Presented by Supermicro/NVIDIA
AI delivers business value and a competitive advantage for enterprise, but there’s one obstacle: graduating from proof of concept to production AI at scale. In this VB Spotlight event, learn how an end-to-end AI platform helps deliver strategic projects and business value fast.
Watch free on-demand!
“AI is as transformative as the internet to the structure of business, how business is being done and its impact,” says Anne Hecht, senior director, product marketing, enterprise computing group at NVIDIA. “Every business and department is starting to use AI and finding opportunities to operationalize, be more efficient and develop more intimate relationships with their customers.”
Consumers are interacting with these AI products every day, from the recommendation engines developed by marketing departments to the intelligent virtual assistants, which enable customers to get results faster, to route optimization for logistics departments (and faster pizza delivery for us). It’s a transformative technology already, but generative AI and applications like ChatGPT are shaking up the way business is done. Enterprises are looking for ways to unlock the potential of AI, and realize cost savings, operational benefits and new business models.
“Despite all these opportunities, we’re finding that enterprises are challenged to move these use cases into full production,” Hecht says. “There’s tremendous potential, and yet only — maybe a third of enterprises are in full production with AI right now.”
The challenges of deploying AI at scale
The challenges range from the technical to the human, says Erik Grundstrom, director, FAE at Supermicro. Cost is always number one, of course. But on the technology side, there’s the technical complexity of migrating disparate systems into a unified platform. Then there’s mapping data from multiple systems to a unified platform, which requires deep understanding of the data structure and relationships between the data.
The application environment often requires multiple teams, each with their own expertise, working together to create a singular platform — and on top of that, ensure the data is still reliable and the applications remain high performing.
“Pulling that team together is probably the biggest challenge today,” Grundstrom says. “Disparate groups within a company are all working on their own models and projects, in their own departments.”
The support team’s environment used to develop a chat bot is very different from the environment and the tools being used by the team doing the recommendation engine, and there’s no unification of infrastructure and resources across all these environments. When everyone’s just doing their own thing, it turns into the wild west.
“Creating a unified structure presents a lot of new challenges at the enterprise level,” Grundstrom says. “But companies that are making that happen are benefiting the most out of predictive analytics and getting the best quality information from their AI at scale.”
The other key issue that makes AI production complicated for enterprises is that it’s much different than a standard enterprise application, Hecht adds. You don’t build it, deploy it and come back and do an update 12 months later. An AI application is continuously run and trained with new data for additional inferencing, to keep it current, make it smarter and ensure it adapts to evolving circumstances. On top of that, you need to consistently ensure the quality and integrity of your data.
“It takes most enterprises, on average, about seven to seven and a half months to develop and train a model,” Hecht says. “Often they’re leveraging a pre-trained model. And then moving it into production. Then they’re still dealing with the fact that almost half of those never make it to production. If we can reduce that time, that’s very powerful for our customers.”
Accelerating the AI pipeline
Enterprises early in their journey commonly have developers and teams building out their own infrastructure, leveraging a cloud instance, or developing on local workstations or PCs. They’re using open-source frameworks and pre-trained models, to do their development work. Those tools can be a great place to start, but where they fail enterprises is their incompatibility. And thus, applications developed in these highly customized shadow IT environments often can’t be deployed into the data center, or end up patched in, rather than assimilated, and it becomes incredibly difficult to scale. AI production becomes a hassle instead of a win.
To solve this, the AI pipeline must be optimized to accelerate every step and get to market with an application within days as opposed to months. Adding acceleration cuts down a lot of the time it takes to train and process the data as well, which means cutting costs, because you don’t need as much infrastructure. An end-to-end production AI platform, which comes along with a partner and tools, technologies and scalable and secure infrastructure, is essential.
The companies that are becoming successful are driving this from a strategic standpoint. They’re taking the time to develop the full business strategy, and approaching AI as a center of excellence, putting together the governance, processes, people and teams. They are making the infrastructure investments, while including security practices, privacy practices and data management practices to make AI core to their business.
“If you start from that standpoint, it’ll naturally reveal what infrastructure you need and which partners you want to work with, so that you build out a comprehensive and streamlined AI infrastructure for your business,” Hecht says. “Something that’s flexible, that can address any AI workflow, any AI opportunity that might present to your organization and to your business.”
To learn more about the infrastructure and partners that are foundational to successful production AI, a deep dive into the power of NVIDIA AI Enterprise and more, don’t miss this VB Spotlight!
Watch on-demand now!
- Why time to AI business value is today’s differentiator
- Challenges in deploying AI production/AI at scale
- Why disparate hardware and software solutions create problems
- New innovations in complete end-to-end production AI solutions
- An under-the-hood look at the NVIDIA AI Enterprise platform
- Anne Hecht, Sr. Director, Product Marketing, Enterprise Computing Group, NVIDIA
- Erik Grundstrom, Director, FAE, Supermicro
- Joe Maglitta, Senior Director & Editor, VentureBeat (moderator)
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.