Nvidia’s Agentic AI Agenda At CES Was Smart, But It Could Go Further

The surprise was not that Nvidia got on the agentic bandwagon at CES. The surprise to me was how well it was done.

Don’t get me wrong: Nvidia is a great company loaded with smart people. But so far, agentic thought leadership has been mostly coming from software or cloud-first companies, so it was interesting to see it coming from a chip company instead. And before someone yells that Nvidia is a software company, too: I know it is, but it’s still GPUs that are driving its business results.

So it was a pleasant surprise that the company’s agentic announcements at CES 2025 were quite solid relative to what we have seen from other leading companies over the past few months. The word I keep coming back to is smart.

What Nvidia Got Right About Agentic AI

So let's talk about what makes Nvidia’s approach here stand out.

1. It’s a visionary announcement — I think that the desire for tech vendors to demonstrate ROI after a couple of years of major AI investment is manifesting in very tactical roadmaps and marketing moves. Often, it’s just the same use cases over and over, along with a slick demo. By contrast, Nvidia did a nice job of mixing ordinary blueprints for functions like customer service with something truly visionary like robotic fleet management via its Mega Omniverse blueprint. (For more on the implications of Nvidia’s work in physical AI, see my colleague Bill Curtis’s article from last week.) It’s also notable that we saw mentions of some of the more advanced agentic development techniques such as agent orchestration and collaboration. All in all, Nvidia’s approach is well balanced between now and five years from now.
2. It’s well targeted — Agents are still new, and while we see a number of companies targeting agents at no-code/low-code developers or business users, Nvidia is taking a different approach with its agentic blueprints. These are not products per se, but more of a toolkit. That said, these blueprints are well-thought-out recipes to help professional devs and DevOps people build enterprise-grade solutions. The specs and deployment plans are well detailed and appropriate for the target users to get started. Overall, it’s just well done and respects the reality that this is new technology that goes beyond augmenting personal productivity.
3. It’s partner-driven — My prediction is that the enterprise AI war will be won with the best ecosystem and not the best model. I think we are starting to see a replay of what we saw with early mobile technology pre-iPhone, where big phone companies built closed platforms. In the mobile space, that mindset was disrupted when Apple blew up the status quo with the App Store and the iPhone. An AI-specific example of keeping things open is how AWS Bedrock has supported a wide range of large language models instead of just the ones it created. In the case of agentic AI, Nvidia did a great job at CES of highlighting both AI startups that are delivering blueprint content as well as its partner Accenture for helping design and deliver the blueprints. As much as I liked it, however, I also see some opportunities to improve on it (see below).
4. Customers want to run AI anywhere — I also believe that the foundation for agentic AI in the enterprise will be in AI development platforms such as AWS Bedock or Azure AI Foundry. But as good as these cloud based frameworks are, they’re still built around the CSPs’ own clouds. By contrast, Nvidia specs its blueprints so customers can deploy them on-premises or in the cloud, and the company provides enough information for the customer to decide which way to go.

Where Nvidia Could Go From Here

Long story short, there’s a lot of good in the company’s approach to agentic AI. But I do see some ways that Nvidia can take it further. First, what Nvidia has announced so far is visionary but not extraordinary. Its picture of the future is definitely well done, but if you look at the most well-baked features available right now, similar capabilities can be found in other platforms such as Bedrock. This makes me wonder whether there are optimizations that Nvidia can deliver in the near term; I tend to think so, considering that the company has more hardware control thanks to the tight integration between its chips and software.

Nvidia could also partner more. I like the Accenture partnership, but again, it’s not necessarily unique given that most of the AI cloud providers have at least one GSI partner. But what if Nvidia chose to go the same route as Nutanix and partner its agentic software stack with Dell, Lenovo or HPE? That could be a big deal, and you’d have to think it’s readily attainable, considering that all three of those big server makers are already partnered with Nvidia for AI datacenter tech. To get a little more inventive, what if Nvidia partnered with one of the new-school cloud providers like Coreweave?

Finally, I think Nvidia needs to get clearer on its software strategy. Consider this comparison: Over many years, Intel has made good software that is sellable. Intel has also made massive contributions to the open source world. But the software was always secondary to selling more semiconductors. That led to a gap in credibility, and it has stymied Intel’s ability to sell software innovation.

If Nvidia truly wants to succeed in agentic, it will need to turn the (smart) blueprints it announced into some sort of true business model — and provide enterprises with more assurance that production will be sustainable. And again, Nvidia does not need to do all of this by itself; we could see even more opportunities for partnering in the months and years to come.

Follow me on Twitter or LinkedIn. Check out my website.