Forwarding Pax Technica: Abundance comes when intelligence commoditizes and labor costs trend to $0. Building next-gen general foundation model for physical AI.
Internet
Joined March 2010
- Tweets1.5K
- Following557
- Followers3K
- Likes633
Pinned Tweet
AI commoditizes intelligence (we are here)
-> Coordination cost drops
-> Trusted outcomes become the scarce resource
-->Civilization becomes Pax Technica
When a toddler starts exploring the world, how much of that is pre-training, mid-training, or post-training?
If humans are the most optimized instrument for our physical world, why should training robots be any different?
The challenge with these businesses come down to cost vs butts-in-seats. Can T800s fighting bring enough of a spectactle beyond a one-time thing? Sure, over time likely yes but at this juncture the cost to operate a league may be premature. It is cool seeing EngineAI early dominate a new demographic just like Unitree saturated the researchers/university market.
Yet another humanoid robot combat league is launching. This time in Australia.
HFC (Humanoid Fighting Combat) kicks off in Melbourne on Nov. 21. They’re pitting eight Terminator-inspired T800 robots from the Chinese company EngineAI against each other
EngineAI continues to win the robot combat format war against Unitree but it’s still early days.
We went from narratives based on research papers --> demos --> accessible. Now missing the at scale deployment piece >>> any research for VCs. Time to walk the walk (literally).
The Dreamlabs announcement is not just usual launch but a well timed release that resets the robotics narrative
The last few weeks, the scare of omni-models that will “win” robotics has been a common topic of discussion. Astra, Argon, Opus might eat the roboticist’s lunch.
Dream-1 challenges that narrative, outperforming all of them on benchmarks. The savior, the robotics community is looking for to not discount native robotics foundational models.
There is a clear tension being explored publicly and privately about what type of model wins in robotics.
Why are physical AI funding rounds so large? Jensen claims physical AI is the final frontier, experts estimate ~$90Tn TAM. Humans designed work under physical world constraints, and so if you could solve any one of those action primitives, that, in theory, could claim a chunk of the above TAM.
If you are a Western founder looking to get into robotics in China @Robo_Tuo should be your first port of call. Highly recommend!
Can't believe our first WeChat group for global founders in Shenzhen has already hit 500 members and is full! We just opened a second group.
We're welcoming founders, builders, and developers from around the world who are working on robotics and Physical AI in Shenzhen to join our community.
DM me if you'd like to join!
This just opened up a whole new dimensional space and race. Key is now dependent on cost-effective deployment.
I have been thinking about this quite a bit – where then does value accrue for physical AI? What still is scarce?
- Software/brains get commoditized (fable vs astra)
- Need for data can be reduced via lighter weight RL + sims complementing the above
- Tactile/touch for long tail precision tasks is still missing
- Hardware gap needs to be solved ie dependency on China (likely will achieve physical AI deployment at scale first)
- Deployment is an ops game but still requires hardware dependency
Next gen robotics companies will come from operational excellence
The right form factor is an agent you naturally engage as if a person. Everything “AI” is abstracted away in the back end.
This is my thesis: no one uses AI. I repeat, absolutely no one. We live in a bubble.
Even among my friends who pay for it, when I ask them to open ChatGPT and show me their queries, it’s the same handful of basic things.
Most don’t even know they can upload a photo and ask questions about it. Connecting Gmail so an agent can read and send emails blows their minds. An agent opening a browser and checking them into a flight? They’ve never even heard of it.
The massive challenge right now is adoption, and then getting people who already signed up to actually use what they’re paying for. Most have absolutely zero clue what’s possible.
Imagine the compute shortage when everyone starts using AI like the top 1% of users do today.
Contrarian take but I believe so. Software exists largely because human labor is expensive in the West to coordinate at scale. Historically, human behavior tended towards instructing labor to conduct a task as opposed to being self-serve. Agents will bring us back to that natural behavior at software scale and margins.
Why browse UberEats or Doordash when you can instruct your agent what food you want and at what time?
Why browse Booking.com or Kayak to book your trip when you can talk it through with an agent?
Etc.
Game theory dictates even if not all compute spend is warranted you don't want to be the one left out of the race. Dollars from rich tech margins flow to hyperscalers' FCF which is used then buy semis, driving profits there. Those profits go into building more supply to satiate the demand.
It felt like we hit a plateau on consumer demand for AI end of last year and then came Opus4.5 and Openclaw which paved the way for products like Meta's Muse, which seem to have found PMF and thus justify the demand. This should unlock continued growth. Zooming out though, IMO this creates a barbell investment strategy: long the leaders/monopolists who own the distribution or scarce resources, and then the upstarts who can continue to widen the use cases. Long the gap, specifically the ones becoming non-cyclical.
A whole decade of builders who entered the job force between 2010 and 2020 now have to contend with a regime shift where hardware matters again – it always has. Hardware dictates the ceiling or physics of what one can build on top and the human behaviors that result (eg self-facing cameras enabling selfies and Snapchat).
What a time to be a builder in Silicon Valley. Not only does culture reward failure (tries) – like having infinite lives in a video game – compressed multiples in the market driven by actual profits mean future multiple expansion back to previous benchmark heuristics could be reflexive.
Just like how startup offices around the world adopted a "WeWork" aesthetic, we're seeing the same play out with this decade's startups' websites that adopt a similar "vibe-coded" aesthetic.
You can't describe it but know it when you see it.
The decades ahead for physical AI progress will be quite exciting.
🇨🇳 The oldest Chinese robotics founder is 36. The youngest is 22.
I walk European delegations through the biggest Chinese robotics companies almost every week. What people notice first is who walks into the room.
Wang Xingxing started Unitree at 26 and took it public in August. Chen Jianyu held a Berkeley PhD and a Tsinghua professorship before founding Robotera at 31.
Jiang Zheyuan left his Tsinghua doctorate unfinished to build Noetix. Yang Fengyu paused a Yale PhD and started UniX AI at 24.
Then there is Zhao Tongyang at EngineAI, who has no university degree at all. His team landed the first front flip a humanoid has ever done.
Five of the ten on this list studied in the United States. All five came home.
The talent loop that used to end in California now ends in Beijing and Shenzhen. And it ends with founders young enough to still be running these companies in 2045.
If you are choosing a robotics partner in China this year, you are choosing who you will still be working with in twenty years. Worth knowing who they are.
Swipe for all ten.
This is fantastic. An expression of Pax Technica at the foundation is a world where nation states move to security substrates and sovereign AI stacks sit in top. For example you will be subscribed to the Elon stack eg SpaceX starlink + xAI Grok + Tesla robotaxi and Optimus ecosystem or a different one eg OpenAI + Figure + Uber. These stacks will become holistic ecosystems where we will spend most of our time engaging.
