TAKEAWAY#1
Serious investment pouring into humanoid robotics from AI companies
The push to build “physical intelligence” isn’t just talk — major players in the AI space are hiring for humanoid robotics roles, funding startups, and forming partnerships to integrate AI with real bodies.
Open AI led a Series A2 funding round of about US$23.5 million in 2023 and later another US$100 million funding round (Series B) for 1X Technologies to help them build its bipedal humanoid robot NEO, scale production of the wheeled android EVE.
Amazon invested US$150 million in Agility Robotics to help create their humanoid Digit. Agility has recently broadened their relationship further with Amazon.
Figure raised US$675 million, with investors including Jeff Bezos, Microsoft, Nvidia, Open AI AND an Amazon-affiliated fund.
MY TAKE
There is real momentum. But money alone doesn’t guarantee useful results. Many incumbents have poured cash into robotics before, and outcomes have lagged. Yet this level of investment suggests humanoid robots aren’t a fringe idea any more, they’re becoming a strategic priority. As Gen AI hype cools down, AI robotics could be the next thing to blow up.
TAKEAWAY#2
The real world is where the hardest challenges lie
Software breakthroughs (LLMs etc.) are impressive in virtual/lightly constrained settings. While large language models are excelling in virtual domains (text, code, etc.), doing things in the real world — moving, balancing, manipulating objects, reacting to unpredictable environments — is hard. Hardware limits, sensing, reliability, and generalization are major obstacles.
A Berkeley expert says there’s a “100,000-year data gap” — robots simply don’t have the breadth of experience/data the way virtual AI models do, making real-world skill acquisition slow.
Engineering analyses point out that perception (sensors, vision), real-time decision making, mechanical design (actuators, power, balancing) are significant bottlenecks.
When Boston Dynamics’ Atlas did a parkour demo, the success rate was very low (e.g. 1 out of 20 times) under non-ideal conditions.
MY TAKE
The disparity between simulated / lab environments and messy real life is huge. Until robots can reliably deal with variability, edge cases, unexpected conditions, the “humanoid future” remains fragile and expensive.
Help us keep sharing these stories
TAKEAWAY#3
Early use cases are in structured environments like warehouses and factories
Before humanoids roam our homes or streets, the first places they will make sense are places with constraints: predictable layout, limited types of tasks, controlled environment, existing infrastructure.
Agility Robotics’ roadmap suggests starting in logistics, manufacturing and warehousing before moving to more variable spaces like retail stockrooms or grocery.
Amazon’s fulfillment centers are working on robots for transporting “non-conveyable items” in environments crowded with pallets, obstacles, people; this shows incremental deployment.
Figure 02, the latest humanoid robot from the Californian company Figure. This is currently being successfully tested at the BMW Group plant in Spartanburg (South Carolina, USA) in a real production environment.
MY TAKE
This makes sense: you want the uncertainty as low as possible when starting. Factories and warehouses let you predict many things, making the return on investment more realistic. It’s smart strategy. But even in those environments, entry costs, reliability, and maintenance remain steep.
TAKEAWAY#4
There’s significant hype vs. reality tension; many demos are over-curated, and hardware still lacks essential capabilities
People see flashy demos, videos of robots doing cool things, but those are often cherry-picked, teleoperated, or in ideal conditions. Actual general purpose humanoids are far from mature.
Tesla is falling well short of the promised 5,000 units target this year for their Optimus model. As of mid-2025 only about 1,000 units were built. Procurement and production for more has been paused due to unresolved hardware challenges.
Technical bottlenecks: battery life, robust balancing, perception failures (lighting, occlusion, sensor noise), efficient actuators — hardware and mechanical systems still frequently underperform relative to what people see in videos.
Tesla Optimus Gen 2
MY TAKE
The hype is useful for investment and visibility, but it also risks overstating what’s possible now. Practical deployment will lag expectations. People excited by demonstrations should keep their skepticism: many things in robotics remain brittle, expensive, and failure-prone. All the investment and vision won’t mean much if the robots can’t reliably operate outside labs or staged demos.
▶ Listen to the Podcast: WIRED Podcast review: Move Aside, Chatbots: AI Humanoids Are Here
TL;DR
Looking back on what this podcast lays out, it’s clear we stand at an inflection point: humanoids may be around the corner, but the road there is messy. The promise of AGI or robots in our homes is seductive, but until these machines can reliably handle everyday chaos, most of what we hear is optimism.
For all the money, all the demos, and all the bold timelines—how many companies will deliver useful humanoid robots outside labs by 2030? I’m betting few. What do you think? Let us know in the comments.
