NVIDIA GTC 25: What To Expect From the Year’s Biggest Artificial Intelligence Event

Several key themes and announcements are expected to shape the discussions:

The new Blackwell architecture. NVIDIA’s latest chips, named for mathematician David Blackwell, are now in production and shipping to customers. Blackwell represents a leap forward in GPU design, the company says. During his March 18 keynote presentation, CEO Jensen Huang will surely tout the technical enhancements of Blackwell, including improvements in speed, memory and power efficiency.  

Project DIGITS. The company plans to sell a first-of-its kind personal AI supercomputer called DIGITS that anyone might purchase for a home office or that businesses of any size might find useful. “Normally a workstation with the same computing power would cost $30,000,” says Sana Gutierrez, senior manager for data and artificial intelligence at CDW. “Now you can have it in this different form factor, and NVIDIA has said, preliminarily, that it’s going to be like $3,000.” The company expects to begin shipping its DIGITS workstations in May.  

NVIDIA Inference Microservices. Better known as NIMs, NVIDIA Inference Microservices are prepackaged software kits that help organizations shorten development cycles. “It’s the Legos that you use to build the plane,” Gutierrez explains. “You still have to have some knowledge and capability to build what you need, but it makes it a lot easier. It’s not out of a box; there’s always going to be some development work, but the speed at which you can develop with NIMs is much quicker.”

DIG DEEPER: Unlock the value of generative artificial intelligence in your organization.

How Businesses Can Build Their Own AI Solutions

There’s no limit to what organizations can create using the NIM tools. Many businesses in the manufacturing, warehouse and logistics sectors have used NIMs to build digital twins, copies of factory or warehouse floors where companies can experiment with new processes or layouts.

CDW has used NIM tools to build several AI solutions, Gutierrez notes, including a virtual assistant, a product description generator, a corporate briefing manager that can speak to outsiders about the business and a customer service agent.

“We’ve built some really cool things using those blueprints, and it highlights our capability to build. But we also know that no two customer situations are the same,” she says, adding that CDW is available to help customers develop their own AI solutions.

Between Project DIGITS, NIM, and NVIDIA’s advancements in the area of robotics, Gutierrez says that GTC is an opportunity for the public to glimpse “really futuristic stuff,” and it’s worth considering how far technology has come in just a few years. At last year’s event, “I don’t think there was nearly enough oohing and aahing at the fact that in our day and age, there are technology companies that are truly defying what we thought was possible, and NVIDIA is one of those.”

Click the banner below to learn how organizations are deploying artificial intelligence.

What Is the Data Center of the Future?

As businesses begin AI development aggressively, they will have to plot out a data strategy, and it will need to answer questions about how and where they will store data associated with AI workloads: in the cloud, on-premises, or some of both. AI is complicating questions about data storage because of the volume of data it generates and the amount of computing power it requires. Organizations that have long planned to grow in the cloud, for example, may need to think carefully about the compatibility of that idea with big AI ambitions.

“You have to consider cost,” Gutierrez says. “Does it get too expensive, from a data egress and ingress perspective, to maintain the data in the cloud? Do you need your workload to be close to your data, which maybe means it’s on-prem?”

To be clear, cloud is an option: NVIDIA has its own cloud-based GPU offering and works closely with all of the major hyperscalers. But for companies that do have physical data centers, or those thinking of creating one, it’s important that they think about AI as they upgrade their hardware. In short, Gutierrez says, they must focus on building “the data center of the future.”

FIND OUT: Can artificial intelligence agents really ease workloads for large enterprises?

“As a business goes to refresh its storage, it needs to be thinking about the platforms that are AI-capable,” she says. “You might not be doing AI today, but how do you ensure that you have a high-performance storage platform that can support your AI workload needs in the future? That’s where some of our partners, like NetApp and Pure Storage, come in.”

She adds that tomorrow’s data center will look a lot like today’s. It “doesn’t physically change; it’s still compute, storage, networking, and power and cooling.” But in an AI world, “the dial on all those things slowly cranks up. The greater your compute requirements are, the more power you need, and the more power you need, the more cooling you need and the more storage you need.”

Keep this page bookmarked for continued coverage from the event and follow us on the social platform X at @BizTechMagazine and the official conference feed, @NVIDIAGTC and at #GTC25.