The race to dominate physical AI is no longer about building the most impressive robot; it's about controlling the invisible infrastructure that makes those robots actually work. Three major developments this week reveal a fundamental shift in how companies are positioning themselves in the embodied AI market: Alibaba is launching a full-stack robotics model, Nebius is acquiring AI optimization software and launching a robotics incubator, and New York is getting its first pop-up store dedicated to humanoid robots. Together, these moves signal that the real competitive advantage in physical AI lies not in the mechanical systems themselves, but in the software, computing infrastructure, and deployment tools that bring those systems to life.

What Is the Physical AI Stack, and Why Does It Matter?

Physical AI refers to artificial intelligence systems embedded in robots, autonomous vehicles, industrial equipment, and other machines capable of sensing, interpreting, and responding to real-world environments. Unlike traditional AI that lives in the cloud and processes text or images, physical AI must operate in unpredictable, constantly changing physical spaces. This requires a fundamentally different approach to how AI models are built, optimized, and deployed.

The "stack" is the layered architecture that makes this possible. At the bottom sits the hardware: graphics processing units (GPUs) and other computing chips that provide raw processing power. Above that sits the software layer, where AI models are optimized to run efficiently. At the top sits the application layer, where robotics companies build their actual products. For years, the focus has been on the bottom layer. Now, the value is moving upward.

Why Is AI Inference Becoming the New Battleground?

Training AI models has dominated headlines for the past two years, but a less visible challenge is becoming equally important: running those models efficiently once they are deployed. This stage is called inference, and it happens every time a user interacts with a chatbot, generates an image, or submits a request to an AI application. For robotics, inference is critical because robots must make decisions in real time, often with limited computing resources.

Faster and more efficient inference can reduce infrastructure costs while improving performance. This is why Nebius, a European AI infrastructure company, just completed a $643 million acquisition of Eigen AI, a California-based company specializing in model optimization and inference technologies. Eigen AI was founded by researchers from the Massachusetts Institute of Technology (MIT) and has developed techniques that are now widely adopted across the AI industry for running models more efficiently.

"The acquisition gives Nebius far more than a software product. It brings a team of researchers whose work already underpins many modern AI deployment techniques," the company noted in its announcement.

Nebius Group

This move reflects a broader industry trend: as AI systems move from experimentation into production environments, the economics of inference become increasingly important. Companies that can optimize how AI models run in the real world will have a significant competitive advantage over those that simply build better models.

How Are Companies Building the Physical AI Infrastructure?

• Full-Stack Integration: Alibaba is launching Qwen-Robot, an embodied AI foundation model designed to bridge large-scale multimodal AI with real-world robotics across industrial and commercial sectors. The model is expected to unlock high-margin, recurring revenue streams through managed AI services (MaaS), potentially adding $2.5 billion in incremental revenue for the company.
• Optimization and Deployment: Nebius is integrating Eigen AI's inference optimization technology directly into its Token Factory managed inference platform, allowing robotics companies to deploy AI models more efficiently and cost-effectively.
• Developer Support and Simulation: Nebius and Nvidia have launched the Physical AI Living Lab, a six-month program designed to support robotics start-ups in Britain and Europe. The program provides access to Nvidia technologies, Nebius cloud infrastructure, and engineering support. Participants will work with engineers from both companies to develop synthetic data, create simulations, test behavior in virtual environments, and evaluate performance under a wide range of operating conditions.

What Does This Mean for Robotics Companies?

For early-stage robotics companies, access to computing resources and specialized software tools has historically been a significant barrier to development. Training modern AI systems requires substantial computing resources, and as robotics developers incorporate increasingly sophisticated AI models into their systems, this challenge has become more pressing.

The Physical AI Living Lab addresses this directly. The first cohort is expected to begin in September, with participants receiving access to computing infrastructure, simulation tools, and expert guidance from both Nebius and Nvidia engineers. This represents a shift in how the industry supports robotics development, moving beyond individual start-up funding toward building shared infrastructure and tools.

Meanwhile, in New York, the opening of the first pop-up store dedicated to general-purpose robots signals that the market is moving toward consumer awareness and early adoption. The store, a partnership between KraneShares and OpenMind, will allow visitors to interact with advanced robots integrated with OpenMind software and place pre-orders for humanoid robots. The storefront opens to the public on June 26 through June 28 at 188 Lafayette Street in SoHo.

How Is Europe Positioning Itself in the Physical AI Race?

Europe possesses many of the ingredients required to build competitive robotics companies. European universities continue to produce influential robotics research, while countries including Germany, the Netherlands, France, and Sweden maintain strong industrial engineering traditions. However, the continent has historically struggled with commercialization and scale.

The Physical AI Living Lab represents an effort to strengthen the supporting infrastructure available to European robotics companies rather than focusing solely on individual start-ups. This approach acknowledges that the bottleneck in robotics development is no longer just innovation; it's access to the computing resources and specialized tools needed to bring innovations to market.

European policymakers have also become more vocal about the strategic importance of AI infrastructure, semiconductors, and advanced computing capabilities. Physical AI sits at the intersection of all three, making it a priority for governments seeking to build technological resilience and reduce dependence on foreign suppliers.

What Are the Financial Implications?

For investors, the shift toward full-stack AI infrastructure has significant implications. Alibaba's Qwen-Robot launch is expected to support a 20 to 40 percent stock upside at 13 to 15 times forward price-to-earnings (P/E) ratio, according to analyst estimates. The model could add $2.5 billion in incremental revenue, supporting higher valuations for the company as it transitions from a general-purpose cloud provider to a specialized AI infrastructure and robotics platform.

Nebius' acquisition of Eigen AI and launch of the Physical AI Living Lab signal that the company is evolving from a so-called "neocloud" provider that rents computing capacity into a more integrated AI platform combining infrastructure, optimization, and deployment capabilities. This vertical integration strategy allows Nebius to capture more value across the AI stack, from raw computing power to optimized model deployment to robotics development support.

The broader trend is clear: companies that control multiple layers of the physical AI stack will have significant competitive advantages over those that focus on a single layer. This is why Alibaba is building end-to-end robotics models, why Nebius is acquiring optimization software, and why Nvidia remains central to the ecosystem despite competition from other chip makers.

What Challenges Remain?

Despite the momentum, significant challenges remain. Commercialization delays, intense domestic competition, and macroeconomic headwinds could slow adoption. Additionally, many infrastructure providers remain heavily dependent on Nvidia hardware and software, limiting the number of alternative ecosystems available to developers. This concentration raises concerns about technological resilience as governments and companies seek greater independence from single suppliers.

The physical AI market is at an inflection point. The companies that win will not necessarily be those that build the most impressive robots, but those that control the infrastructure, software, and tools that make those robots actually work in the real world.