Nvidia's stock is sitting approximately 30 percent below the consensus analyst price target of $298.42 as of this week, and the company is simultaneously unveiling what may be the most significant expansion of its business model since Jensen Huang decided to focus the entire company on AI accelerators in 2016.
Nvidia spent the first half of 2026 announcing robotics infrastructure that changes what the company is.
In March at its annual GTC conference, Huang told a packed auditorium in San Jose: "Physical AI has arrived, every industrial company will become a robotics company." On May 31, Nvidia released Cosmos 3, the company's first fully open physical AI model, and followed it on June 1 at GTC Taipei with the Isaac GR00T Reference Humanoid Robot, an open reference design for a complete humanoid robot platform built on a Unitree chassis and Nvidia's Jetson compute module.
Also at GTC Taipei, Huang unveiled the RTX Spark Superchip, a system-on-chip designed to take Nvidia's AI computing into Windows laptops, competing directly in a PC market that AMD, Intel and Qualcomm had been dividing between themselves.
The data center business remains Nvidia's engine. The robotics and PC businesses are the next cylinders firing.
The stock that has powered the AI investment narrative for three years is now trading at levels that Goldman Sachs, which reiterated Buy on the PC chip announcement, would describe as approximately 30 percent undervalued.
Jensen Huang looked at the recent weakness in tech stocks and said:
"Whatever happened to the stock market, you should be very happy because now you can buy at a discount."
What Physical AI Actually Means
The phrase physical AI is Nvidia's framing for the specific problem that comes after training large language models on text and images, the problem of making AI work in the physical world with physical objects.
An AI that can write code or generate images is useful. An AI that can control a robotic arm, navigate a warehouse, perform surgery or drive a car requires something qualitatively different, understanding of physics, spatial relationships, cause and effect in three dimensions, the ability to handle uncertainty in real environments rather than controlled datasets.
Nvidia's thesis is that the same stack that dominates AI training, GPUs for compute, CUDA for software, an ecosystem of models and frameworks, can be extended to dominate physical AI as well.
The tools it has been building to that end are the Cosmos world models and the Isaac simulation frameworks. Cosmos generates synthetic data, realistic simulations of physical environments, that robots can train on without the cost and time of real-world data collection.
Isaac Sim provides the physics engine that makes those simulations accurate enough to transfer to real robots. GR00T is the robot foundation model that learns from both simulated and real-world data and produces robots that can generalize across tasks.
The ecosystem Nvidia has assembled around this stack is the clearest evidence that the robotics ambition is serious rather than aspirational.
ABB, FANUC, KUKA and Yaskawa are four of the largest industrial robot manufacturers on the planet. Agility, Figure and NEURA Robotics are among the leading humanoid robot developers.
CMR Surgical and Johnson & Johnson MedTech are using Nvidia's simulation tools to train and validate surgical robots before clinical deployment. This is not a developer conference demo. These are the companies that make the actual robots.
The Open Reference Humanoid That Just Got Released
The Isaac GR00T Reference Humanoid Robot announcement on June 1 is the detail that the robotics community is most closely examining.
Nvidia published an open reference design, available on Hugging Face under a commercial license, for a complete humanoid robot that combines a Unitree H2 Plus chassis standing approximately six feet tall and weighing 150 pounds with 75 degrees of freedom, Sharpa Wave five-finger tactile hands and an Nvidia Jetson AGX Thor T5000 compute module delivering 2,070 FP4 teraflops of onboard AI processing.
The significance of making it open is deliberate and strategic. Nvidia is not in the business of selling robots.
Nvidia is in the business of selling the compute and software that makes robots intelligent. By publishing an open reference design, Nvidia gives every robotics developer a standardized hardware platform to build on, which means every robotics developer building on that platform is a customer for the compute modules that run the intelligence.
The ecosystem strategy that made CUDA the dominant AI programming framework by making it free for developers is being applied to the physical hardware layer of the robotics stack.
The GR00T N1.7 model that powers the reference humanoid is now available in early access with commercial licensing, meaning companies can deploy it in production robots today.
GR00T N2, the next generation, has been previewed and ranks first on the robotics evaluation benchmarks that the community uses to measure generalist robot policies.
The PC Chip That Sent Three Competitors Lower
The robotics story is the longer-term Nvidia narrative. The more immediate market catalyst from the past two weeks is the RTX Spark Superchip that Huang unveiled at Computex in Taipei on June 1, a system-on-chip designed to run AI workloads locally on Windows laptops, in partnership with Microsoft.
Huang's framing was specific: Nvidia and Microsoft are going to "reinvent the PC." The RTX Spark is an Arm-based chip that can run AI agents locally, models that process information and execute tasks on the device rather than sending every request to a cloud server. "Look how beautiful it is, this agent could run 24/7, meter free," Huang said, holding up an MSI laptop built around the chip. "No meter anxiety."
AMD, Intel and Qualcomm stocks all fell on the news. Goldman Sachs analyst James Schneider reiterated his Buy rating on Nvidia specifically because of the PC chip, writing that the combination of the new PC chip and Nvidia's AI infrastructure leadership could drive significant momentum.
The Windows on Arm platform, which Microsoft has been trying to advance for years, finally has the Nvidia brand and the Nvidia software ecosystem behind it, which Schneider described as the credibility it has been lacking.
The Stock And What The Market Is Weighing
The tension between what Nvidia is building and where the stock sits is the specific question the Barron's article addresses.
At approximately $205 with an analyst consensus target of $298, Nvidia is trading at levels that imply either that the analyst community is significantly wrong about the company's trajectory or that the stock is genuinely undervalued relative to the business being built.
The bear case, articulated by the investor quoted in Yahoo Finance last week who noted that the stock fell after strong earnings, is that the AI industry is expanding, competition is entering and yesterday's picks-and-shovels winner may not be tomorrow's.
Custom silicon from Microsoft, Google and Amazon reduces hyperscaler dependence on Nvidia at the margins. Competitors like AMD and Cerebras are taking share in specific workloads.
The transition from AI model training to AI inference may favor different economics than the ones that made H100s the most valuable hardware in the world.
The bull case is that Nvidia is not standing still while this happens. The robotics platform, the PC chip, the agentic AI infrastructure, the open models, all of it is the next layer of a company that has built a software moat around its hardware that no competitor has yet reproduced.
The GR00T ecosystem alone has 110 robot brain developers and partnerships with the largest industrial automation companies on the planet.
Jensen Huang told investors the stock pullback is a buying opportunity. The analyst consensus says he is approximately right. What the next chapter of the AI era looks like, and who supplies the compute for it, is the question the rest of 2026 will answer.