It’s a busy time for NVIDIA’s VP of Enterprise Computing, Manuvir Das. After all, it isn’t just AI that’s at an inflection point, enterprise AI too is at crossroads. Das, who reports directly to company chief Jensen Huang, is convinced that the AI revolution in enterprise will be as big as the cloud transformation, if not bigger.
Bigger than the cloud transformation
“Let me explain why the AI transformation will be bigger. With the cloud transformation, it changed how the IT teams worked but AI is expected to impact all the employees of the company. It will make every employee more productive, not just the IT specialists.
“If you consider a bank, customer service agents will become more productive because they will have an AI assistant to help them work faster and more accurately. Finance people will become more productive because they will get better reports generated automatically instead of spending hours and hours. The HR employees will become more productive because they’ll get an automatic analysis of which team is performing well and which one is performing badly. So, the impact will be quite widespread,” Das said.
He explains what the next era of enterprise computing will look like. “Until now, any enterprise company that has benefited from AI has had to learn a lot about AI themselves, so they can be the practitioner. But now with generative AI, you have all these other companies like Slack, Microsoft, ServiceNow and SAP, whose products are used by thousands of companies. And if they incorporate generative AI into their products, then enterprises can benefit from it. This will be the big shift in the industry,” he stated.
Shift in cloud computing segment
But Das admits that integrating generative AI into enterprises seems deceptively simple. “Everybody looks at ChatGPT and thinks, ‘Hey, cool! I’m gonna use ChatGPT too,’ but the reality isn’t so simple,” he said.
Das noted down three things that need to be considered. “The first is that every company has its own data as well that’s highly proprietary and confidential. The second thing is depending on what your company is doing, the skills needed from the model are going to be different. For example, you can teach the model how to write a code or you can teach the model how to generate an expense report.
“And then the third thing is that in a business setting, it’s equally important what the model isn’t allowed to say or do. Say, if I’m a bank, and I’m using one of these models for customer service then you obviously don’t want your chatbot to be answering such sensitive questions. So, we introduced the concept of guardrails,” he stated.
According to Das, giant foundational models like GPT-4 are only the starting point for these enterprises because they have a lot of general knowledge from the internet and some general text prediction. “What an enterprise company actually needs is their own models that are highly specialised for a good performance,” he continued.
Increased competition in chipmaking
The wheels of change will also overhaul other major landscapes. Das predicts that the cloud segment too will alter because of AI. “The way in which cloud has been stored in data centres, it’s been these traditional servers with CPUs not GPUs in them. This traditional model of computing does not really work well for AI, especially in generative AI, because of the lack of accelerated computing that GPUs offer.
If you look ten years from now, most servers in the cloud are probably going to have GPUs because they really are the only efficient and cost effective way to do AI,” Das stated.
But even as the technology itself evolves at a fast pace, the industry itself also changes. A number of Big Tech players like Google and Microsoft are all working on making AI chips increasing competitiveness in an expensive industry. But Das responded saying that NVIDIA doesn’t necessarily fear this. “The work that these companies are doing is a validation of the accelerated computing approach that NVIDIA brought to the world. The way we designed the GPU and the reason why it has stood the test of time is because we built an interface like CUDA and we’ve developed this whole ecosystem. So, our approach to making chips is fairly general purpose and they can be used for many different use cases. Every time we come up with a new generation of chips, it’s on the same software interface so developers don’t have to learn anything new,” he said.
Instead, Das said that NVIDIA encourages innovation. “We want to constantly improve our own GPUs as well at the same time. We’re very comfortable with the curve of innovation and our GPUs because of our years of knowledge. This isn’t the first version that we’ve built,” he said.
NVIDIA’s Enterprise partners
And there’s good reason for Das’ confidence. NVIDIA has a slew of partners to spread its enterprise technology.
“If we take the analogy of cars, NVIDIA builds the engine. Our role is to build a really good platform and keep innovating on it. For this strategy to actually work well, we need lots of partnerships with these other people.
We focus on cloud providers like Azure, Microsoft, Google, Amazon and Oracle, who are all working very hard to make their platform the better choice for enterprise customers, in terms of cost and capabilities. We’ve been working with them for several years now to integrate all our hardware and software products onto the public cloud. So, we don’t compete with them, we help them.
Our second type of natural partnerships are the ones we have with customers who have their own datasets and put servers into their data centres. They buy data centres from Dell, HP or Lenovo and we work with them to design servers with our GPUs.
The third type of partnerships which are newly developing are the next wave of companies like, say, CoreWeave which aren’t as big but want to build a specialised cloud just for accelerated computing. As accelerated computing gains importance, they are becoming providers for this and want to use our GPUs in their data centres. Equinix is another good example of this.
Our fourth kind of partner is a typical enterprise company like a bank which decides to go through cloud transformation, and that becomes a five-year long process. They don’t do it on their own and instead sign up with a Global Service Integrator (GSI) like Accenture or Infosys who we partner with. Like now, we have a big partnership with Deloitte and even work with Accenture, Infosys.
And finally, the reason that NVIDIA has been successful is because there’s a crucial partner that people don’t stop and think about, which is the software developer.
So, when we built our GPU, we built an SDK called CUDA that developers use to access our GPU and write the application. We’ve put in a lot of effort into helping developers use our platform which is why we have around 3.5 million developers in our ecosystem. These are ultimately the people who create the software for AI and they are very precious to us,” Das explained.
Scope in India
For Das, India is a crucial piece of the puzzle considering some of the biggest GSIs in the world are based in India. “NVIDIA has big partnerships with companies like Wipro, TCS and so on. We are also working closely with government institutions in India because, as you know, the Indian government has started a lot of initiatives to be technologically independent. There have been efforts to build supercomputers in India and these have been built using NVIDIA’s tech.
Then, there are other massive companies in the telecom sector like Reliance, or in retail companies which really benefit from AI and accelerated computing and we have partnered with these companies too,” he said.
The post ‘AI Transformation Will be Bigger Than Cloud Transformation for Enterprises’: Manuvir Das, NVIDIA appeared first on Analytics India Magazine.
