Citibank: Data remains the biggest bottleneck for the commercialization of embodied intelligence; the RaaS model may be the key to breaking the deadlock

```

Embodied intelligence is moving from proof of concept to large-scale implementation, but this path is far more complex than digital AI.

After its annual Robotics and Physical AI Leadership Summit, Citi Research released a report, stating that labor shortages, manufacturing reshoring, and regulatory easing are accelerating enterprise demand, while data scarcity, talent bottlenecks, battery endurance, and high deployment costs remain the core friction points restricting industry expansion.

Analyst Heath Terry believes that the ultimate winners will be companies that possess proprietary real-world data, address specific labor bottlenecks, and reduce upfront costs for customers by adopting the “Robot-as-a-Service” (RaaS) model.

From Concept to Deployment: The Industry Turning Point Has Arrived

The Citi summit brought together founders, investors, operators, and industry executives in robotics. The main conclusion: Physical AI is at a critical juncture, transitioning from proof of concept to commercial deployment, but its path to scale is far more operationally challenging than digital AI.

Unlike large language models, the core value of physical AI lies not in general foundational models, but in proprietary task-level data collected in real-world environments, customized hardware, and safety certification systems. At the summit, Instawork pointed out that even if industry data accumulated by 2026 reaches tens of millions of hours, compared to the total required for high-level robot performance, it is only a "basis point" difference rather than a "percentage point" difference.

Meanwhile, compute architecture also faces bottlenecks. Attendees noted that existing semiconductor platforms are mainly designed for data center workloads and do not fit real-time edge inference needs for mobile platforms; power management and chip architecture compatibility are emerging as new technical hurdles.

RaaS Model: The Key to Unlocking the SME Market

On the business model front, RaaS (Robot-as-a-Service) is the key mechanism driving rapid industry penetration. By converting upfront capital expenditure into subscription or pay-per-use models, RaaS significantly lowers the procurement threshold for small and medium-sized enterprises.

The report specifically mentions Symbotic’s “warehouse-as-a-service” products (GreenBox/Exol), stating that this model helps expand warehouse automation solutions to a wider customer base. Many summit guests also emphasized that companies with the fastest commercial progress—whether in humanoid robots, warehouse AMRs, autonomous trucks, or construction robots—share common traits: they tackle specific high-pain labor problems, adopt RaaS to reduce procurement barriers, and prioritize safety and reliability over model complexity.

It’s worth noting that despite humanoid robots attracting significant investment hype, recent visible returns mainly come from specialized AMRs and vertical scenario systems by companies like Locus Robotics and Dexterity, rather than generic humanoid robots.

$20 Billion of Capital Inflow, Industrial Beneficiaries Expected

Capital markets are increasing their bets on physical AI. Over the past two years, the global physical AI sector has attracted approximately $20 billion in investments, with application scenarios spanning warehousing, logistics, trucking, construction, aviation, and defense.

On actual use cases, BMW recently disclosed that upgraded humanoid robots are walking and working on production lines at its Spartanburg plant in South Carolina, marking the entry of humanoid robots into mainstream manufacturing.

The Citi industrial team believes that continued advancement in automation, robotics, and physical AI will provide long-lasting and sustainable growth drivers for industrial companies with high automation exposure. The core factors driving automation adoption include a persistently tight labor market, accelerating manufacturing activity, and capacity expansion. Automation’s value in improving capacity utilization, increasing uptime, and boosting operational efficiency has been validated, supporting healthy investment returns.

Advances in AI and large language models, coupled with increasingly rich real-world and simulation data, are pushing robotics toward higher integration—deep fusion of hardware and software, and a positive flywheel effect where systems continuously "get smarter" in use. For companies with a large installed base, data accumulation itself becomes a significant competitive moat.

Ten-Year Marathon: The Track for Patient Capital

The report’s conclusion is clear: physical AI is a decade-long construction endeavor, whose lasting value will concentrate on companies that master the data flywheel, solve real deployment issues, and meet the highest safety standards.

"Physical AI may ultimately benefit from the laws of scale, but this path will be far slower and more operationally challenging than the chatbot boom." Terry wrote in the report.

This judgment means investors need to approach physical AI with a longer time frame and stronger fundamental screening ability—short-term catalysts are limited, but structural opportunities are gradually becoming clear.

Risk Disclosure and DisclaimerThe market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the individual investment goals, financial status, or needs of specific users. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their specific situation. Investments made based on this information are at one's own risk. ```