SAN JOSE, Calif. — Nvidia has won the business of one of the largest cloud providers with powerful GPUs to run AI models and services. The company is now moving downstream with a broader toolset and army of partners focused on enterprise data centers.
This week, Nvidia GTC, the company’s annual developer conference, attracted thousands of data scientists and electrical and computer engineers looking to learn how to build, deploy, and manage AI-specific software. Technologists were joined by customers and partners who do business and power an industry that analysts say will transform business.
Today’s data centers, with CPUs powering servers running business software, must make room for the infrastructure specific to generative AI models. Deploying and running GenAI models requires a new toolset.
“General-purpose computing is exhausted,” Nvidia CEO Jensen Huang said in his opening keynote here this week. “We need another way to do computing.”
Huang announced version 5 of the company’s AI enterprise platform, featuring new technology that executives described as Nvidia Inference Microservices (NIM). The integrated software simplifies the process of creating and developing his GenAI applications, which leverage Nvidia’s CUDA parallel computing platform and the programming model of the company’s GPUs.
Analysts expect many companies to bring smaller language models in-house so they can be fine-tuned based on corporate data without moving sensitive information to the public cloud. Additionally, running models in your data center can be cheaper than in the cloud.
Nvidia partners target enterprises
Nvidia’s NIM helps by simplifying the process of regularly feeding real-world data to a trained model, a process known as inference, so it can create up-to-date responses. Robin Bordri, chief marketing officer at AI model training platform maker Weights and Biases, said having tools to automate processes related to models means traditional software engineers can replace hard-to-find AI experts. It says it means you can do the job.
Weights and Biases integrates its software with Nvidia’s inference engine, allowing developers to train and infer from a platform that supports 30 underlying models. Currently, Weights and Biases has 1,000 customers, many of whom are government agencies and life science organizations, Bordoli said.
“We’re helping the next customer, the business,” he said. “They’re not going to build a model from scratch, but they want to take an existing model and tweak it based on corporate data.”
Patrick McFadin, vice president of developer relations at DataStax, which provides vector databases for AI applications, said NVIDIA is running containers on Kubernetes, an open source container orchestration platform well-known to enterprises. We have built a NIM that works.
“What we quickly realized was that it was deployed using Kubernetes,” McFadin said. “The people who run the infrastructure at large enterprises are using Kubernetes, so they’re very well connected.”
Nvidia partner Dell Technologies offers a variety of PowerEdge servers featuring Nvidia’s AI Enterprise software and GPUs, H100 and L40S.
Varun Chhabra, senior vice president of infrastructure and communications marketing, said, “What we see in the majority of companies is taking an off-the-shelf model, whether it’s a large model or a small model, and building their own company. It’s about combining it with data.” At Dell.
We believe search augmentation generation (RAG) will become as important as inference within the enterprise. RAG is an architecture that incorporates information retrieval systems to protect private data.
“RAG is a big area of focus for us,” says Chhabra.
Chhabra says the biggest advantage of Nvidia NIM is that it packages many of the microservices needed for inference into a single container. “It’s done turnkey.”
AI software and the accelerated computing required to run it are transforming data centers, Chhabra says.
“We definitely feel like we’re at a tipping point,” he said. “We are nearing a complete rebuild of our data center.”
Nvidia GTC customers
At the conference, Nvidia customers got an early look at how they can work with GenAI. The companies include LinkedIn, global advertising company WPP, cosmetics maker L’Oréal, and German claims management software maker ControlExpert, GmbH.
WPP partnered with Nvidia to develop a content engine to create videos, 3D, and 2D images of clients’ products using Nvidia Omniverse Cloud and GenAI. The system also uses photos from Getty Images and content creation technology from Adobe.
WPP chief technology officer Stefan Pretorius said the quality of advertising art depends on the data available to the AI models that generate the work.
“We find that when we work with clients who have a very clear brand definition, a very precise way of describing the brand’s personality, tone, voice, etc., we get much better results than if it’s vague. ” he said during his presentation.
Essentially, WPP uses AI to mimic the “human content creation process,” Pretorius said. “But given the complexity, the scale, and the amount of data we need to work with, we can’t do something like this without AI.”
Pretorius believes that AI-powered voice communication with website visitors will eventually replace today’s content-driven approach.
“We believe that the future of content consumption will be primarily conversational,” he said.
L’Oréal is testing GenAI to generate images for storyboards and Nvidia Omniverse for 3D rendering of its packaged products. It also uses some AI models.
The company feeds its models thousands of branded images, as well as settings such as background colors, different types of lighting, and elegant Parisian sunsets. Ad creators can use natural language to let models create images to inspire marketing ideas for 37 global brands.
The system helps imagine scenarios such as futuristic beauty salons or advertisements using cosmic phenomena like the Red Nebula.
“There is a reinterpretation of data happening,” Asmita Dubey, L’Oréal’s chief digital and marketing officer, said of GenAI in an interview. “And it’s the speed [of creation]. It can run faster.
Over the past six months, L’Oréal has worked with Nvidia Omniverse and WPP to create custom 3D models of its products, allowing users to change backgrounds, colors, and shading without spending days in a studio with a photographer. . L’Oreal requires him only one studio session to photograph every angle of the product packaging.
The company believes this will save time and money, but it’s still in the early stages of using Omniverse.
Sabri Tauzin, LinkedIn’s vice president of engineering, said during a panel discussion at the conference that the company uses AI for language translation. This allows customer service representatives in Omaha, Nebraska to speak to customers in Spanish, French, or German even if they don’t know the language.
“This allows us to retain people who are really good at understanding our products and giving detailed answers to our customers,” Tozin said.
ControlExpert’s software allows insurance companies to have customers take photos of car damage after an accident and send the images to the vendor through a mobile app.
An AI model analyzes the image, assesses the damage, and returns an estimated repair cost and a list of approved body shops. According to the company, ConrolExpert’s customers include 90% of all insurance companies.
The company trained its model on data collected over 20 years. Sebastian Schoenen, the company’s director of innovation and technology, said the company processes 20 million claims a year.
ControlExpert constantly updates its models, as vehicle design changes occur regularly and repair prices fluctuate. Nevertheless, there are still a minority of cases that require human intervention.
“If we find that our model is incapable of doing something, we point the case to humans,” Schoenen says.
Antone Gonsalves is Editor-at-Large for TechTarget Editorial, reporting on industry trends important to enterprise technology buyers. He has been in technology journalism for his 25 years and is based in San Francisco.
