NVIDIA GTC25 Conference Unveiled: AI Factories and Quantum Computing Lead the Next Wave of Transformation

Jensen Huang’s blueprint in Washington isn’t just about chips—it’s a vision for the future of American tech dominance.

“AI is not a tool; it is the main driver of productivity.” NVIDIA CEO Jensen Huang declared this to the world from the stage at the GTC25 conference in Washington. This event, hailed as the “Super Bowl of the AI world,” is no longer just about technical specifications. Instead, it sketches a grand blueprint for the future of computing from the heart of the U.S. capital.

From bringing chip manufacturing back onshore to using AI to dominate future communication standards, NVIDIA is adopting a stance of “extreme co-design” to help the U.S. solidify its leadership in AI infrastructure and innovation.

01 Chip Revolution: The Computing Leap from Blackwell to Vera Rubin

At GTC25, Jensen Huang unveiled the next-generation Vera Rubin superchip for the first time, marking another self-surpassing leap in NVIDIA’s computing power race.

This revolutionary superchip integrates one Vera CPU and two Rubin GPUs, equipped with 8 SOCAMM modules.

The astounding performance of Vera Rubin stems from its architectural innovations: the CPU section integrates 88 custom Arm cores, totaling 176 threads; each Rubin GPU corresponds to 8 HBM4 stacks with a capacity of 288GB.

Even more impressive, the NVLink-C2C interconnect bandwidth reaches 1.8 TB/s, completely eliminating communication bottlenecks between the Vera CPU and Rubin GPU.

The performance improvement of Vera Rubin compared to the current Blackwell architecture is staggering. With the same configuration (144 GPU chips), the Vera Rubin NVL144 delivers FP4 computing performance, increasing from Blackwell's 1.1 PFLOPS to 3.6 PFLOPS.

With the use of HBM4 memory, bandwidth also increases from 8 TB/s to 13 TB/s.

Huang also previewed the Rubin Ultra architecture for the second half of 2027, which will further upgrade to HBM4E memory. A single rack solution will be equipped with 576 GPUs.

This configuration will skyrocket FP4 inference computing performance to 15 ExaFLOPS, approximately four times that of the Vera Rubin NVL144.

02 AI Factories: Strategic Deployment from Concept to Large-Scale Implementation

The core theme of this GTC conference—"AI Factories"—has moved from concept to reality. Jensen Huang defines them as “a new comprehensive computing platform specifically built for the generation, transmission, and servicing of massive tokens.”

A revolutionary example from the pharmaceutical industry is Eli Lilly’s Blackwell SuperPOD. This pharmaceutical giant has built the largest AI factory in the industry, utilizing the DGX SuperPOD architecture powered by over 1,000 Nvidia Blackwell Ultra GPUs.

This factory's computing power is equivalent to 7 million Cray supercomputers, capable of solving 900 quintillion mathematical operations per second.

Huang emphasized that AI factories are creating, not eliminating, jobs. The Lilly deployment alone will bring 13,000 high-paying manufacturing jobs and 500 positions for engineers, scientists, and technical staff.

Addressing the challenges of hyper-scale AI deployment, NVIDIA officially launched Omniverse DSX—a comprehensive solution covering the design and operation of AI factories ranging from 100 megawatts to several gigawatts.

This blueprint has been fully validated at the AI Factory Research Center in Virginia, achieving co-design of building, power supply, and cooling with the NVIDIA AI infrastructure stack.

03 Quantum Computing Integration: NVQLink Bridges the Classic-Quantum Divide

Discussing the frontiers of computational science, Jensen Huang recalled quantum physicist Richard Feynman's vision from forty years ago: to create quantum computers capable of directly simulating the laws of nature.

He announced, “Today, we are able to prepare stable, coherent logical quantum bits with error correction.”

To seamlessly integrate quantum computing with GPU computing, NVIDIA introduced the quantum-GPU interconnect technology NVQLink. This innovation allows quantum processing units to call the CUDA-Q computing framework in real-time, reducing communication latency to an extreme level of approximately 4 microseconds.

As an open system architecture, NVQLink can closely integrate the ultimate performance of GPU computing with quantum processors, enabling real-time CUDA-Q calls from QPUs to build accelerated quantum supercomputers.

The industry's response to this technology has been swift and positive. Huang revealed that 17 quantum computing companies and multiple U.S. Department of Energy laboratories have joined this collaborative innovation network.

04 National Strategy: Comprehensive Layout for U.S. AI Leadership

Holding the GTC conference in Washington itself is rich with political symbolism, and many of NVIDIA's announcements are closely aligned with U.S. national technology strategy.

Huang announced a strategic collaboration with the U.S. Department of Energy to jointly build seven next-generation supercomputers, providing powerful computing support for future scientific research.

Among them, Argonne National Laboratory will deploy a system named Solstice, integrating 100,000 NVIDIA Blackwell GPUs. Once completed, it will become the world's largest agent-science platform for public research.

Another system, Equinox, equipped with 10,000 Blackwell GPUs, will deliver up to 2200 EFLOPS of AI computing power, dedicated to cutting-edge scientific computing, simulation, and open research.

In the field of communication technology, NVIDIA launched the groundbreaking NVIDIA ARC—an AI-native 6G wireless technology stack centered on U.S. technology.

Huang emphasized, “Telecom technology is the lifeline of the economy and national security, but currently most global wireless technology deployments rely on systems from other countries. This situation where core communication technology is dependent on others must end.”

As part of the strategic layout, NVIDIA has reached an in-depth cooperation with Nokia, acquiring a $1 billion investment to become Nokia's second-largest shareholder, and will integrate ARC technology into Nokia's future base station systems.

05 Economic Impact: Supply-Demand Dynamics in the Future Computing Market

Huang revealed staggering market demand data at the conference: NVIDIA has shipped 6 million Blackwell GPUs in the past four quarters, and the total sales of Blackwell and the next-generation Rubin chips are expected to reach a scale of $500 billion.

More specifically, 20 million Blackwell and Rubin chips will be shipped over the next 5 quarters (until the end of 2026), generating approximately $500 billion in revenue, which is five times the lifetime shipments of Hopper.

This figure significantly exceeds market expectations of 15 million units, representing a 30% surplus.

Market confidence in NVIDIA surged accordingly. Analysts updated their revenue expectations for NVIDIA's fiscal year 2027 to a year-on-year increase of around 75%, significantly surpassing the previous year's nearly 70% revenue growth rate and the market's expectation of 50% growth for the new year.

06 OmniYQ: High-Cost-Effective GPU Computing Solutions

Facing the strong demand for computing power driven by NVIDIA's cutting-edge chips, the OmniYQ computing platform offers high-cost-effective GPU computing solutions for businesses. The platform integrates the latest GPU technology, supports pay-as-you-go billing, and enables elastic scaling, helping AI startups and enterprises access cutting-edge computing resources at lower costs.

In the wave of AI, computing power has become the "oil" of the new era, but not every company can afford the huge investment required to build its own AI factories. It is precisely in this context that computing power rental platforms like OmniYQ are becoming critical infrastructure driving the popularization of AI.

Whether startups or industry giants, all can leverage such platforms to access cutting-edge computing power more cost-effectively and efficiently, thereby maintaining a leading position in the AI competition.

After all, in the AI future depicted by Jensen Huang, the winners will not be the companies with the most algorithms, but those that can utilize computing power most effectively.