In June 2026, NVIDIA entered a pivotal moment.
At the start of the month, during Computex Taipei, Jensen Huang announced the full-scale production of the Vera Rubin platform, officially launching a new generation of AI factory engines. By the end of the month, at the Automate 2026 conference in Chicago, NVIDIA unveiled Halos for Robotics—the industry’s first end-to-end robotic safety system—translating over 18,600 engineering years of safety expertise from autonomous driving into the physical AI domain. On June 24, the NVIDIA 2026 Annual Shareholders Meeting is set to take place, with Blackwell and Vera’s production ramp-up and AI ecosystem commercialization progress at the top of the agenda.
From Grace Blackwell to Vera Rubin and the rollout of robotic safety systems, NVIDIA is building a comprehensive hardware universe spanning data centers, AI factories, and the physical world. This article analyzes the latest moves of this $5 trillion AI giant from three perspectives: product and technology evolution, market dynamics, and investment rationale.
Vera Rubin Mass Production: The Arrival of Third-Generation Rack-Scale Systems
On June 1, 2026, NVIDIA officially announced that the Vera Rubin platform had entered full-scale mass production. This is not just another routine product iteration—it marks the most strategically significant platform upgrade since Grace Blackwell.
Vera Rubin is NVIDIA’s largest POD-scale platform to date—a massive AI supercomputer built from five dedicated cabinets, purpose-built for agent workloads. The platform integrates the NVIDIA Vera Rubin NVL72 system, Vera CPU, Groq 3 LPX, BlueField-4 STX storage, and Spectrum-6 SPX Ethernet racks into a fully integrated system. Compared to the previous-generation Grace Blackwell platform, Vera Rubin delivers a 10x increase in agent throughput at scale.
During his GTC Taipei 2026 keynote, Jensen Huang defined Vera Rubin’s positioning as follows: "Agent AI is a brand-new workload. A single prompt can trigger a computational process spanning thousands of steps, including inference, information retrieval, tool invocation, and response generation. Vera Rubin was born for this. It’s an AI factory engine built to deliver intelligence at scale, with the performance, efficiency, and security required to drive the next industrial revolution."
From a supply chain perspective, Vera Rubin’s production scale far exceeds its predecessor. NVIDIA’s supply chain ecosystem covers more than 30 countries and 350 factories worldwide, with over 150 partners in Taiwan alone. Jensen Huang noted that Vera Rubin’s supply chain is twice the size of Grace Blackwell’s. Major system manufacturers—including Dell Technologies, HPE, Lenovo, and Supermicro—are fully committed to Vera Rubin production. The first batch of products is expected to ship to cloud service and enterprise customers starting in fall 2026.
On the technical architecture front, Vera Rubin introduces several key innovations. Spectrum-X silicon photonics Ethernet technology has reached full-scale production—deeply integrating optoelectronic packaging with Spectrum-X switches to power AI factories at million-GPU scale. The Vera CPU uses NVIDIA’s proprietary Olympus core and a scalable coherence architecture, with official claims that its agent sandbox performance is 1.8 times that of x86 CPUs. For memory, Vera Rubin utilizes HBM4 high-bandwidth memory from Micron, SK hynix, and Samsung.
Notably, Jensen Huang positions the Vera CPU as a "CPU built for agents," rather than a traditional human-driven computing chip. At Computex, he stated that the Vera CPU "will be more popular than GPUs" and will become NVIDIA’s "new primary growth driver." The rationale: agent workloads demand low latency, high single-thread performance, high bandwidth, and strong energy efficiency—areas where CPUs play an irreplaceable role in coordinating tool calls, memory access, and GPU-adjacent workflows.
From Data Center to Physical World: The Full-Stack Safety Logic of Halos
If Vera Rubin answers the question of "how to scale AI factory intelligence production," Halos for Robotics addresses "how AI can safely enter the physical world."
On June 22, at the Automate 2026 conference in Chicago, NVIDIA launched Halos for Robotics—the industry’s first comprehensive, end-to-end safety system for robotics and physical AI. This system extends NVIDIA Halos’ proven safety architecture from autonomous driving to robotics and physical AI scenarios, providing a unified safety framework for machines that perceive, make decisions, and act in the real world.
Halos for Robotics is built on over 18,600 engineering years of safety development and 7 million lines of validated code from NVIDIA’s autonomous driving efforts. The system covers the entire stack, from chips and sensors to operating systems and safety certification.
Architecturally, Halos establishes a four-layer safety framework:
Platform Security Layer addresses hardware reliability. NVIDIA IGX Thor, designed for robotics and industrial AI, features an independent "safety island"—with its own processor, I/O, power, and clock—physically isolated from the main compute system. Even if the primary AI system crashes or malfunctions, the safety island can independently execute critical functions like emergency braking. On the same layer, the Holoscan Sensor Bridge solves latency issues from heterogeneous sensors, unifying all sensor data into the safety compute domain for low-latency, synchronized processing.
Safety Operating System Layer ensures system stability. Halos OS runs atop IGX Thor, supporting either pure Linux or a Linux+QNX hybrid architecture. In hybrid mode, NVIDIA uses a hypervisor to split the system into two isolated domains: Linux handles AI computation and applications, while QNX manages safety-critical tasks—each running completely independently.
Algorithm Safety Layer introduces external perception. The Outside-In Safety Blueprint uses external cameras mounted on ceilings and other vantage points, with independent AI monitoring robot behavior from a third-party perspective. This capability is now open to developers and available as open source.
Ecosystem Safety Layer addresses certification and standardization. The NVIDIA Halos AI Systems Inspection Lab is the world’s first functional and AI safety program recognized by the ANSI National Accreditation Board, helping partners prepare for third-party certifications from leading bodies like TÜV Rheinland and UL.
On the ecosystem front, humanoid robotics company Agility has already integrated Halos into its Digit robots, now deployed in factories of clients such as Amazon, GXO, and Toyota. The Halos ecosystem has expanded to over 43 partners, including Boston Dynamics and Hesai Technology.
Industry observers have likened this strategy to "the Android model for embodied intelligence"—NVIDIA does not manufacture robots directly but opens its safety platform to everyone. This approach aligns with NVIDIA’s positioning in the AI factory era: providing foundational infrastructure rather than occupying the application layer.
SMCI Blueprint Implementation: Mapping the Vera Rubin Ecosystem Across the Supply Chain
The mass production of Vera Rubin is not just a product milestone—it’s a supply chain event.
On June 22, Supermicro unveiled its data center modular solution blueprint based on the NVIDIA Vera Rubin NVL4 platform at ISC 2026. This blueprint offers end-to-end HPC and AI infrastructure solutions, with a single scalable unit housing up to 1,152 NVIDIA Rubin GPUs and 576 NVIDIA Vera CPUs, utilizing liquid-cooled rack designs and scalable unit power up to 3.2MW. Supermicro CEO Charles Liang stated, "With our DCBBS blueprint, research institutions can confidently deploy HPC and AI infrastructure at any scale."
The market reacted swiftly and decisively. On June 22 (Monday), SMCI shares surged 15.66% in a single trading day, closing at $35.46, with intraday gains peaking at 19%. Trading volume reached 128 million shares. On the same day, NVIDIA closed at $208.65, down 0.97%, while the Nasdaq fell 1.32% to 26,166.60.
SMCI’s standalone rally reflects the market’s structural demand for AI infrastructure hardware. Despite broader Nasdaq pressures, hardware suppliers directly linked to Vera Rubin have seen significant valuation premiums. Analyst firms have raised SMCI’s target price to $48. This price signal suggests the market is repricing system integrators within the Vera Rubin ecosystem—reassessing the value distribution of hardware in the AI investment cycle.
Shareholders Meeting Preview: Blackwell, Vera, and the Trillion-Dollar Revenue Outlook
At 00:00 Beijing time on June 25 (9:00 AM Pacific Time, June 24), NVIDIA’s 2026 Annual Shareholders Meeting will be held online. Key topics include: the production ramp-up of Blackwell and the new Vera architecture chips, AI ecosystem commercialization progress, and capital return plans for NVIDIA’s massive cash flow.
Looking back at the 2025 shareholders meeting, several key messages emerged: NVIDIA is entering the early stage of a "decade-long AI infrastructure buildout"; AI and robotics are two major growth opportunities; the era of robotics and autonomous driving has arrived. On the day of the meeting, NVIDIA’s stock price rose 4.3%, closing at a record high of $154.31.
In terms of product cadence, NVIDIA has committed to launching a new AI chip generation every year: Blackwell architecture in 2024, Blackwell Ultra in 2025, and a new platform in 2026 comprising Vera CPUs and Rubin GPUs. The Blackwell series, as the flagship for 2024–2025, remains in tight supply. In Q1 of fiscal 2026 (ending April 2026), NVIDIA’s data center revenue reached $75.2 billion, up 92% year-over-year and 21% quarter-over-quarter, driven mainly by the widespread adoption of Blackwell 300 products.
At the GTC Developer Conference, Jensen Huang predicted that the Blackwell and Rubin product lines alone would generate a combined $1 trillion in revenue across 2026 and 2027. This forecast underscores NVIDIA’s confidence in the ongoing AI infrastructure investment cycle. Whether the shareholders meeting will update this revenue guidance, and whether Vera’s mass production pace will impact Blackwell’s capacity allocation, are key points for the market.
From a valuation perspective, NVIDIA’s current market cap is around $5 trillion, with a forward P/E of about 23x based on 2026 earnings estimates. As AI infrastructure capital expenditures continue to expand, the reasonableness of this valuation hinges on Vera Rubin’s ability to deliver incremental revenue as planned and the sustainability of AI factory capital spending.
The Structural Logic of AI Infrastructure Investment
The mass production of Vera Rubin and the launch of Halos both point to a broader trend: AI infrastructure investment is shifting from "model training" to "large-scale deployment."
In 2026, AI infrastructure capital spending faces three core bottlenecks: power, memory, and optical bandwidth. Vera Rubin’s focus on energy efficiency, HBM4 memory integration, and Spectrum-X silicon photonics is aimed at engineering solutions for these challenges. The rollout of SMCI’s liquid cooling solutions and NVIDIA’s doubling down on supply chain scale are fundamentally about lowering the deployment barriers and operational costs of AI factories.
Jensen Huang’s remarks at GTC Taiwan offer a key insight: "Compute is revenue, compute is profit." Metrics like performance per watt, reliability, deployment speed, and system lifespan are becoming central economic indicators for AI infrastructure operators. If this logic holds, the value of AI hardware suppliers depends not only on peak chip performance but also on their ability to reduce total cost of ownership at the system level.
Within this framework, Vera Rubin’s 10x agent throughput, Halos’ standardized safety architecture, and SMCI’s end-to-end deployment solutions together form a complete value chain from chips to systems. NVIDIA is transforming from a GPU company into an AI infrastructure company—with the goal of becoming the core supplier for over 100GW of new global AI factory capacity by 2030.
Conclusion
In June 2026, NVIDIA is advancing on three fronts: Vera Rubin’s mass production pushes AI factory scalability to new heights; Halos for Robotics debuts, extending safety architecture from autonomous driving to physical AI; and the upcoming shareholders meeting will see the market scrutinize Blackwell and Vera’s production cadence and revenue outlook.
From Blackwell to Vera Rubin and robotic safety systems, NVIDIA’s "complete universe" is not a closed hardware ecosystem but a full-stack infrastructure system spanning data center compute to physical world deployment. The commercial value of this system depends on how quickly AI evolves from "conversational" to "agentic," and on the pace at which AI factory capital expenditures scale from gigawatt to hundred-gigawatt levels.
For those tracking AI infrastructure investment logic, Vera Rubin’s production ramp-up, the speed of Halos ecosystem expansion, and the capacity and revenue signals from the shareholders meeting will be key markers for assessing where we are in this investment cycle.
