NVIDIA is betting that the next major computing wave will be robots and autonomous machines, not just cloud-based AI systems. The company has declared a "ChatGPT moment for physical AI" and framed humanoids and labor automation as a potential $40 trillion market opportunity. With Amazon expanding its GPU partnership and adopting NVIDIA's full robotics stack, the infrastructure shift appears to be accelerating faster than many investors anticipated.

What Is Physical AI, and Why Does It Matter?

Physical AI refers to robots and machines that can perceive their surroundings, handle objects, and make decisions independently. Unlike traditional AI systems that live in data centers and process text or images, physical AI must operate in the real world, making it fundamentally different from the large language models (LLMs) that dominated AI headlines over the past two years. This shift represents a new frontier for computing hardware and software, one that NVIDIA believes will dwarf the current AI boom.

The market opportunity is enormous. CEO Jensen Huang stated that humanoids and labor automation represent a $40 trillion addressable market, roughly equivalent to the entire global economy. If even a fraction of that materializes, it would justify NVIDIA's aggressive push into robotics infrastructure.

How Is NVIDIA Building Its Physical AI Ecosystem?

NVIDIA's strategy mirrors the approach that made it dominant in data center AI. Rather than selling individual components, the company is bundling an entire ecosystem designed to handle every step of the robotics pipeline:

• Training Infrastructure: AI factories and data centers that train the foundational models robots will use to understand and interact with the world.
• Simulation and Design: Omniverse and Cosmos platforms that let engineers simulate physical environments and test robot behavior before deployment.
• Robot Foundation Models: GR00T, a general-purpose foundation model specifically designed to teach robots how to perform tasks and adapt to new situations.
• Edge Compute: Jetson chips that embed intelligence directly inside robots, allowing them to make decisions locally without constant cloud connectivity.

This full-stack approach proved highly effective in the data center, where NVIDIA's CUDA programming framework became so entrenched that switching to competitors became prohibitively expensive. The company is now attempting to replicate that lock-in effect in robotics.

Why Amazon's GPU Expansion Signals a Turning Point?

Amazon's August 2026 announcement that it would deploy an additional 2 million graphics processing units (GPUs) from NVIDIA's Blackwell, Rubin, and Rubin Ultra generations over 2027 and 2028 represents a significant validation of NVIDIA's physical AI strategy. What makes this particularly noteworthy is that Amazon Robotics, the company's own robotics division, chose to adopt NVIDIA's full-stack physical AI platform spanning Jetson, Omniverse, Isaac, and Cosmos for developing and training robots.

This decision is striking because Amazon has been actively building its own custom AI chips, including Trainium processors, specifically to reduce its reliance on NVIDIA. Yet for the foundational layer of robotics work, Amazon determined that NVIDIA's ecosystem was the better choice. Even more telling, Amazon decided to use NVIDIA's NVLink technology for the scale-up network at the heart of its data center connectivity, pulling even custom silicon into NVIDIA's ecosystem.

"For forty years, you launched apps. Click. Type. With RTX Spark and Microsoft Windows, you ask, and the PC does the work," said Jensen Huang, founder and CEO of NVIDIA.

Jensen Huang, Founder and CEO, NVIDIA

What Does This Mean for NVIDIA's Valuation?

From an investment perspective, NVIDIA's valuation metrics remain compelling despite the stock's significant run-up. The company trades at 18.7 times forward 12-month earnings, below the S&P 500's 19.9 times multiple, while its projected annual earnings-per-share growth of 52.7% for 2026 through 2028 far outpaces the broader index's 20.1% growth rate. This suggests the market may still be underpricing the durability of NVIDIA's earnings cycle as physical AI scales.

Analysts point to NVIDIA's 0.48x price-to-earnings-to-growth (PEG) ratio as particularly attractive, indicating that the company's growth rate justifies its current valuation even after accounting for its premium stock price. In Q2 2027, NVIDIA's revenue surged 106 percent year-over-year to $96.2 billion, with Q3 guidance of $108 billion in revenue and a 74 percent non-GAAP gross margin, demonstrating the scale and profitability of its current business.

What Are the Risks to NVIDIA's Physical AI Thesis?

Despite the bullish outlook, several risks could derail NVIDIA's dominance in physical AI. Supply constraints remain a persistent challenge, as demand for advanced GPUs continues to outpace production capacity. Custom AI silicon from competitors like Amazon, Google, and Meta could eventually erode NVIDIA's market share if those companies successfully reduce their dependence on NVIDIA chips. Geopolitical restrictions, particularly around semiconductor exports to China, could also limit NVIDIA's addressable market and growth trajectory.

However, NVIDIA's full-stack strategy and the stickiness of CUDA create substantial switching costs that make it difficult for customers to migrate away, even if they develop custom silicon. The company's ability to integrate custom chips into its broader ecosystem through technologies like NVLink Fusion means that wider adoption of custom processors may not translate directly into lost revenue for NVIDIA.

When Will Physical AI Hardware Reach Consumers?

The shift toward local, on-device AI is already beginning to materialize in consumer hardware. Acer announced the SFF RTX Spark, a compact desktop built around NVIDIA's RTX Spark superchip designed to run local agentic AI workloads. The system pairs up to a 6,144-core Blackwell RTX GPU with up to a 20-core NVIDIA Grace CPU, offering up to 1 petaflop of AI compute and up to 128 gigabytes of high-speed unified memory.

This hardware can run large language models with up to 120 billion parameters and context windows of up to 1 million tokens, meaning it can process roughly 100,000 words at once. RTX Spark laptops and compact desktops will be available in fall 2026 from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI, with models from Acer and GIGABYTE following after the initial launch window.

The broader ecosystem is also mobilizing around this platform. More than 100 Windows software providers and game developers are embracing RTX Spark, including Adobe, which is rearchitecting Photoshop and Premiere to deliver up to 2x faster AI, editing, coloring, and effects on the platform. Game developers like Riot Games, Remedy Entertainment, and XBOX are also optimizing their titles for the new hardware.

As physical AI extends beyond the data center into real physical space, NVIDIA's position as the foundational infrastructure provider appears unlikely to prove a passing benefit. The company's ability to bundle hardware, software, and ecosystem partnerships into a cohesive platform suggests that its dominance in AI infrastructure may simply be shifting from one domain to another, rather than diminishing.