are older stories (NightCafe, Zeet, Odyssey's McDonald's metaverse experience, Bit192 GPT-NeoX-20B).Source
• HiringDirector of Product Marketing, Agentic AI (open, posted June 17, 2026): position CoreWeave "as the premier infrastructure platform for agentic AI"; build "use-case playbooks" for "autonomous coding agents, multi-agent reasoning pipelines"; and "Build and scale a customer proof program featuring case studies and reference customers across AI labs, enterprises, and high-growth AI startups deploying agents in production"; plus thought leadership "white papers, technical blogs, benchmark reports, webinars".Source

Publish a yearly, aggregated report: failure causes, time lost to restarts, goodput by cluster size and what improved it, with the method explained. Launch it at Fully Connected and NeurIPS.

Original data earns press, analyst citations and assistant answers, and makes reliability CoreWeave's category.

• Customer confidentiality

  Aggregate and suppress small groups.

• Your websiteSUNK page proof points: "up to 96% training goodput, 97-98% effective training time (ETTR), 10x longer mean time to failure (MTTF), and demonstrated >50% model FLOPs utilization (MFU)". The blog celebrates "One Billion Runs" on Weights & Biases.Source
• Your websiteHomepage: "The essential cloud for AI"; "The only AI cloud to earn SemiAnalysis' Platinum ClusterMAX rating three times", "9 of 10 leading foundation model providers rely on CoreWeave", "First NVIDIA Exemplar Cloud for training on NVIDIA GB200 NVL72", "Up to 98% effective training time", "Minutes to a human solving your issue". Forge is "free to build with any cloud, model, or framework". Menu programs: ARENA, Direct-to-Expert, Ventures, Academic Research, Zero Egress Migration.Source
• HiringSenior Product Marketing Manager, AI Training Infrastructure (posted August 25, 2026): owns the narrative tying "CoreWeave Kubernetes Service, SUNK, and SUNK Anywhere into a coherent market story", the competitive framework "against alternative infrastructure approaches", where "Ray, SkyPilot, and Kueue fit", how the Weights & Biases integration "shapes researcher perception", and a customer evidence strategy that is "not a library of case studies"; measured on "deal velocity, contract value, competitive win rates".Source
• HiringSenior Manager, Marketing Research & Intelligence (posted September 10, 2026): "Develop research and measurement frameworks for new / emerging audience and platform discovery channels, including Generative AI / LLMs"; size TAM and SAM; "Track market share against key competitors"; identify buyer groups by use case; and "push account targeting signals into marketing automation and CRM platforms"; success includes "identifying accounts and prospects that are ready to buy".Source

Release the harness (workloads, metrics, reporting) as open source, invite other clouds and the community to run it, and publish CoreWeave's own results. Run a hackathon around it.

Owning the standard method for comparing AI clouds puts CoreWeave in every evaluation, and its published results already lead.

• Rivals tune for the benchmark

  Use real workloads and publish the method.

• Your websiteARENA: "Prove production readiness", a pass with "14 days free" and "Expanded inference credits"; founder stories from Axiom Math, Rime and Anam; "Not ready for ARENA? Start building free".Source
• Competitor researchRivals' own homepages: Lambda, "Supercomputers for training and inference"; Nebius, "The Ultimate AI Cloud"; Crusoe, "The AI factory company"; Together AI, "the AI Native Cloud"; Runpod, "The AI Developer Cloud". Hyperscalers (AWS, Google Cloud, Azure) sell GPU capacity alongside everything else.See the research Source
• Your websiteBuilder page: CoreWeave Hacks hackathons, VAST Builders Challenges in San Francisco, New York and London, build notes (ARIA, Agent Reinforcement Trainer), Discord, X and LinkedIn. npm: @coreweave/cwsandbox has about 1,082 weekly downloads; Forge Agent Lens packages exist for Codex, Claude Code and OpenClaw.Source
• Your websiteHomepage: "The essential cloud for AI"; "The only AI cloud to earn SemiAnalysis' Platinum ClusterMAX rating three times", "9 of 10 leading foundation model providers rely on CoreWeave", "First NVIDIA Exemplar Cloud for training on NVIDIA GB200 NVL72", "Up to 98% effective training time", "Minutes to a human solving your issue". Forge is "free to build with any cloud, model, or framework". Menu programs: ARENA, Direct-to-Expert, Ventures, Academic Research, Zero Egress Migration.Source

Publish three reference architectures (coding agent, research, voice agent) built with named partners, with code, cost and latency numbers, sandboxes and RL post-training. Co-launch with each partner.

Builders adopt working examples; partners promote what makes their framework look good.

• Frameworks change fast

  Maintain with partners and version each architecture.

• HiringDirector of Product Marketing, Agentic AI (open, posted June 17, 2026): position CoreWeave "as the premier infrastructure platform for agentic AI"; build "use-case playbooks" for "autonomous coding agents, multi-agent reasoning pipelines"; and "Build and scale a customer proof program featuring case studies and reference customers across AI labs, enterprises, and high-growth AI startups deploying agents in production"; plus thought leadership "white papers, technical blogs, benchmark reports, webinars".Source
• HiringDirector, Partner Marketing (posted July 31, 2026): joint go-to-market with "NVIDIA, leading model and agent framework providers, and Weights & Biases"; "better together" narratives; "co-branded case studies"; "Track and report on partner-sourced and partner-influenced pipeline"; lead partner presence at Fully Connected, NVIDIA GTC, NeurIPS and AI Engineer World's Fair.Source
• Your websiteAgentic AI solution page: H1 "Agentic AI that paves the path to superintelligence", sections on training, inference and observing agents, "Exemplary agent inference. Validated by NVIDIA." and 11 FAQs, but no customer story or named agent builder.Source
• Your websiteBuilder page: CoreWeave Hacks hackathons, VAST Builders Challenges in San Francisco, New York and London, build notes (ARIA, Agent Reinforcement Trainer), Discord, X and LinkedIn. npm: @coreweave/cwsandbox has about 1,082 weekly downloads; Forge Agent Lens packages exist for Codex, Claude Code and OpenClaw.Source

Add llms.txt to www.coreweave.com pointing to pricing, product pages, case studies and topic explainers, and document how a coding agent can sign up, check prices and launch a Template.

Engineers increasingly ask coding agents to compare and provision clouds; a clear agent path makes CoreWeave the easy answer.

• Little immediate traffic

  Treat it as a low-cost bet tracked monthly.

• Your websiteSitemap (892 URLs): 244 blog posts, 190 resources, 68 Fully Connected sessions, 59 events, 23 topic explainers ("What is Slurm?", "What is vLLM?") plus /topics/test, a live page with the H1 "Test" and an empty title and description; /pricing/classic is still live next to /pricing; an "AI cloud comparison guide" ebook and an "AI cloud TCO pricing comparison" report. coreweave.com answers without redirecting to the canonical www host. The docs site publishes llms.txt and an MCP endpoint; www.coreweave.com/llms.txt is a 404.Source
• Your websiteBuilder page: CoreWeave Hacks hackathons, VAST Builders Challenges in San Francisco, New York and London, build notes (ARIA, Agent Reinforcement Trainer), Discord, X and LinkedIn. npm: @coreweave/cwsandbox has about 1,082 weekly downloads; Forge Agent Lens packages exist for Codex, Claude Code and OpenClaw.Source
• Your websiteLaunches this week: Physical AI Launchables (ready-to-run Templates for Isaac Sim, CARLA, MuJoCo, AlpaSim and Cosmos, with agent skills) and the Zero Trust AI Framework; Fully Connected 2026 sessions are on the site.Source

Collect papers that acknowledge CoreWeave or W&B, list them by field with links, and invite authors to short talks. Give program members a ready acknowledgement line.

Researchers trust other researchers, and today's students run tomorrow's labs.

• Few papers found

  Start with W&B citations, which are common.

• Your websiteForge pricing: Free at $0 a month, Teams "Starts at $60/month" with a 30-day free trial, Enterprise, and "Free forever for academic research".Source
• Your websiteHomepage: "The essential cloud for AI"; "The only AI cloud to earn SemiAnalysis' Platinum ClusterMAX rating three times", "9 of 10 leading foundation model providers rely on CoreWeave", "First NVIDIA Exemplar Cloud for training on NVIDIA GB200 NVL72", "Up to 98% effective training time", "Minutes to a human solving your issue". Forge is "free to build with any cloud, model, or framework". Menu programs: ARENA, Direct-to-Expert, Ventures, Academic Research, Zero Egress Migration.Source