Physical AI and Specialty Chemicals: What Is the Long-Term Link?

Over the long term, physical AI is likely to give specialty chemical suppliers a new market and a new source of customer data at once. Robots need lubricants, may need thermal fluids, and patented joints can report lubricant condition. This forecast holds if robot fleets grow as the International Federation of Robotics expects and suppliers keep application knowledge structured.

• The global operational stock of industrial robots rose 9% to a record 5 million units in 2025, with more than 600,000 new units installed during the year (IFR, 2026).

• About 7,000 full-size humanoid robots, above 140 cm tall, were sold globally in 2025, and the IFR has identified 174 humanoid robot manufacturers (IFR, 2026).

• The IFR forecasts average annual industrial robot growth of 8% in Asia, 6% in Europe and 2% in the Americas for 2027 to 2029 (IFR forecast, 2026).

• In a robot maker's endurance tests, iron powder in reducer grease rose gradually with wear and then rapidly just before failure (Kawasaki Jukogyo, US 11577394 B2, 2023).

• A strain wave gear maker lists four areas within its gears that require lubrication and tells users to replace the lubricant at the recommended service interval (Harmonic Drive LLC, accessed 2026).

What did the IFR's World Robotics 2026 report find about physical AI?

The International Federation of Robotics (IFR) reported on 24 September 2026 that the global operational stock of industrial robots reached a record 5 million units in 2025. In its World Robotics 2026 release, the IFR wrote that the stock "surged 9% to a record 5 million units in 2025," driven by an 11% jump in annual installations to more than 600,000 new units (IFR, 2026). China installed 354,000 of those units, almost three of every five worldwide (IFR, 2026).

The IFR's outlook is a forecast. For 2027 to 2029 it projects average annual growth of 8% in Asia, 6% in Europe and 2% in the Americas (IFR forecast, 2026). Among the technology trends it lists are AI as an enabler for robotics and robots as part of connected production systems with factory data integration (IFR, 2026).

A second IFR release on 30 September 2026 covered service robots. Professional service robot shipments rose 24% to almost 250,000 units in 2025, a count from 238 of 1,001 manufacturers that the IFR does not project to the whole industry, and about 7,000 full-size humanoid robots were sold globally (IFR, 2026). The IFR also describes China's robotics industry as "moving towards becoming an innovator in physical AI and a developer of intelligent industrial robotics" (IFR, 2026). Physical AI here means AI that acts in the physical world through robots and autonomous machines. The IFR lists labor shortages among the factors that increase demand for automation (IFR, 2026), the same pressure behind the skilled labor shortage in process industry maintenance.

Why will physical AI create demand for specialty lubricants and thermal fluids?

Physical AI creates demand for specialty lubricants because robot joints transmit motor power through reducers that run on grease or oil suited to the gear, its size and its temperature. In one robot maker's patent, a six-axis robot joint holds a reducer, and "Grease (a lubricant) 10 is filled inside the reducer 8 to lubricate gears and bearings configuring the reducer 8" (Fanuc, US 11045960 B2, 2021). A strain wave gear maker lists four areas within the gear that require lubrication: the wave generator bearing, the Oldham coupling if one is included, the gear teeth and the inner diameter of the flexspline (Harmonic Drive LLC, accessed 2026).

The same maker says users "must choose the option that best suits your gear model, size and application," lists one grease for operating temperatures from 0°C to +40°C and another for -10°C to +70°C, and offers an NSF H1 certified grease for food processing machines (Harmonic Drive LLC, accessed 2026). Each is a specialty product chosen against conditions, like NSF H1 food grade grease elsewhere in a food plant.

Thermal fluids are the more speculative half of this forecast. A university motor lab states that excessive heat in humanoid joint modules "severely restricts their long-duration and high-load operation" and is developing liquid-cooled joint actuators (HKUST Guangzhou AMDT Lab, accessed 2026). If liquid cooling spreads to production robots, coolant demand is likely to follow; that is an inference, not a market figure.

How will robot condition data change how specialty chemical suppliers serve customers?

Robot condition data is likely to move lubricant service from fixed replacement schedules toward replacement when the data calls for it, and that changes what customers ask a specialty chemical supplier. Robot makers have patented the means. One patent places a sensor that "measures concentration of iron powder, which is impurities included in the grease" in the reducer's grease supply port, estimates grease temperature from motor and friction heat, and calculates the grease's remaining life and a recommended replacement date (Fanuc, US 11045960 B2, 2021). Another describes endurance tests in which iron powder in reducer grease "gradually increases with the progress of wear" and "increases rapidly due to abnormal wear just before a failure occurs" (Kawasaki Jukogyo, US 11577394 B2, 2023).

When a robot can estimate its own grease life, the customer's questions shift. A maintenance team stops asking only which grease to buy and starts asking why one axis uses up its grease faster than another, whether a different product would last longer under that duty, and what a rising iron reading means. Those are application questions. Answering them takes the customer's operating history, the lubricant's data sheets and earlier cases from similar sites. The practice resembles oil analysis in industrial condition monitoring, with the robot supplying the readings. This forecast is an inference from the patents, not a measured shift in the market.

Why does a specialty lubricant for a robot depend on the application?

A specialty lubricant for a robot depends on the application because the rate at which a reducer uses up its grease additive differs with the robot type and the joint's position. In its endurance tests, one robot maker found that the additive in reducer grease was consumed linearly and that "when the additive is consumed completely (a consumption rate of 100%), a failure (abnormal wear) occurs" (Kawasaki Jukogyo, US 11577394 B2, 2023). Further tests changed the robot type, the installation place and the test time, and found that "even for a speed reducer for the same joint (JT2), if the type of robot is different (machine name: BX300), the numerical values of the additive consumption rate are different" (Kawasaki Jukogyo, US 11577394 B2, 2023). The same tests found the consumption trend was linear in every case, so the pattern carries across robots while the rate has to be read per installation (Kawasaki Jukogyo, US 11577394 B2, 2023).

For a specialty lubricant supplier, that finding has a direct reading. A grease recommendation for a robot cell cannot be copied from the robot's model number alone. It is likely to depend on the axis, the load, the cycle and the conditions at that customer's site, and the evidence for it builds up one installation at a time. A supplier that keeps a record of which product went into each customer site under which conditions holds the start of that evidence. A supplier that keeps it in email threads has to rebuild it each time a customer asks.

How do robot makers and robot fleets compare as specialty chemical customers?

Robot makers and robot fleet operators are likely to be two different specialty chemical customers, and the difference lies in the data each one holds. The table below compares the two; cells without a source are inferences from the sources above, not survey results.

The row that matters most is what data exists. A robot maker qualifies a grease once, under test conditions. A fleet operator with condition monitoring would generate readings from every robot in every shift, and each reading raises a question only someone who knows that site's history can answer well. Suppliers that serve fleets are likely to be judged on how well they interpret those readings. The products may be the same in both columns; the service around them is not, and it depends on records kept per customer and per site. A supplier that already answers gear and bearing questions with the site's history in hand is practicing the second column before any robot arrives.

Is physical AI too early to matter for specialty chemical suppliers?

Physical AI is too early to matter for humanoid robot volumes, and the IFR's own assessment points that way. About 7,000 full-size humanoids sold in 2025 is a small market next to more than 600,000 industrial robots installed in the same year (IFR, 2026). The IFR's market presentation states that humanoid robots "are still on the verge of making the leap from laboratories to pilots" and lists safety and functional reliability, lack of training data and the lack of an economically viable business case as major challenges (IFR, 2026). A supplier that plans on humanoid grease volumes before the end of the 2020s is betting on volumes the IFR's 2026 materials do not forecast.

The opposing view is weaker for the installed base. Industrial robots already run in millions of joints, and the grease-life patents cited above have priority dates in 2017 and 2018 (Fanuc, US 11045960 B2, 2021; Kawasaki Jukogyo, US 11577394 B2, 2023). Robot condition data is not a humanoid story. It arrives through ordinary six-axis robots in ordinary plants. The IFR also expects humanoids to act as visible innovation that will spill over to established robot forms (IFR, 2026), which suggests their components and lessons reach conventional robots first. The fair reading is that humanoid demand is a long-term forecast, while robot condition data is a practice already described in patents.

What should a specialty chemical supplier's knowledge system have for physical AI?

A specialty chemical supplier's knowledge system for physical AI should keep each robot application as a structured case that customer data can be added to over time. Five conditions matter most:

• Each customer, project and site visit connects in one record, so a robot cell's history sits with the account rather than in one engineer's inbox.

• Data a customer shares on request, such as grease analysis results from a robot cell, becomes part of that customer's project record once approved.

• Every recommendation shows its basis: the technical data sheet (TDS), safety data sheet (SDS) and field history behind it.

• Missing information is not made up: an estimate is marked as an estimate, or the point stays unconfirmed.

• Each case ends as one document that the next similar robot application can draw on.

None of this needs a robot customer to begin. The cost of waiting is that every gear, bearing and motor case answered only by email leaves nothing for the robot work to build on. A supplier can start with the next gear or automation inquiry it receives, run it as a recorded case, and let the record grow one case at a time. The application knowledge that fleet operators will ask for is the same knowledge, kept in a form that can be found again.