A Closer Look at a Critical Partnership for Hybrid Generative AI

Hybrid AI is one of the big trends for 2023, but not all hybrid AI solutions meet an essential condition.

• By Rob Enderle
• January 11, 2024

In October, 2023 Lenovo and NVIDIA announced an interesting and powerful new partnershiphybrid generative AI. Hybrid AI is one of the big trends I’m following this year because virtually every chip company is on board with this move. However, NVIDIA is still the AI king in terms of technology with broad partnership across most of the companies in a variety of industries working on this technology. Lenovo is an interesting partner because it has been aggressively moving against its peers with ever more innovative solutions in a variety of areas, including water-cooled server solutions (an advance on what it got from IBM when it bought IBM’s server business).

The thing about any current AI deployment is experience — because there is still so little of it in the market. Many of the vendors are building their own solutions using NVIDIA hardware but aren’t engaging as deeply with NVIDIA to gain access to the company’s uniquely deep understanding of AI, particularly when it comes to training and testing that AI at machine speeds with simulation rather than doing it live.

Hybrid AI solutions are even more difficult now because neither the server nor the desktop side of this solution is optimized for AI. This is why so many initial implementations of this technology are behaving suboptimally.

The Key Characteristic of Hybrid AI

When someone refers to something as “hybrid AI,” a lot of what I see is AI that is mostly running on the desktop to reduce cloud or server loading and the related costs. To me, this isn’t truly a hybrid.

For instance, if we talk hybrid learning, we aren’t just talking about people working from home — we are talking about a situation where some are at home and some are in school real-time. True hybrid AI is an implementation that is concurrently using (when possible) both desktop and server or cloud resources at the same time, load-balancing the effort and dynamically shifting that load based on the needs of the entity providing the service and the needs of the user.

To do the former (where software comes from the server and is run on the desktop largely in isolation) is easy because we know how to run things on servers and PCs independently of each other. This means you really don’t need any true commonality on either the server or the PC, so you can create solutions from a variety of mix-and-match product suppliers. However, if you want to run AI concurrently on both the laptop and the server/cloud and provide dynamic loading that factors in users’ needs and the need to optimize the use of the firm’s resources, you need a lot of commonalities between the client and the server solution so that transitions between the two components are seamless and invisible to the user.

This is the unique and critical role NVIDIA plays. Not only is it expert in AI in general, it also exists in both server and desktop environments. Much of the AI technology today is designed to run on NVIDIA’s GPUs because of their early dominance of the AI market. By working together, Lenovo and NVIDIA are bringing what may be the only true hybrid AI solution to the market.

Final Thoughts

Hybrid AI is one of the big trends for 2023. Everyone and their brother, from Microsoft to IBM, is working on related solutions. For true hybrid solutions, however, you need common technology at both ends to reduce the need for emulators and to assure the seamless movement of all aspects of the solution between the two extremes or balance across those extremes based on network access and need.

Although I don’t expect them to be the only partners for long, for now Lenovo and NVIDIA showcase the closest potential match to what should be a true hybrid AI solution.