Barclays Questions the Humanoid Robot Hype With Three Hard Truths

Global financial institution Barclays, headquartered in London, has pushed back its forecast for the mass commercialization of general-purpose humanoid robots by five years, now projecting the milestone will arrive around 2035, as reported by 로봇신문 (Robot Newspaper), citing outlets including AllWeatherMedia.

In a research note released on the 18th, Barclays laid out three central debates surrounding the humanoid robotics industry: the timing of mass commercialization, the real scale of data center demand growth tied to humanoid deployment, and whether the humanoid form itself is the best vehicle for physical AI.

Commercialization timing pushed back

Barclays observed that many of the humanoid demonstrations currently shown by robotics companies rely heavily on pre-programmed routines, teleoperation, or narrowly scripted automation — far from what could be called true general-purpose autonomy.

The bank attributed the delay less to hardware limitations than to the state of the underlying intelligence. AI models capable of the perception, reasoning, and action needed for humanoids, it argued, are not yet mature enough. Barclays described a three-way bottleneck: scale is needed to bring costs down, but sufficient intelligence is required to prove commercial value, and achieving scale, cost efficiency, and intelligence simultaneously remains extremely difficult. The bank compared this trajectory to autonomous driving, which took more than a decade to reach commercial deployment. As a result, Barclays suggested the earliest investment opportunities may lie not in finished robots but in the underlying infrastructure — computing, data, and models — that could eventually trigger a "GPT moment" for humanoids.

Edge computing, not hyperscale data centers

On the debate over computing demand, Barclays argued that real-time inference — a robot perceiving, judging, and acting on its surroundings — will largely have to run on dedicated onboard edge processors due to constraints around latency, power consumption, reliability, and security. This means, the bank said, that inference workloads won't directly translate into massive new data center demand.

However, Barclays noted that data center compute will still be needed for simulation, synthetic data generation, and the training and post-training of foundation models — meaning computing demand is likely to rise ahead of any large-scale robot deployment, not necessarily because of it.

Is the humanoid form even optimal?

Barclays also expressed skepticism about whether humanoids represent the best form factor for physical AI. The bank suggested the first major disruptive breakthrough in physical AI may not come from humanoids at all, pointing to collaborative robots, autonomous mobile robots (AMRs), AI-powered drones, and quadrupedal robots that are already expanding their influence across commercial, industrial, and defense applications.

As a concrete example, Barclays cited Amazon, which operates more than one million robots across its warehouses — the majority of which, the bank noted, are task-specific machines rather than humanoids.

Still, Barclays acknowledged real advantages to the humanoid form. Because factories and warehouses were built around human bodies — with stairs, doors, handles, workbenches, and tools designed for human use — humanoid robots could potentially be deployed without major infrastructure changes.

The bank also pointed to versatility as a key strength: unlike dedicated robots built for a single task, a humanoid capable of learning multiple tasks could offer a new way to address labor shortages across industries.