From Chips to Ecosystem, From Ground to Space: Huang Renxun Showcases "Trillion-Dollar Ambition" and Unveils Nvidia's New Blueprint

From March 16 to 19 local time, NVIDIA (NVDA.US) GTC 2026 Conference will be grandly held in San Jose, California. At 2:00 AM Beijing time on March 17, NVIDIA CEO Jensen Huang delivered the highly anticipated keynote speech at GTC 2026, covering the roadmap release, infrastructure unveiling, and a large number of charts and chip milestones as always.

One of the clearest signals Huang Huang sent in this year’s keynote is that NVIDIA has far surpassed its identity as a chip company. According to Huang, NVIDIA now designs complete architectures, including GPUs, CPUs, networking architectures, rack-level systems, AI software platforms, and developer ecosystems.

At GTC, Huang proposed the concept of an “AI factory,” whose infrastructure is not only used for storing and processing data but also aims to manufacture intelligence at scale. These factories produce not software but tokens. Every time an AI system performs inference, generates text, writes code, or executes tasks, tokens are consumed. This makes tokens the new economic unit of AI. Huang repeatedly emphasized metrics like tokens per watt, indicating that the true bottleneck of AI infrastructure is no longer GPUs. The power of megawatt-level AI factories is becoming the next stage of global infrastructure.

Specifically regarding products, Huang announced that its latest accelerated computing platform is ushering in a new era of space innovation, bringing AI computing into Orbital Data Centers (ODC), geospatial intelligence, and autonomous space operations. By integrating data center-level performance into environments constrained by size, weight, and power (SWaP), NVIDIA enables AI applications to run seamlessly from ground to space and within space, supporting increasingly complex tasks.

NVIDIA Space-1 Vera Rubin modules are the latest addition to NVIDIA’s space acceleration platform. Compared to NVIDIA H100 GPUs, the Rubin GPU on this module offers up to 25 times the AI computing power for space inference tasks, providing next-generation computing support for orbital data centers, advanced geospatial intelligence processing, and autonomous space operations.

Additionally, NVIDIA is installing Groq LPU, Vera CPU, and Bluefield-4 DPU into new data center racks. Huang announced the launch of a new inference server rack based on Groq, called NVIDIA Groq 3 LPX, which will be released in the second half of this year alongside Vera Rubin NVL72 racks, Vera CPU racks, and BlueField-4 STX storage racks.

Regarding the recently popular “little lobster,” Huang praised OpenClaw as the fastest-growing open-source software ever and announced the launch of “Minimal Shrimp Farming” — NemoClaw, a deployment toolchain optimized specifically for OpenClaw, with installation requiring only two commands. Huang stated that this simplicity is deliberate, not accidental. NVIDIA aims to enable every GPU server to seamlessly connect to the OpenClaw ecosystem, tying computing power and agent frameworks together.

In terms of graphics, NVIDIA officially released DLSS 5 at this year’s GTC, calling it the most significant breakthrough in computer graphics since real-time ray tracing in 2018. Using real-time neural rendering models, it injects “cinematic” lighting and material details into pixels, aiming to achieve Hollywood-quality interactive visuals in games. Huang introduced that DLSS 5 will be available for mainstream games this fall and has received support from major companies including Bethesda, CAPCOM, NetEase (NTES.US), Tencent, and Ubisoft. Notably, Huang framed DLSS 5 as an example of a broader computational shift, implying that this approach could extend far beyond gaming and into enterprise computing.

As the global AI computing race continues to heat up, Huang Huang brought a more aggressive growth outlook to the market. He stated that by 2027, the revenue opportunity for the company’s AI chips (including the new generation AI accelerator architecture Blackwell and the next-generation Rubin products) will reach at least $1 trillion (about 6.37 trillion RMB). Huang did not provide further details on this forecast, but it significantly exceeds NVIDIA’s previous quarterly earnings guidance of approximately $500 billion (about 3.19 trillion RMB) by 2026.

Following this news, NVIDIA’s stock price surged to a new daily high, rising about 4.8% at one point. However, it quickly retraced more than half of the gains, closing up 1.63%. Earlier this year, NVIDIA’s stock had already fallen over 3%, and concerns about the sustainability of the AI investment cycle had put pressure on the stock.