Late last year, a Klimt sold for the highest price ever paid for modern art at auction.

An outlier sure, but it wasn't a fluke. U.S. auction sales grew 23.1% in 2025. The $1-5mm segment even grew 40.8% YoY.

Meanwhile, Apollo’s chief economist Torsten Slok said to expect ‘zero in return in the S&P 500 over the coming decade.’

So, that's where it gets interesting.

Each environment is unique, but after dot-com, post war and contemporary art grew about 24% annually for a decade. After 2008, about 11% for 12 years.

It’s also had near-zero correlation with the S&P 500 since ‘95.*

Now, Masterworks lets you invest in shares of artworks featuring legends like Banksy, Basquiat, and Picasso.

Shares can sell quickly, but my subscribers can skip the waitlist:

Skip the waitlist

^{*Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.}

If you’re new, not yet a subscriber, or just plain missed it, here are some of our recent editions.

_{Interested in sponsoring these emails? See our partnership options here.}

Subscribe to the Life Self Mastery podcast, which guides you on getting funding and allowing your business to grow rocketship. 

Previous guests include Guy Kawasaki, Brad Feld, James Clear, Nick Huber, Shu Nyatta and 350+ incredible guests.

On March 20, Unitree Robotics filed a 363-page prospectus with the Shanghai Stock Exchange, applying to list on the STAR board and raise RMB 4.2 billion. The headline numbers will carry the coverage. Revenue of RMB 1.71 billion in 2025, up 335%. Adjusted net profit of RMB 600 million. Gross margins of 60.27% — closer to Kweichow Moutai than to any company that ships physical hardware. The number one position in humanoid shipments: 5,500 units, while Figure and Agility still count theirs in the hundreds.

This is the strongest commercial record humanoid robotics has put on paper. It is also not what it looks like.

Read the revenue breakdown and the business narrows fast. Research and education buys 74% of the humanoids. The consumer units mostly stand at store entrances to pull foot traffic. Industrial deployment is 9%, and most of that is reception desks and tour guides. Margins climbed 16 points while humanoid ASP fell 71.7%.

So the question the prospectus raises is not whether Unitree is profitable. It is what a $5.8 billion valuation is paying for — the company that exists today, or the one in the founder's closing line: "achieve humanity's ultimate dream: AGI."

Unitree calls itself an embodied intelligence company. The prospectus runs to world models, embodied large models, and a founder's letter that closes on AGI. Strip it back and the structure is plainer.

Unitree is not a robot company yet. It is a vertically integrated parts factory that sells its own components as finished machines.

The moat is the cost structure. Unitree self-designs the motors, reducers, encoders, and joint modules — 40–60% of a humanoid's bill of materials. Purchased parts run 14–18% of cost; domestic sourcing sits above 90%. That is why gross margins climbed to 60.27% — Kweichow Moutai territory — while humanoid ASP fell 71.7%. Costs dropped faster than price. The factory is the asset, not the robot.

What sits on top of it is thinner. Research and education buys 74% of the humanoids. Consumer units mostly stand at store entrances. Industrial deployment is 9%, most of it reception desks and tour guides. The shipments lead is real — 5,500 units, more than anyone — but the demand is rented, not embedded.

The financials carry the point: RMB 1.71 billion in 2025 revenue, up 335%, adjusted net profit of RMB 600 million. Profitable, in an industry built to burn cash.

So the question the listing raises is not whether Unitree can build robots cheaply. It can. It is whether the half of the RMB 4.2 billion raise going to the model layer arrives before actuators commoditize — or whether $5.8 billion is paying for the founder's closing line.

Wang Xingxing built his first quadruped in his apartment. Self-taught, working hand-made because in 2016 there was no Chinese robotics supply chain to buy from. That constraint became the company. Forced to fabricate motors, reducers, and joint modules himself, he ended up owning the manufacturing stack that every competitor now outsources.

Unitree was a robot dog company for most of its life. Quadrupeds for research, inspection, industry — State Grid, Sinopec, JD.com walking machines through substations and chemical plants. Humanoids were a rounding error: 1.9% of revenue in 2023, five units sold that year.

Then the flip. Two CCTV Spring Festival Gala appearances. A G1 on Jensen Huang's GTC stage in 2024. The brand exposure converted, and by the first three quarters of 2025 humanoids crossed half of core revenue.

The financials track the turn exactly. RMB 123 million in 2022. 159 million in 2023. 392 million in 2024 — the year it turned profitable. 1.71 billion in 2025.

So the founding story is not a garage-genius myth. It is the reason the margins exist. The apartment had no supply chain, so Unitree built one — and that is now the moat.

The company admits it does not fully understand how industrial buyers use the robots. The sales lead does not come from owning a use case. It comes from cost-performance and iteration speed. Strip that down: there is no scenario barrier. Nothing locks a customer in except being cheaper and faster, and both are copyable.

The battery curve shows how fast that erodes. Lithium-ion packs fell from roughly $1,200/kWh in 2010 to $137 by 2020, an 89% collapse compounding near 20% a year. By the 2015 midpoint the pack had stopped being a differentiator. CATL, BYD, LG, and Panasonic converged, and the winners won on manufacturing scale, not battery novelty.

Unitree's curve is steeper. Quadruped per-unit cost fell from $3,300 in 2022 to $1,800 by mid-2025, 46% in three and a half years. The humanoid curve is shallower at 15%, but humanoids trail the quads on scale, not on physics. The difference is the head start: batteries spent the 2010s building a supply chain that did not exist. China already makes motors, reducers, and encoders at scale, and the carmakers on Unitree's cap table already build most of those parts. So the commoditization timeline is closer to 3–6 years than 10. The thing that makes Unitree special is the thing that copies fastest.

Demand is rented. Q3 2025 revenue fell 26% once the Spring Festival Gala bump faded, humanoids down 32%. That swing is attention, not embedded demand. Unitree says the hype is cooling itself.

Competition is arriving capitalized. Tesla's Optimus, plus BYD, Geely, and Xiaomi, who can absorb a price war 60% margins cannot. The IP is thin: 262 patents, only 20 domestic invention patents.

So the risk is not that Unitree fails to grow. It will. The risk is it grows into a commodity before the RMB 2 billion model bet builds the moat the hardware no longer can.

The market numbers in the prospectus are the kind that justify any valuation. Morgan Stanley projects 1 billion humanoids and $5 trillion in annual revenue by 2050, with the industry compounding at 54% a year over the next decade. At those figures, $5.8 billion is a rounding error.

The numbers are not the opportunity. The gap between them and today is.

Today the addressable market is research labs. 74% of humanoid demand is education and research — the academic buyer who has anchored Unitree since 2022. That is not the $5 trillion. That is universities buying one machine to study. Real, repeatable, and small.

The quadruped side shows what a mature market looks like, and it is instructive. Only a third of quad revenue is research. Over 40% is commercial. State Grid, Sinopec, PetroChina, JD.com run them through substations and pipelines for actual inspection. The use case is established because the technology is. Humanoids are not there yet — Unitree says so, putting industrial deployment at 9% and most of that at reception desks.

So the opportunity is a timing bet, not a sizing one. The TAM is not in doubt. The arrival date is.

Unitree's own target frames it: 75,000 humanoids and 115,000 quadrupeds annually within five years — roughly 14x the 2025 humanoid volume. The number admits how early this is.

The question the market opportunity raises is not whether the $5 trillion exists. It is who is still standing, and still cheap, when the labs become factories.

Unitree sells two product lines, and the prospectus treats them as one story. They are not. They are two different businesses at two different stages.

The quadrupeds are the mature business. Go2 for consumers and research, B2 for industrial, A2 above it. Over 60% global market share in four-legged robots. The cost curve tells the story: per-unit manufacturing fell from roughly $3,300 in 2022 to $1,800 by mid-2025, a 46% drop. These machines do real work — inspection runs through chemical plants, substations, coal mines, pipelines. The use case is settled because the technology is.

The humanoids are the bet. H1, H2, G1, R1. The G1 is the one in the viral clips — 1.32 meters, 35 kilograms, doing push-ups on voice command. Manufacturing cost fell from $10,800 to $9,200 over the same window, a far shallower curve than the quads. Less mature technology, less scale, less cost compression. The machine that drives the narrative is the one furthest from profitability per unit.

Underneath both sits the thing that actually matters: the components. Motors, reducers, encoders, joint modules, dexterous hands, LiDAR, cameras — self-designed, self-built. The robots are the visible product. The parts stack is the real one.

So the product line is not a catalog. It is a hedge. The quads pay the bills with established industrial demand. The humanoids carry the valuation on a use case that does not exist yet. And the components let Unitree build both cheaper than anyone selling either.

Unitree's business model reads, on the surface, like any hardware company. Design robots, build robots, sell robots at a margin. The prospectus describes it in those terms.

It is not a hardware company in the way that framing suggests. It is a component manufacturer that sells finished machines as its distribution channel.

The economics live in vertical integration. Actuation — motors, reducers, joint systems — is 40–60% of a humanoid's bill of materials. Unitree self-designs all of it. Purchased parts run 14–18% of cost; domestic sourcing sits above 90%. Only commodity items get outsourced: battery cells, flash storage, the compute board. Everything that carries margin, it owns.

That ownership produces the signature move: margins rising while prices fall. Humanoid ASP dropped 71.7%. Gross margins climbed to 60.27%. Costs fell faster than revenue per unit, because each new product line reuses the same self-built components and the same production staff. Scale spreads fixed cost. The model compounds.

The demand side runs on discipline, not expansion. Production-based-on-sales — a monthly production-supply-sales meeting matching output to orders. Even with Spring Festival Gala traffic, Unitree did not build ahead. Production-sales ratio: 95.95% on humanoids, 98.56% on quads. No inventory backlog, no trapped capital.

So the model is not "sell robots." It is "own the parts, spread the cost, build only what is sold." The robot is the package the components ship in.

The vulnerability is the same as the strength. If actuators commoditize, the channel is all that is left.

Wang Xingxing – Founder, Chairman, General Manager and Chief Technology Officer

Wang Xingxing founded Unitree in Hangzhou in August 2016 and serves as its controlling shareholder, Chairman, General Manager, and CTO. Born in 1990 in Yuyao, Zhejiang, he developed his first quadruped robot, XDog, during his master's studies in mechanical engineering at Shanghai University, then left a position at DJI to build the company around it. Through the dual-class structure, he holds 23.82% of shares directly and controls a further 10.94% through the Shanghai Yuyi equity incentive platform, but his Class A shares carry ten votes each, giving him 68.78% of voting power before listing and an expected 65%-plus after.

Wang Feng – Chief Financial Officer

Wang Feng serves as Chief Financial Officer. The prospectus discloses that he holds no equity in the company and received annual compensation of RMB 466,800. He is one of only three senior managers named in the filing.

Fu Fenghua – Board Secretary

Fu Fenghua has served as Board Secretary since May 2025. Born in 1988, he holds a bachelor's degree. From September 2010 to May 2018 he worked at Tianjian Certified Public Accountants, rising from auditor to senior project manager. From June 2018 to May 2025 he served at Zhejiang Hangke Technology (688006) as deputy finance director, then finance director and board secretary, and ultimately deputy general manager.

Unitree's cap table reads like a roster of Chinese tech. Meituan at 9.6%. HongShan China, formerly Sequoia China, at 7.1%. Matrix Partners China at 5.5%. Behind them, through direct stakes or affiliated entities: Tencent, Alibaba, Ant Group, Xiaomi, ByteDance, plus the carmakers BYD and Geely.

The valuation trajectory frames the raise. Unitree's June 2025 round priced it at RMB 12.7 billion, roughly $1.7 billion. The IPO targets $3–7 billion. At the lower bound that is already a doubling in under a year; at the upper bound, a forward price-to-sales multiple in the low-to-mid 20s on 2025 revenue.

So the question the cap table raises is not whether Unitree can raise money. The strategic capital is already in. It is whether those same investors are hedging — backing the leader and the challengers at once.

The humanoid robotics market divides across hardware integrators, component suppliers, and the model layer, with global players, Chinese entrants, and capitalized cross-entrants competing at each. Unitree takes a different position: the volume leader, vertically integrated, shipping more humanoids than anyone at a price others cannot touch, on over 60% global share in quadrupeds.

The Obvious Competition

Global humanoid makers — Tesla's Optimus, Boston Dynamics' Atlas, Figure AI, Agility Robotics' Digit. Unitree competes on volume and price: 5,500 humanoids shipped in 2025 against rivals counting theirs in the hundreds, at a $16,000 entry price the cost structure of US peers cannot match.

Chinese peers — AgiBot, UBTech, Fourier Intelligence. UBTech reaches public markets alongside Unitree as the other pure-play benchmark. Unitree competes through manufacturing scale and iteration speed rather than any locked-in use case.

Cross-entrants — Xiaomi's CyberOne, plus carmakers BYD and Geely. These are the capitalized threat the prospectus names directly: balance sheets that can absorb a price war Unitree's margins cannot.

How Unitree Competes

The moat is cost, built on owning the components — not a use case. That is also the exposure. Unitree concedes its lead rests on cost-performance and iteration, not an irreplaceable scenario. Cheaper and faster wins until someone with deeper pockets is cheaper too.

Unitree's financials are the rare kind in frontier robotics: they make money. Revenue grew from RMB 121 million in 2022 to RMB 1.71 billion in 2025 — up 335% in the final year alone. The company turned profitable in 2024 at RMB 94.5 million net income, then expanded to RMB 288 million in 2025 on a GAAP basis. Adjusted net profit reached RMB 600 million, roughly 35% margins. Operating cash flow ran positive at RMB 672 million.

The number that does not behave like hardware is the gross margin. 60.27% for full-year 2025, up from 44.22% in 2023. Software companies live there. Hardware companies run 30–40%. Unitree sits closer to Kweichow Moutai than to anyone shipping physical product.

What makes it remarkable is the direction. Margins climbed while prices collapsed. Humanoid ASP fell 71.7%, from RMB 593,400 to RMB 167,600. Quad ASP fell 29%. Costs dropped faster than revenue per unit, because vertical integration compounds with scale — own the components, spread the fixed cost, reuse parts across product lines.

The margin story gets harder in 2026. Q1 broke the pattern: revenue grew 68%, but adjusted net profit fell 52% to RMB 40.3M, a 9.5% net margin against roughly 35% for 2025. The model spend is landing in the P&L. Two variables decide where it settles: R&D-plus-sales load, rising, and gross margin, falling. Best case holds near 37% net. Base case, with margin halving and the model ramping, lands near 21%. Bear case, price war plus full load, hits 8%, close to what Q1 printed. The valuation assumes the best row. The latest quarter came in on the bear.

Then the most recent quarter complicates the story. Q1 2026 revenue rose 68%, but adjusted net profit dropped 52% to RMB 40.3 million. R&D and sales spend running ahead of the top line — the model bet starting to show in the P&L.

So the financials are not the simple beat they read as. The base business is genuinely profitable. The question is what the margin looks like once the RMB 2 billion model investment is fully loaded, and whether 60% survives the price war the filing warns about.

Unitree files the strongest commercial record humanoid robotics has produced. Profitable, 60% gross margins, more humanoid volume than anyone. The filing is also unusually frank about what that record is not.

The bull case: this is the company that solved manufacturing. Vertical integration drives margins up while prices fall, the cost curve keeps compounding, and the RMB 2 billion model bet lands before the hardware lead erodes. Demand broadens from research labs into industry, the way quadrupeds already did. At $5.8B, an early entry into a $5 trillion market.

The bear case: the demand is rented. 74% of humanoids go to research. Q3 revenue fell 26% when one TV appearance faded. The moat is cost, and cost is copyable — BYD, Geely, Xiaomi sit on the cap table and the competitive set at once. Actuators commoditize like batteries, margins compress, and the model bet does not arrive in time.

Anuraag Gutgutia is the founder of True Foundry which provides provides an enterprise-grade AI Gateway that encompasses an LLM Gateway, MCP Gateway, and Agent Gateway, enabling enterprises to securely connect, observe, and govern access to models, tools, guardrails, and agents from a single control plane.

If you enjoyed our analysis, we’d very much appreciate you sharing with a friend.

Here are the options I have for us to work together. If any of them are interesting to you - hit me up!

What do you think about my bi-weekly Newsletter? Love it | Okay-ish | Stop it