Key Points:
- Nvidia CEO Jensen Huang announced that the company’s largest-ever supercomputing platform, the Vera Rubin NVL72, has officially entered full-scale production.
- The rack-scale system unifies 72 Rubin GPUs and 36 Vera CPUs to deliver 3.6 exaflops of artificial intelligence inference and 75 terabytes of fast memory.
- Vera Rubin is specifically co-designed for “agentic AI” workloads, delivering 10 times the throughput at one-tenth the token cost of the previous Blackwell architecture.
- The advanced system-level launch includes the new Vera CPU, featuring 88 Olympus cores and up to 1.2 terabytes per second of high-speed memory bandwidth.
The global semiconductor giant Nvidia Corp. has officially fired the starting gun on what it describes as the most significant infrastructure buildout in human history. Speaking at the Computex 2026 technology exposition in Taipei, Taiwan, on Monday, June 1, 2026, Nvidia CEO Jensen Huang announced that the company’s largest-ever supercomputing platform, the Vera Rubin, has entered full-scale production. This advanced, rack-scale system is specifically designed to power “agentic AI”—an emerging class of autonomous software agents capable of executing complex, multi-step logical reasoning and decision-making workflows with minimal human oversight.
The transition to agentic workloads represents a fundamental paradigm shift in how humanity interacts with computing power. Historically, users have treated artificial intelligence as a simple, prompt-and-response search tool. By contrast, agentic systems act as autonomous teammates. “Agentic AI is a new kind of workload. One prompt can launch a thousand-step journey of reasoning, retrieval, tool use, and response generation,” Huang explained to a packed audience. To handle these immensely complex, recursive computational demands without breaking corporate budgets, tech developers require entirely new hardware architectures co-designed from the ground up.
The physical crown jewel of this new architecture is the Vera Rubin NVL72, a massive, fully integrated rack-scale supercomputer. The platform unifies 72 advanced Rubin GPUs and 36 standalone Vera CPUs into a single, cohesive processing unit using second-generation NVLink-C2C interconnect technology. This extreme co-design allows the single-rack system to deliver a staggering 3.6 exaflops of real-time AI inference, 75 terabytes of ultra-fast system memory, and a massive 1.6 petabytes per second of high-bandwidth memory (HBM4) bandwidth.
This massive consolidation of processing power delivers unprecedented levels of operational efficiency for data center operators. The Vera Rubin platform targets up to 10 times the computational throughput per watt while simultaneously slashing the marginal token cost of running advanced reasoning models to just one-tenth of the cost of the previous-generation Nvidia Grace Blackwell platform. For enterprise clients and cloud hyperscalers struggling with rising electricity bills and carbon emissions, these efficiency gains represent a vital economic lifeline that significantly lowers the barrier to wide-scale AI adoption.
A key technological driver behind these performance gains is the standalone Vera CPU, which represents Nvidia’s aggressive push to challenge Intel and AMD’s traditional dominance in central processors. The chip features 88 high-performance, proprietary Olympus cores and a cutting-edge LPDDR5X memory subsystem that provides up to 1.2 terabytes per second of memory bandwidth. Nvidia’s internal testing demonstrates that the Vera CPU delivers 1.8 times faster task completion than traditional x86 server processors for complex workloads such as reinforcement learning, data processing, and agentic reasoning.
The financial markets have reacted with immense enthusiasm to the new CPU and supercomputing lineup. Analysts at JPMorgan estimate that the Vera CPU alone will unlock a massive $200 billion market for Nvidia, with the company already recording a staggering $20 billion in advance orders in 2026 alone. To demonstrate the maturity of the production line, Jensen Huang personally hand-delivered the first production-scale Vera Rubin NVL72 units to the CEOs of the world’s leading AI labs, including Sam Altman of OpenAI, Dario Amodei of Anthropic, and Elon Musk’s SpaceX (xAI), as well as the cloud giant Oracle.
Specialized cloud providers are already racing to bring this next-generation hardware online. CoreWeave has officially become the first AI cloud provider to successfully bring up and validate the Vera Rubin NVL72 platform, with Lambda and Oracle Cloud Infrastructure set to follow shortly. Because these massive, $8 million rack systems operate under extreme power and thermal constraints, hardware partners like Supermicro are deploying advanced, direct-to-chip liquid cooling systems. Supermicro’s custom racks utilize an innovative coolant with 1,000 times higher electrical impedance than traditional mixtures, ensuring that minor leaks will not short-circuit the multi-million-dollar hardware.
The successful rollout of the Vera Rubin platform depends on a highly coordinated global manufacturing footprint spanning multiple continents. Nvidia confirmed that hundreds of supply chain partners across 350 specialized factories in 30 countries are actively ramping up production of Rubin-based systems. This extensive ecosystem includes 150 dedicated partners in Taiwan alone, with manufacturing giants such as Foxconn, Quanta Cloud Technology, Wistron, and Wiwynn running their assembly lines around the clock to meet unprecedented demand from global hyperscalers and sovereign cloud providers.
Ultimately, Nvidia’s transition from a graphics chip designer into a highly integrated systems and infrastructure provider represents a historic milestone for the global tech economy. With its market capitalization sitting at an all-time high of $5.11 trillion, the company is proving that its technological lead remains highly secure. By bringing the Vera Rubin NVL72 platform into full production and establishing a clear focus on agentic AI, Jensen Huang has provided the physical foundation for the next major epoch of human-machine collaboration. As these massive AI factories begin operating worldwide, they will undoubtedly transform how businesses, researchers, and societies harness the power of artificial intelligence.
