Enterprises have been slow to adopt AI in a big way. Recent forecasts predict that is about to change, and NVIDIA wants to reap the benefits.
Enterprise IT organizations can easily be put off by the amount of new technology they will have to master in order to adopt AI and benefit from all the new capabilities it can offer. Many are just completing Big Data initiatives and are loathe to start down an entirely new journey requiring new skills as well as new technology. Nonetheless, 50% of enterprises plan to spend more on AI and ML this year, with 20% saying they will be significantly increasing their budgets, according to research by Algorithmia. The research also found that 76% of enterprises will prioritize AI and machine learning (ML) over other IT initiatives in 2021. Skill shortages remain a primary barrier, but complex AI technology also present significant hurdles.
NVIDIA is way out in front
While most AI semiconductor companies are just trying to get fast chips to work with adequate software that they can be used by ninja AI experts (see our analysis of MLPerf), NVIDIA has built a massive lead with application-specific frameworks as well as user interfaces, development and management tools, and of course very fast hardware to meet customers wherever they are in the journey to exploit AI.
To ease the AI onramp and spur enterprise adoption, NVIDIA has been putting together a suite of AI technologies over the last couple of years that should help enterprises get going. Many of the previously announced tools are now ready to deploy, so NVIDIA put a nice wrapper around the collection and presented the comprehensive suite of frameworks, software, hardware, services and partners. While this can be a bit dizzying, the comprehensive nature mirrors the myriad of enterprise AI use cases it intends to serve. And we were impressed by how much progress the AI leader has made since January, 2020; the company highlighted ten offerings that are all new in the last 18 months.
Since 2020, NVIDIA has launched no less than 10 new frameworks, software tools, hardware platforms and services with the intent to lower the barriers to adoption of AI in the enterprise. New products are circled in green. Source: NVIDIA and Cambrian-AI Research
What is notable in the stack?
NVIDIA runs the risk of confusing some by having so many distinct products in this space, so let me help sort it out. At the top level of the figure above are the NVIDIA application frameworks. Not to be confused with the AI frameworks like TensorFlow and Pytorch, these platforms provide the tools to simplify building specific classes of AI applications, including Smart Cities, Healthcare, Conversations, Recommendation engines, video conferencing, robotics, and Cybersecurity. NVIDIA invested in preparing these frameworks because it believes these to be the low-hanging fruit for enterprise adoption.
Beneath these lie the software for building and running AI’s in the data center and on the edge. NVIDIA AI Enterprise is now available (having been pre-launched at GTC earlier this year) providing a comprehensive cloud-native suite of AI software all built to run on VMware vSphere. Enterprises already comfortable with vSphere and AI concepts can just use AI Enterprise. For those who want to access NVIDIA AI from a managed and hosted platform, NVIDIA Base Command provides provides a turnkey platform of hardware, networking, and software hosted by NVIDIA Partners. Base Command can manage AI creation workloads on infrastructure in an NVIDIA AI LaunchPad environment. In the future, Base Command will be able to orchestrate workloads across EGX-based mainstream servers running NVIDIA AI Enterprise.
All this makes it easy for teams to get started, prep their data and experiment with AI models. Once ready for deployment, customers can continue to run on Base Command or move to on-prem hardware or cloud services. Fleet Command manages the AI inference processing on edge AI devices.
NVIDIA software supports the entire AI development workflow. source: NVIDIA
At the bottom of the stack is the NVIDIA hardware platforms available from NVIDIA and through OEM partners. Of note here is the NVIDIA AI LaunchPad which offers ready-to-run infrastructure for the complete lifecycle of AI from development to deployment. It includes NVIDIA Base Command built on DGX SuperPOD with NetApp storage offered in an Equinix-hosted environment, with pricing starting at $90k/month. These are completely configured systems with the software stack described above, all ready to go for initial enterprise experimentation and deployment.
Conclusions
When people think of NVIDIA’s ecosystem advantage, many often refer to it as CUDA. As you can see, the enterprise offerings expand so far above the CUDA libraries that it is not even mentioned. By making AI easy to buy, easy to develop, easy to deploy, and easy to manage across the cloud and edge, NVIDIA’s ecosystem is massive and will remain a durable defensive moat competitors will struggle to cross.