As AI unlocks new capabilities, firms and investors are asking: Where is automation actually creating value? With available labor supply at a premium and labor costs rising, more firms are leaning on robotics and automation to ease the pressure. That was the focus of a recent roundtable discussion we hosted with two leading experts in the field:
- Gavin Curran, former COO at Astrak Global and former director at GXO Logistics
- Helen Fang, Head of Industrials Research, Asia-Pacific at HSBC
Below, we recap the key takeaways from that discussion, covering humanoid demand, hardware costs, automation returns, software and data, and geopolitics.
Humanoid Robots on the Factory Floor
Helen Fang said she initially viewed humanoid robots as secondary to industrial robots. But on a recent visit to suppliers at the World Robotics Conference in Beijing, she saw signs of massive demand: As a result, Fang estimates that China will ship roughly three times more humanoid robots in 2026 compared to last year.
Fang broke down current humanoid robot demand into three main categories. In factories (accounting for roughly 50% of humanoid demand, Fang said), humanoids are sorting parcels, moving boxes, and handling parts on auto assembly lines. Another 30% of deployments are in government and private sector applications where robots are easing pressures from underemployment and auto-sector overcapacity. Finally, 20% of humanoid demand draws from exhibitions and demonstrations.
Of these applications, Fang highlighted auto assembly as the main factory use case. She noted that a line can employ about 1,200 people and that a chassis alone can have more than 70 parts. She explained that these lines are narrow, uneven, and expensive to redesign, and that a robot that walks on two feet can work in them as they are.
[W]here you have the humanoid robot, the [value added is in] the feet. Because [the factory line] is relatively narrow, difficult for you to navigate through, there's ups and downs, it's expensive to change the overall floor planning. It's relatively cheaper to buy…humanoid robots with a two-footed solution. …So I began to realize that there are some good ROIs on those kinds of applications.
A high-quality robotic hand is even more important in factory applications: “A truly capable hand is at least [eight times] more valuable than the body itself,” said Fang. The most advanced ones offer 22 degrees of freedom — enough to play the piano — versus just four or five degrees of freedom for lesser models, according to Fang. The race to develop a more cost-effective yet functional model could benefit European manufacturers, which produce some of the most advanced hands on the market today, Fang noted.
Chinese Supply Chains Commoditizing Hardware
Fang said that Chinese suppliers are driving down prices for actuators, the joints that control a robot's strength and movement and make up about 60% of a humanoid's total materials. She described a Western supplier quoting $1,000 per linear ball screw for Tesla's Optimus program and a Chinese supplier offering the same part for around $75.
Industrial robots went through the same shift, Fang noted. The “Big Four” of industrial robotics (Fanuc, Yaskawa, KUKA, and ABB) held about 80% of global share when she began covering the sector around 15 years ago. She said Chinese buyers took about 60% of the roughly 500,000 units shipped each year, while Chinese makers held less than 10% of the market. Now, the Big Four hold about 40% combined, with most of that decline coming in the past five to six years as cost-conscious Chinese electric vehicle makers opened the door to local robot makers. Her current base case calls for Chinese companies to hold about 50% of the humanoid market.
Gauging Warehouse Automation ROI Potential
In warehousing, Gavin Curran expects rising labor costs, an aging workforce, and capacity constraints at older sites to drive automation investment. Right now, he sees most of the capital flowing into autonomous mobile robots and goods-to-person systems that bring items to a stationary worker.
When assessing a site’s potential for automation, Curran said he starts by asking whether the operation is already well run. 90% of the time, that’s not the case, due mostly to aging back-end systems, he said. He also assesses time lost to walking, double handling, and waiting across the site, measuring across the full workflow and over a long time period to rule out seasonal effects.
Curran sees a role for humanoid robots at existing sites that are short on space and labor as volumes grow. Yet humanoids priced at $20,000 to $50,000 are “probably early” overall, especially for new-build “greenfield” sites. In fact, both speakers indicated that the ROI math is more compelling for existing “brownfield” sites with throughput constraints than for greenfield sites.
But the purchase price is only one consideration in the broader business case, alongside factors like reliability, maintenance needs, and expected downtime. Additionally, projected ROI and actual returns achieved in a live operating environment can differ greatly, which complicates the ROI math. While vendor simulations can show AI-driven productivity gains, those gains need to be validated before Curran trusts them for modeling. This is why he advises operators to be involved as early as possible to minimize the chances of an error or oversight making it into production, where a fix would be much costlier.
Software and Data: Where the Value Accrues
With the actual hardware increasingly being commoditized, Fang said much of the investment occurring in Beijing was in the robot's “brain.”
As an example, Fang pointed to fingertip force data, which is how a robot knows how much pressure to use on a champagne glass compared with a hot coffee mug. This data must be captured from real human hands. Fang estimated the industry has gathered about 10,000 hours of this type of data, but even so, it may need 10 times more data to reach a true “ChatGPT moment.”
Fang expects a recurring model to win out whereby manufacturers buy operating data from customers and charge for software and model upgrades. But both speakers described data sharing as a constraint on that model. Fang said machine vision customers share data in exchange for upgrades but often stay with one supplier. For example, she expected German automakers to stop buying from KUKA after Midea acquired it, because they would worry their data could reach Chinese competitors. Meanwhile, Curran said IT, legal, and risk teams new to automation usually start by refusing to share any data. He favors separating commercially sensitive data from the operational data a device needs to improve.
China Central to Robotic Automation Scaling
Escalating geopolitical tensions and U.S. protectionist measures remain a headwind to progress in humanoid robotics adoption. However, Fang expects Chinese supply to reach other markets one way or the other, “through the front door or through the back door.” No other region has a supply chain with the required depth and flexibility to serve as a China substitute, in her view. She noted that Korean and Japanese makers already rely on China's supply chain. Once the dust settles, Fang expects seven or eight humanoid makers worldwide: one or two Korean, two or three U.S., and the rest Chinese.
Curran pointed to raw material availability constraints and growing supply chain complexity as key hurdles to humanoid development outside of China. In this environment, firms have to build flexibility into contracts, partnerships, and the full supply chain, he said.
For something that's going to happen in a number of years’ time, it will change during that life cycle and that investment cycle, whatever we're looking at. So I think we need to be adaptable in our approach, and certainly we're looking at investment and our contract arrangements and who we're working with and who the partners are and thinking about the whole end-to-end supply chain.
Never Miss a Game-Changing Development
Separating signal from noise is a challenge even for the most sophisticated analyst teams. In robotics and automation, those signals come from factory floors, warehouse operations, component suppliers, and trade policy. As more robotics projects move from pilot to full-scale deployment, research teams will need to connect what operators are seeing on the ground with what analysts are discussing behind the scenes.
SuperAnalyst, AlphaSense’s always-on agent, was designed for this purpose. Built to act like an extension of your team, SuperAnalyst operates as a power user of the platform with the judgment to know when and how to use the right content, tools, and workflows. With SuperAnalyst, you can be confident that every decision rests on evidence you can trace and trust.
See how AlphaSense’s SuperAnalyst can help you gain a competitive edge. Start your free trial of AlphaSense today.
