NVIDIA DevRel flags 40% AI failures

NVIDIA and NetApp are pushing a blunt message to enterprise buyers: most artificial intelligence work breaks before it reaches production because the data are not ready. (nvidia.com) In NVIDIA’s GTC 2026 session “Turn Your Data Estate Into AI Fuel,” presented with NetApp, the companies said enterprises often hold 100 exabytes or more of unstructured data. The session’s central claim was starker: 95% of AI pilots fail to reach production because those data are not “AI-ready.” (nvidia.com) “AI-ready” in this pitch means company files, emails, PDFs, videos and other records can be found, governed, secured and fed into models without copying them into new silos. NVIDIA’s AI data platform materials say unstructured data makes up 70% to 90% of organizational data and creates governance problems because it is large, varied and messy. (nvidia.com) That is why the argument has shifted from model quality alone to the plumbing around models. NVIDIA now describes the modern data center as an “AI factory,” a system built to move data, train or fine-tune models, and run inference at scale rather than just store information. (nvidia.com 1) (nvidia.com 2) Inference is the part users actually touch: the chatbot answer, the coding suggestion, the search result. NVIDIA’s NIM product is built around that layer, offering prebuilt inference microservices that the company says can run on NVIDIA-accelerated infrastructure in the cloud, data center, workstation or at the edge. (nvidia.com 1) (nvidia.com 2) For developers, NIM is meant to package a model like a containerized service instead of a custom integration project. NVIDIA says Developer Program members can use downloadable NIM microservices for development, testing and research, including support for models such as Meta’s Llama 3.1 8B and Mistral 7B Instruct. (developer.nvidia.com) NetApp’s role in the story is the storage and data-management layer underneath that software. Its March 16, 2026 announcement said the NetApp AI Data Engine was launching to address “complex data challenges,” with NVIDIA listed as an early access customer and partner. (netapp.com) NetApp and NVIDIA describe the combination as an “AI-ready data platform” or “AI-ready data pipeline,” built to avoid duplicated storage, reduce data movement and keep governance controls in place. NetApp’s own materials tie that setup to retrieval-augmented generation, agentic artificial intelligence and production inference workloads. (netapp.com 1) (netapp.com 2) NVIDIA’s broader 2026 State of AI report points in the same direction from a different angle. The company said 38% of respondents cited a lack of artificial intelligence experts and data scientists to implement data and scale projects from pilot to production. (nvidia.com) So the headline is less about one viral “40% failure” soundbite than a bigger sales and engineering push from NVIDIA and NetApp. Their case is that enterprise AI adoption now depends on data preparation, governance and inference operations as much as on the model itself. (nvidia.com) (nvidia.com)