Nvidia has enjoyed a massive increase in revenues over the last couple of years as server and public cloud suppliers rush to address AI applications by coupling its GPUs into x86 servers. Seemingly IBM missed the boat – offloading its System x business to Lenovo in October 2013 – just at the point when GPU acceleration was moving beyond technical supercomputing to the x86-based enterprise server market. However this is far from the truth. In this post I’ll look at the on-going collaboration between the two companies and how it benefits their customers.
One month after it left the x86 server market IBM announced a collaborative plan with Nvidia to deliver Tesla GPU-accelerated versions of the software running on its Power Systems. In those days the theme was ‘Big Data and Analytics’ rather than AI. Arguably IBM was the AI supreme at the time, since – two-years earlier – it had demonstrated its capabilities by beating ex-champions Ken Jenkins and Brad Rutter in the TV-based Jeopardy! Game show, using a 10 TeraFLOPS Power-based server housed in 10 racks.
In 2013 the two companies announced their intention to couple Power8 processors to Tesla GPUs for ‘ he scientific and technical communities for computing tasks like space exploration, decoding the human genome and speeding new products to market’. Nvidia stated that the partnership would ‘bring supercomputer performance to the corporate data center, expanding the use of GPU accelerators well beyond the traditional supercomputing and technical computing markets’. For IBM it helped build on the establishment of the OpenPOWER consortium four months before, of which Google, Mellanox (acquired by Nvidia in April 2020) and Tyan (a motherboard manufacturer and part of MiTAC) were also members.
In March 2025 the two companies underlined new aspects of their collaboration. In particular:
- IBM has now announced the availability of Nvidia NVIDIA H200 Tensor Core GPUs on the IBM Cloud in addition to the H100 GPUs added in October 2024.The next version of IBM Storage Scale (to be announced this quarter) will include Content-Aware Storage (CAS) capabilities allowing enterprise users to extract meaning from their unstructured data for inferencing in order to expand and enhanced AI applications such as Retrieval-Augmented Generation (RAG) and AI reasoning. For this IBM Storage Scale will be integrated with Nvidia’s BlueField-3 DPUs and its Spectrum-X networking. Multi-modal document data extraction workflows will be able to use Nvidia’s NeMo Retriever microservices created through its Nvidia Inference Microservices (NIM) software.
- CAS will also be embedded in the next update of IBM Fusion planned for the second quarter of this year.
- IBM Consulting is adding AI Integration Services; specifically adding Nvidia Blueprints for industry-specific agentic AI applications such as autonomous inspection and maintenance for manufacturers and proactive video data analysis and anomaly response for energy companies. IBM Consulting will continue to help users build and manage compute-intensive AI workloads across cloud environments leveraging RedHat OpenShift and Nvidia offerings including AI Foundry, NeMo, AI Enterprise, Blueprints and Clara.
For some vendors the phrase ‘hybrid multi-cloud’ is nebulous and undefined; for some it’s just slide-ware. For IBM it’s a platform for helping its customers to modernize, optimize and run enterprise applications irrespective of where the data and/or processing resides. As with SAP it continues to enhance its partnership with Nvidia in order to stay relevant and grow its AI business in a typically idiosyncratic way. For Nvidia the partnership has been a vital component of its rise to becoming (however briefly) the most successful IT supplier of all.