Robotics has long been an area of promise but (I think) limited returns
Robots are a branch of industrial manufacturing machinery. That is not, historically, a high-margin business. It also demands high reliability and long machine life.
Interestingly, there's a trend towards renting robots by the working hour. It's a service - the robot company comes in, sets up robot workers, services them as needed, and monitors them remotely. The robot company gets paid for each operating hour. Pricing is somewhat below what humans cost.[1]
[1] https://bernardmarr.com/robots-as-a-service-a-technology-tre...
The end user usually doesn't have the expertise to even maintain the systems, nor does it make sense for them to do it in-house.
Charging per item of work (operating hour or thing processed) allows use of consultants but keeps incentives aligned between all parties (maximize uptime/productivity).
Lots of dotcom busts in the late 90s were concepts that worked 10-15 years later. We just did not have broadband and smartphones. Battery and AI tech is quite likely to be the missing piece robotics lacked in the past.
Cheap semiconductors as well.
Fabricating a chip on a 28nm and 48nm process is extremely commodified nowadays. These are the same processes used to fabricate an Nvidia Tesla or an i7 or Xeon barely a decade ago, so the raw compute power available at extremely commodified prices is insane.
Just about every regional power has the ability to fabricate an Intel i7 or Nvidia Tesla equivalent nowadays.
And most regional powers have 3-7 year plans to build domestic 14nm fabrication capacity as well now. A number of firms like Taiwan's PSMC have made a killing selling the end-to-end IP and workflow for fabrication.
This will probably take off once Amazon finally gets robots that can do unboxing, picking, and boxing. They've been trying for years to get that to work. Amazon already has robots doing most of the lifting and carrying, but people still handle each item.
People have been trying to do bin picking fulfillment with robots since the 1980s. Swisslog, Brightpick, and Universal Robotics have all demoed this, but so far it's not working well enough to take over. It's getting close, though.
"Slow but steady" I would call it.