Good morning, Per Aspera readers. First, I want to throw love to all at SpaceX (and ex-SpaceX’ers) on a quiet, yet major milestone: your launch of the 10,000th Starlink satellite.
👇 Now, for today’s main thought: we’re in a global frame where the question of who builds the technologies of tomorrow matters more than ever. After all, that’s at the heart of our Renaissance. So, let’s take a quick state-of-play on two pivotal fronts for industrial leverage: (i) humanoid robots and (ii) the rare earths essential to building them.
Thank you for being here and — even more — for driving our great 🇺🇸 Renaissance forward.
THIS WEEK IN THE RENAISSANCE
🤖 Humanoid robots, meet capital markets
⛏️ Latest rare earths play-by-play
✍️ Friends in high, hard places
⚡ The grid, exo-shoes, etc.
Forwarded this? Subscribe for more.
A 200-page teardown of Boston Dynamics' Spot robo-dog has been making the rounds online recently. Written entirely in Chinese, the document is forensic in its detail. The report’s authors put Spot through its paces, stripping the quadruped down to its motors, mapping payload interfaces, and even accidentally dunking it in a flooded drainage channel.
Spot’s components
Spot’s engineering brilliance is undeniable, the engineers note. The robot’s leg actuation achieves a level of fluid locomotion that looks straight out of the animal kingdom, the kind that has made Boston Dynamics world-famous. But the teardown also exposes the robot’s limits: dustproof-not-waterproof sealing, avionics-grade components ill-suited for industrial disassembly, and a tendency to slip on wet surfaces. (Though, to be fair to Spot, who among us doesn’t have this tendency? 🙋)
Call it competitive intelligence or industrial espionage with receipts…either way, it's old news. The teardown is from 2021. The robotics ecosystem has moved on. Which brings us to today’s main story: How a cottage industry of lab-bound curiosities has transformed into a capitalized, battle-tested production juggernaut…in China, at least.
Two Humanoid IPOs, In This Economy?!?
It seems as if not one but two of China’s leading humanoid robot makers will IPO on the Shanghai STAR stock exchange within the next year, in a sign of how fast the sector has matured arrived.
Leju Robotics has raised $210M+ in pre-IPO funding, with preparations for a public listing well underway. The Shenzhen-based humanoid maker is partnered with 40+ companies, including giants like Tencent, Huawei, Alibaba, and China Mobile, to deploy its Kuavo bot in factories.
And, earlier this year, Unitree Robotics filed for a $7B IPO, recording $140M+ in 2024 revenue and four years of positive cash flow (a rarity in robotics).
Unitree, Unpacked
The firm aims to be to ground robotics what DJI is to the skies — to dominate the market (both humanoid and quadruped) with accessible, mass-producible, and aggressively priced hardware. And so far, so good. DJI sells 70-75% of all drones globally. Unitree, meanwhile, commands a ~70% share of the quadruped market.
Source: Unitree
And Unitree is no one-trick-(robo)pony. Its product family spans multiple form factors:
Go2, Go2-W: quadrupeds with LiDAR and GPT-enabled control for developers and hobbyists.
B2, B2-W: industrial wheeled–leg robots capable of carrying 40–120 kg payloads across rough terrain at 20 km/h.
G1: mid-size bipedal robot (127 cm tall) with 23–43 joint motors, capable of running, squatting, and backflipping.
H1: full-scale bipedal robot (180 cm tall) with 19–27 joint motors.
H2, unveiled earlier this month: a 70 kg humanoid with enhanced limb articulation, 7 DoF arms, and modular Nvidia Jetson compute units.
Anatomy of an Automototon
The story behind the story is unit economics and the BoM (bill of materials). We look at the trendlines for both Western and Chinese ecosystems. Below, you’ll see an estimated BoM for Optimus, a robot from the most vertically integrated Western robotics developer: Tesla.
Tesla Optimus bill of materials (BoM) & estimated costs. Of any Western humanoid developers, Tesla is as vertically integrated as they come. Source: Morgan Stanley Research.
Morgan Stanley researchers peg Western humanoid costs, today, at $50,000-$60,000 a unit. This is trending in the right direction, down from $200,000+ just a few years ago, but Western robots are exquisitely priced relative to Asian counterparts. Case in point: A Unitree humanoid now retails for as low as $5,900, which we fixated on almost ad nauseam in The Last Hardware Problem. That’s aggressively low, likely at or below cost. But it’s still remarkable.
What Drives the Cost Gap?
Three structural advantages for China:
Deployment density: China out-deployed the U.S. with industrial robots nearly 9-to-1 last year. (A similar story persists across all form factors, whether they’re wheeled or quadrupeds).
Vertical integration: Chinese OEMs often build their own actuators, motors, and reducers in-house. If they don’t, they source from Shenzhen suppliers down the street.
Cost arbitrage: Western supply chains depend on these same Chinese components.
Actuators, for instance, consume 30-55% of total humanoid BoM.
A humanoid built with Chinese components costs $46,000 versus $131,000 using non-Chinese supply chains.
Actuators alone: $22,000 versus $58,000.
China controls ~63% of the humanoid supply chain and dominates rare-earth processing (90%+), battery manufacturing (70%+), leaving Western roboticists dependent on Chinese components, with few, if any, viable alternatives.
State of Play, Game Theory, & So What?
Humanoids are inverting the smartphone/semiconductor model. Whereas the latter is design in the West, develop in the East, diffuse everywhere, humanoids are design, deploy, diffuse in the East, with the West mostly just designing.
The bet is straightforward: that the BYD/DJI playbook will also work for quadrupeds and bipedal humanoids. It worked for batteries. Worked for drones. Why not robots? Well, maybe because humanoids are harder. Maybe the “general-purpose robot” is like self-driving: the last 20% of real-world reliability and long-tail edge cases eats 80% of your development time.
Still, you don’t have to buy into the humanoid hype to take this seriously. Why? Because the infrastructure one needs to build for humanoids — actuators, sensors, advanced manufacturing capacity, embodied AI training loops, supply chain muscle, etc — can just as readily be used in a dozen other critical technology areas. So, China has two paths to victory.
If humanoids “work” in the near term…China wins decisively, with 1M+ units shipped, insurmountable cost advantages, and unmatched real-world data that the West cannot easily replicate.
If humanoids stall, the investment still “breaks even.” The toolchain doesn’t go to waste. Instead, it’s repurposed for logistics, drones, advanced manufacturing, and autonomy.
Per Aspera’s 2¢: So, there you have it. The West is betting on software, autonomy, and design excellence. China is betting on owning the end-to-end, full-stack industrial base. No matter how you game this out, there’s only one rational move for the U.S., and that’s to also make the full-stack bet. Don’t just design, build better. Own the supply chain. Control the manufacturing. That’s the only strategy where we don’t lose.
The Latest Play-by-Play on Rare Earths
“Not a ban, I swear" Chinese rare earth exports, by tonnage, fell ~31% MoM in September, as Beijing’s tightening export regime began squeezing downstream supply chains.
The contraction hit before the sweeping Oct. 9 controls even took effect, with Chinese rare earth magnet firms facing heightened scrutiny on licensed applications starting in September.
Despite Chinese officials’ diplomatic charm offensive to assure trading partners that this is “not a ban,” markets aren’t buying it, as inventories tighten globally and spot prices edge higher.
Into that vacuum steps Canberra. Australia pitched its vast resource holdings as the democratic world’s rare-earth insurance policy.
The Aussies told the Americans that they hold deposits of 43 of the 55 minerals Washington designates as “critical.”
Fact check, true. Australia is like “a periodic table that lights up like a Christmas tree.”🎄 Mining is ~15% of Australian GDP, vs. 1.3% in the U.S.
Australia’s pitch worked. Last week, President Trump and Australia PM Anthony Albanese inked a $8.5B critical minerals pipeline pact.
The deal commits ≥$1B from each partner over six months to jumpstart a priority project pipeline spanning mining, refining, and processing.
Offtake agreements and strategic stockpiling mechanisms are included in the partnership, to buy down commercial risk for projects in both nations.
China’s response: Beijing expressed immediate displeasure, with state-run
mouthpiecesmedia blasting the deal as “strategic containment dressed as supply chain security,” framing it as another attempt to encircle China economically. (As a reminder, China still controls ~90% of refined rare earths, 70% of lithium, and 88% of cobalt globally.)Now all eyes are on Australia. The stable, resource-rich mining powerhouse is already scaling refining infrastructure that matches China in technical capability if not yet in volume. Australia’s transformation from raw-ore exporter to processing hub is a structural shift years in the making, which now has serious geopolitical momentum at its back.
Meanwhile, in Europe…As the EU prepares its own “trade options” to counter China’s, Beijing is reportedly forcing German firms to hand over sensitive supply chain information in exchange for REE export licenses.
Project Amplify
Nike has unveiled Project Amplify, a “powered footwear system” that pairs a carbon fiber-plated running shoe with an ankle-mounted motor, drive belt, and rechargeable battery cuff. The system runs (ha) on algorithms derived from the Nike Sport Research Lab.
The normal caveat with this kind of stuff applies here: You can’t buy this today, and Nike is targeting “a broad consumer launch in the coming years.” Still, credit where it’s due: Nike is developing robotic assistance technology and dreaming of consumer micro-exoskeleton products. (Ryan owns a pair of Nike’s discontinued “self-lacing” shoes, which he still loves.)
With everyone fixated on humanoids, we’re missing the more immediate human augmentation stories. And this one’s bound to entertain: If Project Amplify ever ships, we’ll see sneakerheads camping out overnight for limited-edition exoskeletons and hypebeasts reselling human-augmented hardware on StockX.
Five-Alarm Fire?
The U.S. grid still earns “very high” reliability marks, but near-misses and small-scale events are piling up. NERC CEO Jim Robb called the situation a “five-alarm fire” last week, citing shrinking reserve margins, glacial permitting, extreme weather, cyber threats, and one new digital wildcard that’s breaking all the traditional assumptions about industrial loads.
What, oh what could that last one be? Why, datacenters (DCs), of course! “AI factories” could jump from ~4.4% of U.S. electricity consumption today to 12% by 2028. And unlike steel mills or chemical plants that ramp predictably, these facilities spike from zero to gigawatt-scale demand in minutes.
The grid was built for factories, not server farms that ignite like afterburners…
FERC is racing to issue colocation guidance for DC/power plant colocation, while planners ditch their priors to adapt to fast-ramping modular loads and their stability impacts (one example: trading annual consumption models for real-time spike analysis). Transmission and interconnection reforms are also in the works.
Georgia case study: The state approved 7.1 GW of new capacity, with DCs signing 15-year contracts to cover infrastructure costs. Projections suggest residential rates could drop $2.64/month as large users subsidize grid expansion. Whether that math holds when every state needs gigawatts remains to be seen.
The hyperscaler ask: A lot more, please. OpenAI wants the U.S. to add 100 GW of new energy capacity annually. For context: China added 429 GW last year. The U.S. added 51 GW.
Current reality: MISO’s operations chief summed it up well: “Where we are today, I would say, is not safe.” Translation: Time to build!!!
Friends in Hard, High Places
Muon Space is the first external partner to tap into SpaceX’s space-to-space laser relay network, integrating Starlink’s 25 Gbps mini laser terminals directly into its Halo satellite platform.
The problem: For decades, satellites have operated with long data delays and patchy coverage, limited by short ground station passes.
The solution: Some time ago, Starlink started adding optical inter-satellite links to its satellites. The near-infrared lasers form a mesh that relay data across thousands of kilometers at the speed of light, all without touching ground. Each satellite maintains connections forward, backwards, and sideways to adjacent satellites.
The impact: For Muon’s FireSat constellation, this means wildfire data that once took 20 minutes to deliver can now be transmitted in seconds, giving first responders streaming insights to detect ignition, alert crews, and stop blazes before they spread.
Why this just works: By opening its laser mesh to third parties, SpaceX gets to monetize excess capacity and turn sunk constellation capex costs into sweet, sweet recurring revenue. Meanwhile, operators like Muon get to tap into a commercially shared high-speed backbone (à la Starlink) and unlock new low-latency applications (that are more societally important &/or economically valuable). They get infrastructure and global reach that they couldn’t easily or economically otherwise attain. This feels like a genuine 1+1=3 moment, as both parties win and American industry advances orbital infrastructure and resiliency.
Skyryse is partnering with Dynamic Aviation to bring its SkyOS automation system to fixed-wing aircraft, starting with the Cirrus SR22 (pictured above). This is a big step toward a new era of fleet resilience across commercial, military, and emergency aviation! 🛫
Editor’s Picks
Nvidia is reportedly planning its own robotaxi challenger to Waymo, Tesla // Crusoe, lead developer of Stargate, raises ~$1.4B Series E with freshly minted decacorn status // Amazon intends to tap robots to automate up to 75% of its operations, and avoid hiring a total of 600,000 jobs through 2033, according to leaked docs —here’s a look at the company’s key warehouse workhorses // the amount of VC into AI labs for materials and chemistry is comparable to the NSF’s annual budget for materials and chemistry // The world needs more spaceports; Oman wants to help // Stop saying NIL will replace EUV // YC’s Summer 2025 batch has ~6% “industrials” startups, with the vast majority of companies working on some flavor of AI (voice, coding, agents, infra) // AR goggles news: Samsung launches new Galaxy XR headset, retailing for $1,800 and weighing 545g; Amazon tests AR smart glasses for delivery drivers; Apple has moved MFG of latest Vision Pro to Vietnam; Snap raising $1B for its own glasses // Astera Labs, a ~$28B company that designs high-speed connectivity solutions for datacenters, will acquire aiXscale Photonics, a German firm specializing in ultra‑high‑precision optical glass interposer gear … we think this neatly validates Per Aspera’s Photonic Thesis // NASA plans to reopening lunar lander contract, per interim administrator Sean Duffy (no relation to Ryan Duffy) // Musk fired back at Duffy (not Ryan) in signature style // U.S. Army Secretary Daniel Driscoll and Treasury Secretary Scott Bessent convened Apollo, Carlyle, KKR, Cerberus, and the other big PE groups to pitch “meaty” strategic projects to help fund the service’s $150B infrastructure overhaul.
|