SAP, Nvidia partner to boost Business AI development
SAP and Nvidia are working together to combine platforms and services that help customers build business-specific generative AI capabilities across SAP's cloud products.
SAP and Nvidia are teaming up with integrations designed to enable customers to build and deploy generative AI capabilities across SAP's cloud products.
The partnership is intended to boost the adoption of SAP's Business AI, which focuses on delivering business-specific AI capabilities such as SAP's Joule AI assistant through its cloud applications.
The new integrations were unveiled at this week's Nvidia GTC 2024 and are expected to be available by the end of the year.
SAP and Nvidia will integrate generative AI (GenAI) into SAP's cloud applications portfolio, including SAP Datasphere for data management, SAP Business Technology Platform (BTP) for application development and Rise with SAP for managed cloud services, according to the companies. SAP will use Nvidia's GenAI foundry service to fine-tune large language models (LLMs) for domain-specific scenarios and deploy applications using the Nvidia Inference Microservice platform, which was also unveiled this week.
SAP's partner strategy now rests on the idea that it doesn't make sense for SAP to build certain technologies that already exist, according to Philipp Herzig, SAP's newly installed chief AI officer.
Because AI technology is evolving so rapidly, SAP needs to focus on enabling a flexible architecture to install or change new components, Herzig said in a TechTarget Editorial interview prior to GTC.
"Overall, it's a very open strategy, where we select the best model for the task at hand that we want to use," he said.
The Nvidia partnership integrates SAP GenAI LLMs on various layers of the Nvidia stack, such as at the GPU layer or at the DGX cloud platform layer.
SAP is also expanding the partnership into Nvidia's software layer, including its use of Nvidia's Neural Modules (NeMo) framework for building, customizing and deploying GenAI models, Herzig said. NeMo includes a retrieval-augmented generation capability to securely access SAP and third-party data running in SAP applications.
"For example, we are using within Joule what's called retrieval-augmented generation, where there are certain parts of their NeMo frame for content moderation or retrieval, which helps with semantic search algorithms," he said. "We will store it to vectors on HANA Cloud Vector Engine, but if you then do retrieval, you need to do re-ranking to get a good embedding model."
Partnership could help with AI adoption
The growth of SAP's partnership with Nvidia is not surprising, as Nvidia has become the easiest choice for most enterprise vendors to build portable AI applications, said Holger Mueller, an analyst at Constellation Research.
"For now, this is all about SAP packaging AI for their packaged cloud apps, but if and when this will open up for SAP to allow its customers to build their in-house custom AI on BTP is still open," Mueller said. He expects SAP to make AI application development a focus at its Sapphire conference in June, which he said is a move the enterprise tech vendor needs to exploit to remain competitive.
"Oracle is ahead, but it's not really a fair competition as Oracle runs on Oracle Cloud and SAP needs to run on AWS, Google and Azure," Mueller said. "There are benefits and complexities to that, but [SAP is] for sure slower out of the gate."
That's something customers should factor into their thinking on SAP Business AI and how to best consume these innovations. Whether you can do effective AI on premises is a worthwhile discussion, but SAP's direction here is very clear.
Jon ReedCo-founder, Diginomica
Jon Reed, co-founder of Diginomica, an enterprise industry analysis firm, called the deepening partnership smart for SAP. Successful AI adoption requires both compute power, which Nvidia brings, and domain-specific data, which comes from SAP, he said.
However, Reed issued a word of caution, saying it's important to keep in mind that despite Nvidia's stratospheric stock valuation and SAP's surging stock price, the ROI of Business AI is still a work in progress.
"It's early days, and the overall cost of compute for GenAI needs to go down. That's something Nvidia and SAP will continue to work on," he said.
He also noted that the partnership focuses on SAP's cloud applications, not on-premises.
"That's something customers should factor into their thinking on SAP Business AI and how to best consume these innovations," Reed said. "Whether you can do effective AI on premises is a worthwhile discussion, but SAP's direction here is very clear."
SAP's cloud focus is helpful for roadmap planning, but doesn't close the door on on-premises AI options for customers, he said. Still, SAP will need to clarify why it has chosen this direction and what the tradeoffs are for customers.
"It's a complicated discussion. Cloud is helpful for AI services, but the key is the state of your data model and your data governance," Reed said. "This is early days, so we'll need to see how it plays out in terms of what approaches to AI work best."
Jim O'Donnell is a senior news writer for TechTarget Editorial who covers ERP and other enterprise applications.
