Behind the talk of petaflops and bidirectional throughput at Nvidia’s developer conference was what could end up being a game-changing software.
What happened: While the industry was focused on Nvidia’s new Blackwell chips — which are really fast — the company also debuted Nvidia Inference Microservices (NIM), a platform to make deploying AI faster and easier.
- NIM is a package with all the software needed to get an AI model off the ground, which clients can tweak to their needs and run on any system with a Nvidia GPU.
- Meanwhile, Nvidia claims Blackwell can train a model the size of OpenAI’s GPT-4 with a quarter of the GPUs and 27% of the power, compared to its previous chips.
Why it matters: The industry was watching for Nvidia to present a chip exciting enough to fend off competitors, and the chip maker seems to have delivered. But things like NIM could be what helps it not just defend its position, but grow its business by being more than ‘the AI chip company.’
- These microservices could also create ongoing revenue for Nvidia, as opposed to one-time hardware transactions.
- It also announced new partnerships in sectors including cybersecurity, quantum computing, medical tech, and mixed reality headsets.
Nvidia had a wealth of other announcements in high-growth areas:
-
GR00T, a platform to help companies develop AI for humanoid robots. It is already working with major players in the growing industry, including Boston Dynamics, 1X, Figure, and Vancouver-based Sanctuary.
-
Hyperion 8, the latest version of a toolkit to develop autonomous vehicles, including the sensors, software, and computing manufacturers need.
- Earth-2, a set of APIs to predict the impacts of climate change and extreme weather events.
Bottom line: Nothing good lasts forever, and some analysts have questioned how long Nvidia can maintain its red-hot growth once companies finish stocking up on AI chips. Diversifying Nvidia’s business even further might make any fall it experiences less severe.