Listen to this story
Over the weekend, NVIDIA became the 7th company in the world to reach a trillion dollar market cap. However, it seems that all the riches in the world isn’t enough for this AI giant. NVIDIA CEO Jensen Huang spoke at the keynote for the Computex conference in Taiwan earlier today, outlining the GPU giant’s plans for an AI-powered future.
After being in the right place at the right time to capitalise on the generative AI boom, NVIDIA is now building the future of AI compute. By bringing both AI products and services to the market, NVIDIA is diversifying its AI portfolio to make its next trillion. Here are eight ways NVIDIA is looking to create its AI future.
1. Advancing AI compute with Grace Hopper supercomputers
The headliner of the talk was the new lineup of enterprise infrastructure powered by Grace Hopper chips. Huang unveiled the DGX GH200, which brings together up to 256 Grace Hopper chips into a ‘single GPU’. This computer has an exaflop of performance with 144 terabytes of memory, making it a supercomputer in its own right.
Google Cloud, Meta, and Microsoft already have access to his device, which can be used as a blueprint for “future hyperscale generative AI infrastructure”. NVIDIA is also planning to create its own supercomputer called Helios, which will use 4 DGX GH200 systems linked together.
2. Creating the MGX architecture for custom servers
Huang also announced the NVIDIA MGX architecture for building custom accelerator servers. Using this modular architecture, manufacturers can build over 100 different configurations based on the users’ requirements.
The architecture supports all NVIDIA products as well as x86 and ARM processors, fulfilling both AI and HPC requirements. Server manufacturers QCT and Supermicro will release MGX products in August, with manufacturers like ASRock Rack, ASUS, Gigabyte, and Pegatron following suit later down the line.
3. Solving networking pain points in AI clouds
Even as NVIDIA creates some of the fastest and most capable servers in the world, much of today’s server infrastructure is crippled by Ethernet. With its manufacturing chops, NVIDIA set out to recreate Ethernet fabrics, unveiling the Spectrum-X networking platform.
Spectrum-X brings together NVIDIA’s BlueField-3 data processing units with traditional Ethernet switches. This reportedly results in a 1.7x gain in performance and power efficiency.
4. Disrupting advertising with Omniverse
Even as big tech shrugs off the metaverse, it seems that NVIDIA’s Omniverse platform is going nowhere. Team Green just announced a partnership with WPP, the world’s largest marketing company, to supercharge creative workloads in advertising.
Using the Omniverse platform, companies can create digital copies of products and use generative AI to create various ads, videos, and 3D experiences based on the product. According to Huang, this will lead to ads that are generated, not retrieved as they are today.
5. Bringing GenAI to games with ACE
Huang also unveiled the NVIDIA avatar cloud engine (ACE), which will allow game developers to create and build custom models for speech. ACE uses models like NVIDIA NeMo, Riva, and Audio2Face to create a complete package for game developers to create realistic characters in games.
These models can be optimised on a variety of parameters to keep the processing and inferencing times low. Moreover, the models are also highly optimised for latency, which is a must-have in games.
6. Creating autonomous robots with Isaac AMR
The CEO also launched the Isaac AMR platform to bring autonomy to robots. This platform can add advanced mapping and simulation capabilities to robots, and can even manage fleets of autonomous mobile robots. By using AMR, factories and robotics companies can effectively deploy large fleets of robots and manage them in a single place. NVIDIA has partnered with companies like ADLINK, Deloitte, and Siemens for this platform.
7. Powering 5G and 6G applications in Japan
NVIDIA has tied up with SoftBank to build out a network of data centres in Japan. By combining various aspects of its technology such as Grace Hopper, BlueField DPUs, Spectrum switches, and MGX, NVIDIA is building out 5G infrastructure in Japan.
8. Optimising manufacturing with AI
NVIDIA also wants to bring optimisations to the manufacturing process using AI. Not only is it pushing Omniverse to a new level for pre-manufacturing processes, it is also bringing in Isaac Sim and NVIDIA Metropolis to further reduce overhead.
After increasing the market cap by $184 billion in the last week, it seems that Jensen is putting all his eggs in different baskets, united by a green banner. Considering the current state of the compute market and the amount of diversification NVIDIA is undertaking, future growth is all but inevitable.