IBM Taps Nvidia GPUs For AI-Turbocharged Data Mart
March 23, 2026 Alex Woodie
Nvidia became the world’s most valuable company thanks to its powerful GPUs, which can be used to train massive AI models containing trillions of parameters. But those GPUs can also be used to accelerate traditional SQL analytics on structured data sitting in a data lake, as IBM’s customer Nestlé found out in a recent deployment.
Nestlé is the world’s largest food manufacturer, with more than 2,000 brands and operations spanning 186 countries. The Swiss company, which recorded revenues of CHF 89.5 billion ($113.5 billion at current exchange rates), operates 335 factories and employs about 271,000 people.
The company, which is a longtime user of SAP enterprise software and IBM infrastructure, uses a single order-to-cash data mart to coordinate every order, delivery, and invoice across its global operations. This mart provides a consolidated data foundation that provides Nestlé with a “single version of truth” across its farflung operations, according to IBM.
Nestlé is the world’s largest food manufacturer.
Updating the 44 tables in this data mart on its old CPU platform used to take about 15 minutes, and it could only be done several times a day. While the data was accurate, it sometimes was not as fresh as some users would have preferred.
After adopting a new platform based on Nvidia GPUs and the Presto SQL query engine running in IBM’s watsonx.data lake, Nestlé reports that the database update has been reduced to three minutes. The overall system costs 83 percent less than the old system, which hashes out to a 30x price-performance increase, according to IBM.
According to Sam Werner, the general manager of IBM Storage, the work with Nestlé was made possible as a result of collaboration between IBM and Nvidia at the software and storage levels. In Presto query engine, IBM has adopted Velox, which is an open-source, C++ database acceleration library from Meta. It uses vectorization to accelerate query processing in SQL engines like Presto and Apache Spark.
IBM also tweaked Presto to use cuDF, which is an open source DataFrame library from Nvidia that is part of the Cuda framework for programming software to run on Nvidia GPUs. By adopting cuDF, IBM is able to utilize the massive parallel performance of Nvidia GPUs to accelerate SQL processing on structured data, such as the 44 tables Nestlé maintains in its data mart.
Nestlé is using GPUs to accelerate its Presto SQL query engine.
“We’ve been leading the innovation around Velox, which gives you this acceleration,” Werner told IT Jungle at Nvidia’s GTC conference in San Jose last week. “Moore’s Law kind of hit its end on being able to improve database performance with CPUs. But now we’re finding that you can actually significantly improve your cost and performance by accelerating now on GPUs.”
While Nestlé’s use case doesn’t involve IBM i or Db2 for i, it does provide an example of how any of IBM’s enterprise customers could potentially improve query performance. IBM i customers are adopting IBM Storage Scale (formerly Spectrum Scale) storage arrays to store their data and potentially even to house data in a data lake environment.
There’s a lot of innovation occurring in these storage systems to enable not only new AI use cases – such as language models or computer vision models on unstructured data – but also to accelerate traditional SQL queries on structured data residing in databases or data lakes.
Werner explained how IBM holds a variety of technical advantages over competitors with its Storage Scale products, such as automatic vectorization of data, storage virtualization, multi-protocol support, and integrated support for tape drives. He talked about the FPGAs embedded into FlashCore Modules that can handle a range of data management tasks, as well as the new IBM Fusion program that combines servers, storage, networking, and even GPU nodes in a single hyperconverged rack running Red Hat OpenShift.
None of this is IBM i-specific, but it is IBM i-adjacent. IBM is outfitting its enterprise storage offering with all sorts of tricks and capabilities designed to help its customers – including IBM i shops – to leverage their data in new ways, including for AI.
“We have trust from our clients right to go in and build AI with us. And IBM’s proven internally and with our clients that we know how to do AI,” he said. “We have the leadership capabilities. It’s about making people aware. Once they know and we have a conversation, we show them our capabilities.”
