Unitree Robotics priced its IPO at RMB 150.80 a share, valuing the Chinese robot maker at roughly RMB 61 billion, while DeepSeek poured more than RMB 140 million into the offering. At the same time, U.S. startup Hadrian closed a $1.4 billion financing round that lifted its valuation to $7.9 billion in just seven months. The twin headlines signal a flood of capital into “physical AI” – the blend of machine-learning software with real-world robotic hardware – and hint at a market that could reshape manufacturing, logistics and consumer services within a few years.

Why the money matters now

Robotics has long been a capital-intensive sector, but the recent deals show investors are betting that smarter, more adaptable machines will finally break out of laboratory prototypes. Unitree’s IPO is the first large-scale public offering from a Chinese firm focused on agile, dog-type robots that can navigate complex indoor spaces. DeepSeek’s stake, though a modest slice of the total raise, matters because the venture fund backs AI models that run on edge devices – the compute needed for on-board perception.

Hadrian’s raise does something similar on the other side of the Pacific. The company builds modular manipulators for warehouse and factory lines, and its rapid valuation climb suggests that customers are ready to replace static conveyor belts with flexible, AI-driven arms. Together, the two rounds illustrate a broader shift: investors see physical AI as a revenue engine, not just a research curiosity.

The market backdrop

China dominates the patent landscape for humanoid body structures, holding 73 % of global filings. The sheer volume shows in the 40,000+ humanoid robots produced in the first half of 2026. Yet less than 5 % of those units sit on actual factory floors; most linger in labs, demo stages or entertainment venues. The United States, while trailing in patent counts, leads in per-patent quality, suggesting a complementary ecosystem where Chinese manufacturing capacity meets American algorithmic expertise.

South Korea plans to field 1,000 AI-enabled robots annually across ten major industries, adding another layer of demand. If the rollout proceeds, it will create a steady stream of orders for sensors, actuators and the AI software that interprets sensor data in real time.

Technical leaps that justify the bets

A handful of breakthroughs reported this week underline why capital is flowing. DreamWAM, a learning framework, lifted real-robot success rates in unseen scenarios from 55.6 % to 74.4 %, showing robots can now generalize to new environments without costly re-training. BridgeVLA++ injects spatiotemporal memory into 3-D models, letting a robot “remember” where an object was even after it disappears – a capability essential for bin picking or inventory audits.

Mind-VLA narrows the focus of vision-language models to specific target objects, sharpening fine-grained manipulation. Tactus pairs inexpensive pressure sensors with natural-language queries to identify objects, lowering the entry barrier for tactile perception. Deltoris accelerates diffusion-based vision-language architectures by 34.2 × compared with typical mobile GPUs, making high-quality visual reasoning feasible on embedded hardware.

These advances shrink the performance gap between simulation and the messy physical world, a gap that has long limited robot adoption beyond highly controlled settings.

Bottlenecks that could slow the surge

Even with soaring valuations, the sector faces hard constraints. Transmission components – harmonic reducers and lead screws – account for 68 % of the bill of materials for robots such as Tesla’s Optimus humanoid. Their high precision requirements and short service lives keep unit costs high and supply chains fragile. Scaling production while maintaining micron-level tolerances needed for smooth motion remains a major engineering hurdle.

The pressure worsens because existing humanoids sit idle. If only a handful of robots per thousand actually work in factories, manufacturers can’t justify the investment, and capital may shift to more immediately profitable automation, such as fixed-axis pick-and-place arms.

What to watch next

  • La roadmap post-IPO di Unitree – L'azienda si è impegnata a espandere la propria linea di prodotti oltre i quadrupedi, puntando a piattaforme più grandi. Monitorare i libri ordini e gli annunci di partnership rivelerà se il capitale si tradurrà in quote di mercato.
  • La scala di implementazione di Hadrian – Il prossimo round di finanziamento, se avverrà, sarà probabilmente legato a contratti su larga scala. I primi successi nei centri di distribuzione e-commerce potrebbero fungere da modello per altri settori.
  • Riforme nell'approvvigionamento dei componenti – Qualsiasi progresso nella produzione di trasmissioni a basso costo e alta precisione – magari attraverso tecniche additive o nuovi materiali – ridurrebbe i prezzi dei robot e accelererebbe la loro adozione.
  • Ambiente normativo – I governi di Cina, Stati Uniti e Corea del Sud stanno elaborando politiche relative alle macchine basate sull'IA. Cambiamenti negli standard di sicurezza o nelle tariffe di importazione potrebbero spostare l'equilibrio dei costi tra fornitori nazionali ed esteri.

In sintesi

La contemporanea quotazione in borsa di Unitree Robotics, l'investimento strategico di DeepSeek e la raccolta di 1,4 miliardi di dollari di Hadrian segnalano che i mercati dei capitali ritengono che l'era dell'IA "fisica" stia arrivando prima che tardi. Il progresso tecnico sta riducendo il divario tra percezione e azione, ma i costi consolidati della catena di approvvigionamento e i modesti tassi di implementazione nel mondo reale frenano l'ottimismo. La traiettoria del settore dipenderà dalla capacità dei produttori di trasformare queste macchine più intelligenti in asset redditizi prima che i colli di bottiglia dei componenti e gli ostacoli normativi ostacolino lo slancio.