What New Story Unfolds in Jensen Huang's "Agent Factory"?

On June 1, 2026, Jensen Huang, founder and CEO of NVIDIA, delivered a keynote speech at the NVIDIA GTC Taipei event held during COMPUTEX 2026.

It had been only three months since the previous GTC.

At that time, NVIDIA released the "chip suite" for Vera Rubin, including: Vera CPU, Rubin GPU, Groq 3 LPU, ConnectX-9, BlueField-4 DPU, and Spectrum-6 switches. These six chips constitute a rack-level AI supercomputer, reducing the number of GPUs required to train large MoE models to one-quarter, increasing inference throughput per watt by 10 times, and lowering the cost per token to one-tenth.

Unlike the previous emphasis on system-level solutions like the "chip suite" and "computing power suite," at COMPUTEX three months later, Jensen Huang shifted his focus to the target these infrastructures serve—Agents.

Huang revealed in his speech that Vera Rubin has officially entered mass production, with Vera CPUs beginning global delivery. DGX Station entered enterprise desktops in Windows form for the first time, Cosmos 3 reconstructed the perception framework for Physical AI, and DSX became the operating system for AI factories. NVIDIA also collaborated with Unitree to release H2 Plus—the first humanoid robot reference design based on Isaac GR00T—extending the boundaries of Agents from the digital world to physical forms.

NVIDIA is reorganizing its complete technology system, from chips, data centers, and models to software and robotics platforms, around the Agent ecosystem.

Huang stated, "The era of Agent AI and practical artificial intelligence has arrived. Now, tokens are the unit of profit, AI is a 'GDP generator,' and the number of software engineers is increasing. People talk about AI reducing jobs, which is complete nonsense; in fact, more software engineers are being hired."

Same AI Factory, 10x Agent Tasks

The Vera Rubin platform is now in full production.

Unlike previous generations primarily oriented toward large model training and inference, Vera Rubin was designed from the outset with Agents as a key workload.

Huang stated in his speech that an Agent task often involves not just a single model inference but multiple steps including reasoning, search, tool invocation, code execution, and result validation, potentially involving thousands of steps behind the scenes. Future data centers will need to handle not just single model requests, but largely numerous continuously running, collaborative Agent tasks.

The platform is defined as a massive, unified compute-unit-level AI supercomputer, built specifically to handle agent workloads ranging from inference and retrieval to tool usage. In ultra-large data centers of the same scale, running autonomous AI agent tasks on the new Vera Rubin platform achieves processing efficiency 10 times that of the previous-generation Grace Blackwell platform.

Beyond the computing platform itself, networking has also become a key focus of the Vera Rubin upgrade.

In past data centers, data transmission between GPUs mainly relied on traditional optical modules and switch architectures. However, as cluster scales continued to expand, power consumption, heat dissipation, and deployment complexity rose rapidly. To address this, NVIDIA introduced the Spectrum-X Ethernet Photonics networking system in the Vera Rubin platform.

This marks NVIDIA's first large-scale introduction of Co-Packaged Optics (CPO) technology into AI data center networks.

Simply put, while traditional solutions require plugging optical modules into the exterior of switches, CPO directly integrates optical components inside the switch, thereby reducing energy consumption and signal loss.

Furthermore, security is a core capability heavily emphasized in this Vera Rubin platform.

To this end, NVIDIA has extended Confidential Computing capabilities across the entire Vera Rubin platform. Through trusted execution environments, hardware-level verification, and end-to-end encryption mechanisms, enterprises can obtain higher levels of security assurance when handling private data, industry-sensitive information, and critical models.

Huang revealed that Vera Rubin has entered the mass production stage. As the third-generation MGX rack-level system, it involves over 150 partners, more than 350 factories, and a supply chain system covering over 30 countries and regions. According to NVIDIA's announced plan, Vera Rubin will begin formal shipments this autumn.

Processors "Born for Agents"

NVIDIA launched Vera, a new processor designed specifically for the agent era, which is now in full production.

Huang pointed out that advancements in memory systems will drive innovation and modernization in storage systems. All CPUs to date have been built for humans, whereas Vera is a CPU designed for the AI era, built for agents.

As the successor to Grace, Vera adopts NVIDIA's self-designed "Olympus" CPU core architecture, increasing the core count from 72 to 88 and significantly enhancing memory and data processing capabilities. According to NVIDIA, in Agent-related workload tests, Vera's task execution speed is 1.8 times that of contemporary x86 server CPUs.

More important than pure performance improvement is the change in the relationship between Vera and the Rubin GPU: Vera connects to the Rubin GPU via second-generation NVLink-C2C, achieving an interconnect bandwidth of 1.8TB/s, further reducing the overhead of data transfer between the CPU and GPU during Agent operation.

Huang stated that Vera Rubin uses HBM (High Bandwidth Memory) from Micron, SK Hynix, and Samsung, with a supply chain scale "twice" that of the previous-generation Blackwell. However, while deploying a large Blackwell rack takes two hours, the time for Vera Rubin has been compressed to the 5-minute level.

Moving AI Factories from "Construction" to "Operation"

The DSX launched by NVIDIA this time can be understood as a "toolkit for AI factory construction and operation."

In the past, when building AI data centers, customers had to separately consider servers, networking, power, cooling, server room design, and operations systems, with many links relying on coordination among different suppliers. What DSX aims to do is integrate these previously scattered links into a single framework, providing customers with a referable and verifiable standard solution for design, simulation, construction, and operation.

At the launch event, Huang stated: "NVIDIA is not just selling chips; we are providing infrastructure builders with a complete blueprint for AI factories."

The most important new capabilities of DSX this time are mainly two.

The first is DSX MaxLPS. It addresses the most practical problem of AI factories: how to fit more GPUs and generate more Tokens under a fixed power budget.

According to NVIDIA, MaxLPS, combined with liquid cooling and intra-rack power optimization, allows operators to run up to 40% more GPUs without significantly impacting performance.

The second is DSX OS. It serves as the operating software for the AI factory, responsible for lifecycle management, intelligent scheduling, health monitoring, fault recovery, multi-tenant management, and other tasks. Simply put, if the AI factory is a complex plant, DSX OS ensures its continuous and stable operation.

Within the DSX product matrix, Reference Design provides AI factory reference designs, telling customers how to set up server rooms, racks, networking, power, and cooling systems; DSX Sim handles simulation, allowing customers to verify design feasibility before construction; DSX Flex connects the AI factory to the power grid, enabling data centers to adjust tasks based on electricity prices, load, and demand response signals; and DSX Exchange is responsible for bridging data interfaces between IT systems, operations systems, energy, and cooling systems.

In terms of ecosystem, cloud partners such as CoreWeave, Crusoe, and Lambda are deploying DSX Sim, MaxLPS, and DSX OS to reduce risks and improve GPU utilization. Manufacturers such as Dell, HPE, Lenovo, Supermicro, as well as ASUS, Foxconn, Gigabyte, and Quanta Cloud Technology, are building systems supporting DSX.

Aligning with Windows and ARM

During the live speech, Huang officially announced the debut of the "DGX Station for Windows" workstation, defined by NVIDIA as a desktop-level AI supercomputer for the Windows ecosystem.

Hardware-wise, it is equipped with the GB300 Grace Blackwell Ultra Desktop Superchip, connecting the Blackwell Ultra GPU with a 72-core Grace CPU via NVLink-C2C, offering up to 748GB of unified memory and 20 PFLOPS FP4 performance, along with network capabilities of up to 800Gb/s.

The focus of this product lies in the change in Agent deployment methods.

NVIDIA hopes enterprises can run multiple Agents in a local, secure, and manageable Windows environment, integrating them into workflows for design, engineering, data science, inference, and Physical AI. Simultaneously launched, OpenShell is responsible for Agent runtime security, limiting unauthorized operations or leakage of credentials and private data through isolated sandboxes and system-level policy controls.

In addition to products for enterprise desktops, Huang also unveiled a system-level SoC at the conference—the RTX Spark SoC, which integrates the N1X CPU and Blackwell GPU onto a single chip with a unified memory architecture, specifically for thin-and-light laptops and small desktops.

Among them, N1X is the first PC processor jointly created by NVIDIA and Microsoft. It is based on the Arm architecture, custom-designed by MediaTek, and manufactured using TSMC's 3nm process. It will debut this autumn in laptops from Microsoft, Dell, HP, ASUS, Lenovo, and MSI, with over 30 models in the first batch, focusing on high-end thin-and-light laptops.

This is the "super chip" NVIDIA has prepared for the AI PC era, which Huang views as an important reconstruction of the PC form factor.

The "Two Brains" of Agents

At this launch event, NVIDIA announced the latest progress in two core model product lines, corresponding to two scenarios for Agents: one running within enterprise systems and the other in the physical world.

NVIDIA released Nemotron 3 Ultra, a Mixture-of-Experts model with 550 billion parameters, providing top-tier intelligence capabilities for code development, scientific research, and long-term agents in enterprise business processes. Compared to mainstream open-source frontier models of the same class, this model increases inference speed by up to 5 times and reduces usage costs by up to 30%, helping agents complete various tasks more efficiently and at lower cost.

Centered around the Nemotron open model, NVIDIA released a series of software, open-source models, and collaboration updates, aiming to enable enterprises to build "digital colleagues" that assist employees in scenarios such as engineering design, healthcare, software development, and business operations.

In this combination, Nemotron provides foundational model capabilities, NemoClaw organizes the model into an Agent, OpenShell handles runtime security, and the Agent Toolkit turns NVIDIA software libraries like CUDA-X into tools that Agents can directly invoke. Agents can use tools, call data, execute tasks, and integrate with existing enterprise systems in a controlled environment.

Huang stated that global software companies are bringing AI Agents into real work systems to help employees complete complex tasks faster. NemoClaw provides the open components needed to build long-running Agents, including capabilities for orchestration, context, memory, tool invocation, and security control.

In the past, when enterprises discussed AI, the focus was more on what models could answer; now, NVIDIA aims to solve how Agents can securely access tools, data, and business processes, and operate continuously in real work.

Then there is Cosmos 3, the official third generation of the Cosmos series, which is also an architectural reconstruction.

Cosmos 3 is a world foundation model for Physical AI, providing the underlying capabilities to "understand the physical world, predict what will happen, and decide what to do."

Compared to previous versions of Cosmos, which mainly targeted robot and autonomous driving developers with video generation and physical world simulation—essentially a relatively unimodal generative framework—Cosmos 3 adopts a new architecture—a hybrid Transformer—that unifies visual reasoning, world generation, and action prediction into a single system for the first time.

It can natively understand and generate text, images, video, ambient sound, and actions, achieving leading levels of physical accuracy, and is the world's first fully open omnimodal model. NVIDIA claims it has the potential to compress the training and evaluation cycle for Physical AI from months to days.

Huang predicted that thanks to breakthroughs in multimodal reasoning languages, vision, and world models, the explosion of Physical AI is imminent.

The open frontier omnimodal models in the Cosmos 3 series provide developers with generational leap capabilities for building robots, autonomous vehicles, and visual AI that can perceive, reason, plan, and act in the physical world.

Lowering the Barrier to Physical AI

NVIDIA and Unitree jointly released H2 Plus—a humanoid robot prototype machine prepared for researchers and developers.

The meaning of "prototype" here is: Unitree is responsible for the robot body, while NVIDIA handles the software and computing platform. Both sides pre-integrate hardware and software so that development teams can start skill development immediately upon receipt, without spending time solving underlying integration issues. It is also the world's first open humanoid robot built on the NVIDIA Isaac GR00T development platform.

This prototype targets a long-standing pain point in humanoid robot development: hardware integration, data collection, simulation, training, evaluation, and deployment are often siloed, making the entire process highly fragmented.

NVIDIA stated that research teams often spend a significant amount of time piecing together underlying layers when they receive a robot body, pushing back actual skill development repeatedly. What H2 Plus attempts to do is clear this path, allowing research teams to skip underlying integration and proceed directly to skill development and real-scenario validation.

In Huang's view, humanoid robots will bring Physical AI to the world's largest industries, opening up trillions of dollars in economic opportunities, and H2 Plus is the starting point for pushing frontier research into real scenarios such as factories, warehouses, and logistics systems.

Additionally, NVIDIA announced the official open-sourcing of a Physical AI Skills toolkit, covering core scenarios such as robotics, autonomous driving, visual AI, and industrial digital twins.

So-called "Skills" can be understood as NVIDIA standardizing the usage methods of its platforms such as Cosmos, Omniverse, Isaac, and Metropolis, writing them into operation instructions that agents can directly read and execute. Packaging these instructions for open source constitutes the toolkit released this time.

When an agent receives a task, such as generating a batch of training data for defect detection, it knows which model to call, what format to output, and how to verify the results, completing the entire process automatically without requiring human step-by-step operation of each link.

Upgrading AI Storage: From "Running Fast" to "Managing Securely"

At the GTC in San Jose in March, NVIDIA released the Vera BlueField-4 STX, where Huang focused on the "AI-native storage architecture," with the core selling point being high-performance KV Cache storage support for long-context reasoning by agents.

Now, based on STX, NVIDIA has announced the addition of a set of security capabilities, shifting the focus from "storage performance" to "storage security."

The core logic and thinking here stem from the changing way enterprises use AI. Many enterprises are now actively deploying agents. When Agents access enterprise systems and continuously read/write and share information across systems without direct human supervision—who is accessing what data, whether there is unauthorized access, or whether there is leakage—these are headaches for enterprises.

NVIDIA's solution is to add a layer of security capability on top of accelerated storage—relying on a unified NVIDIA DOCA security software suite and directly enforcing policies in hardware within the BlueField-4 chip. Platforms based on STX can real-time check and manage interactions between agents, data, and contextual memory, helping enterprises achieve continuous policy enforcement on the AI data path.

Huang explained: "Agents turn enterprise data into a real-time, living system, and this system must be protected wherever data moves, wherever context is stored, and wherever agents act. What Vera BlueField-4 STX aims to do is enforce trust at AI speeds within the chip through inherently secure design."

"Mutual Suppliers" with TSMC

A very interesting point in this conference was the cooperation between NVIDIA and TSMC—currently, TSMC is using NVIDIA technology to improve turnaround time, energy efficiency, yield, and operational productivity in advanced wafer fabs.

The relationship between TSMC and NVIDIA has had only one form for the past thirty years: TSMC manufactures chips for NVIDIA. But now, the roles have subtly changed, with NVIDIA beginning to help TSMC "manage the factory."

Huang stated: "NVIDIA and TSMC have cooperated for nearly thirty years, constantly pushing the limits of computing. TSMC is introducing NVIDIA's AI and accelerated computing into its wafer fabs, using simulation, optimization, and AI to address the world's most complex design and manufacturing challenges, thereby improving the speed, efficiency, and yield of next-generation chips."

The relationship between the two has evolved from a unilateral client-vendor dynamic to a bidirectional mutual dependence.

Conclusion

Looking back at this launch event, NVIDIA is piecing together a new blueprint around "Agents."

Vera CPU schedules tasks for Agents, Vera Rubin provides computing power for Agents, BlueField-4 STX guards data security for Agents, Cosmos 3 enables Agents to understand the physical world, Nemotron+NemoClaw+OpenShell allows Agents to be organized, invoked, and constrained, DGX Station for Windows brings Agents to enterprise employees' desktops, H2 Plus gives Agents a body, and DSX and Skills allow all of this to be mass-produced and mass-deployed.

From this perspective, what Huang tries to depict is a new computing era. This also echoes his opening remark that "the era of Agent AI and practical artificial intelligence has arrived."

Ultimately, the one thing Huang wanted to convey this time is: when Agents become AI infrastructure, NVIDIA can be present at every layer.

Risk Warning and Disclaimer

The market carries risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this content is at the user's own risk.