Barclays moves mass humanoid commercialisation to 2035, five years later than 2030

Barclays has moved its expected date for mass commercialisation of general-purpose humanoid robots to around 2035, five years later than 2030, in a research note dated September 18, 2026, according to Korean coverage by Newspim (September 21) and Robot Newspaper (September 26). The note argues that AI capability, not hardware, is holding the robots back.

Why it matters: the delay comes from a bank that published a $200 billion 2035 market scenario in January, and it lands the same month Goldman Sachs raised its own forecast. Two large banks now read the same evidence in opposite directions.

Why did Barclays delay its humanoid timeline to 2035?

Barclays says the robots are not yet intelligent enough. According to Robot Newspaper’s summary, the note finds that current humanoid demonstrations lean heavily on pre-programmed motion, remote operation and narrow automation scenarios, rather than robots that perceive, reason and act on their own. The bank locates the bottleneck in AI models rather than in actuators, batteries or manufacturing. Newspim describes the new timing as mass deployment “around 2035” against the 2030 date the market had expected.

That matches what the first hard data shows. The International Federation of Robotics counted about 7,000 humanoids sold in 2025, many of them bought for research or data collection, as we reported in IFR’s first humanoid count.

Will humanoid robots drive data centre demand?

Less than bulls expect, according to Barclays. The note argues that the real-time loop of perception, reasoning and action will run on edge processors inside each robot, because latency and reliability rule out a round trip to the cloud. Data centres keep a role in simulation, synthetic data generation and foundation model training, but not in live operation. For investors, that shifts the humanoid compute story from GPU clusters toward on-robot chips.

The training side is still large. Newspim, in the same article, points to Figure AI’s compute agreement with Crusoe announced September 3: an initial $3.5 billion allocation, expandable beyond $6 billion, for up to 100,000 NVIDIA Blackwell GPUs from the second half of 2027. Barclays’ argument is about where each robot thinks, not whether model makers will buy compute.

Does a robot need to be humanoid?

Barclays questions it. The note points to Amazon, which runs more than 1 million robots in its logistics network, built for specific tasks rather than in human form. The bank accepts the two standard arguments for humanoids: they fit buildings, stairs, doors and tools made for people, and one machine can do many jobs. But its reading is that specialised robots will be deployed first and at scale, and humanoids later.

How does Barclays compare with Goldman Sachs?

The two banks moved in opposite directions within a week. Goldman Sachs, in coverage published September 13–14, raised its forecast to about 6.5 million humanoid units and $138 billion in 2035, with 890,000 units in 2030, as detailed in our report on the Goldman upgrade. Barclays’ own January 14, 2026 release sized the humanoid market at $2–3 billion today and up to $200 billion by 2035 in optimistic scenarios. The September note does not, in the coverage we have seen, withdraw that figure; it moves the date at which general-purpose robots reach volume.

Barclays’ own research is not uniform. A separate Barclays chemicals analysis, by analyst Katie Richards and cited by Humanoids Daily on September 25, projected about 10,000 tonnes of PEEK plastic demand from 1 million humanoids by 2030. Yu Fei, robotics technical general manager at Chinese materials maker Kingfa, told Humanoids Daily that most current PEEK use in embodied AI is “hype, though not absolutely”, and said today’s humanoids use 6–10 kg of engineering plastics in total, against the roughly 10 kg of PEEK alone that the Barclays figure implies. A 1-million-unit 2030 supplier case sits awkwardly next to a 2035 commercialisation date.

Our view: the 2035 call is the better-evidenced of the two. It rests on what robots do in deployments today, while the bullish forecasts rest on cost curves and scenario modelling. Our tracker of humanoid forecasts against shipments will show which one the next year of data supports.