By 2027, AI Chip Revenue Will Reach 1 Trillion Dollars! Jensen Huang: Tokens Are the Foundation of the AI Era, "Lobster" Will Become Personal Operating System

NVIDIA, a leader in AI chips, announced several major updates at GTC, with Jensen Huang stating that tokens are the cornerstone of the AI era.

On March 16th, local time, NVIDIA CEO Jensen Huang delivered a keynote speech at the GTC annual developer conference. During this over two-hour presentation, Huang announced multiple products including the Vera CPU designed for Agentic AI, the Groq 3 language processor (LPU), and the new neural rendering technology DLSS 5.

Although Huang did not reveal many additional details about the company’s product roadmap, he sent an optimistic signal to the market. Huang predicts that by 2027, the total purchase orders for the Blackwell and Vera Rubin series chips will reach $1 trillion. In October last year, Huang stated that in the next five quarters, the company’s Blackwell and Rubin chip sales could total $500 billion.

Multiple Chips Released, Vera Rubin Platform Achieves Major Breakthrough in Inference Performance

At the start of his speech, Huang noted that this year marks the 20th anniversary of NVIDIA’s CUDA architecture. He said, “We spent 20 years building hundreds of millions of GPUs and computing systems running CUDA worldwide. We have penetrated every cloud and every computer company. We serve nearly every industry.”

NVIDIA CUDA architecture ecosystem cycle. Source: live broadcast

First, Huang discussed the consumer GeForce graphics cards, calling them “NVIDIA’s greatest marketing campaign” and stating, “You could say NVIDIA was built on GeForce.” He explained that the company’s early push for CUDA technology in graphics laid the foundation for its current position. The features initially developed for GeForce GPUs ultimately sparked the “AI explosion.”

This led to the first major new product of the day—neural rendering technology DLSS 5, which Huang called the “GPT moment” in graphics. The new DLSS 5 system combines traditional 3D graphics data with generative AI models, injecting “cinematic” lighting and material details into images through real-time neural rendering models. This is described as NVIDIA’s most significant breakthrough in computer graphics since the launch of real-time ray tracing in 2018.

Next, Huang showcased NVIDIA’s “circle of friends.” He said that “AI-native companies” are thriving, similar to the internet bubble era when many small companies later became essential giants in AI computing transformation.

Huang emphasized that as model capabilities improve, the demand for computing power is accelerating. Over the past two years, demand has increased by 1 million times. With AI models reaching an “inference inflection point” and beginning to operate as agents performing real tasks, the need for reasoning and thinking has surged. Today, computing power has become the biggest bottleneck.

On that day, NVIDIA officially launched the Vera Rubin computing platform, consisting of seven breakthrough chips, a “revolutionary supercomputer” for driving agents. Huang said its computing capacity has increased 40 million times compared to ten years ago. The platform integrates the new 88-core Vera CPU, which has been optimized in several ways compared to traditional x86 CPUs: threefold increase in memory bandwidth per core, about double the energy efficiency, and approximately 1.5 times performance boost for AI tasks.

The new Vera Rubin NVL72 rack achieved a major breakthrough in inference performance. By connecting 72 Rubin GPUs and 36 Vera CPUs with NVLink 6 technology, it can complete MoE (Mixture of Experts) training with only a quarter of the previous generation’s GPUs. Under the same power consumption, its AI inference throughput is increased 50 times, and the cost to generate one million tokens is reduced by about 35 times.

Vera Rubin NVL72 system becomes the “Inference King.” Source: live broadcast

Meanwhile, NVIDIA announced the Groq 3 LPU, expected to ship in Q3. This new product is also integrated into the Vera Rubin platform. In December last year, NVIDIA acquired the core team of chip startup Groq for $20 billion, marking its largest acquisition to date.

Additionally, NVIDIA launched the “Space Computing” platform, including Space-1 Vera Rubin modules, IGX Thor, and Jetson Orin. Huang emphasized, “Space computing has arrived at the final frontier. As satellite constellations deploy and deep space exploration advances, intelligence must exist where data is generated.” He also mentioned that space environments pose unique challenges, and the company is researching how to address radiation issues.

Introducing the “One-Click Lobster” Deployment Solution and Advocating for All Companies to Prepare for an Agent Strategy

During Huang’s speech, the open-source agent framework OpenClaw (commonly called “Lobster”) became a hot topic. Huang highlighted that OpenClaw will become the “operating system for personal AI,” signaling a new renaissance in the software industry.

Huang said that OpenClaw is timely, and afterward, every company worldwide should prepare an “agent strategy”: “Every IT company, every SaaS (Software as a Service) company, will become a GAS (Generative AI Service) company in the future.” Therefore, tokens will become the future cornerstone of the AI era, and all engineers will need to consume tokens to boost productivity.

To support this, NVIDIA launched the NemoClaw software stack, designed for “Lobster,” emphasizing a “one-click installation” experience. It focuses on ensuring that proxy programs can be used safely without leaking proprietary data. Users can optimize OpenClaw with one click, with built-in OpenShell providing isolated sandbox environments to ensure data privacy and network security.

NVIDIA launches NemoClaw software stack. Source: live broadcast

Huang also announced the formation of the Nemotron Open Model Alliance, aimed at “advancing open-source models.” Members include AI startups like Perplexity, Reflection, and Black Forest Labs. NVIDIA also released a series of open-source models to support the development of agent AI, physical AI, and medical AI.

Huang further stated that the “ChatGPT moment” for autonomous vehicles has arrived. NVIDIA announced collaborations with top automakers such as BYD, Geely, Isuzu, and Nissan to develop L4 autonomous vehicles based on the NVIDIA DRIVE Hyperion platform. NVIDIA also expanded its partnership with Uber, planning to deploy a fully NVIDIA DRIVE AV software-driven autonomous fleet in 28 cities across four continents by 2028.

Of course, Huang did not forget to emphasize his favorite field—robotics—saying, “I can’t think of any robot manufacturing company that hasn’t partnered with NVIDIA.” The event concluded with the debut of “Sven,” a robot from Disney’s Frozen, co-developed with NVIDIA, marking the end of this content-rich conference.

On March 16th, NVIDIA (Nasdaq: NVDA) stock price surged nearly 5% during the keynote, then retreated, ending the day up 1.63% at $183.19 per share, with a total market value of $4.45 trillion.

(Source: The Paper)