10 m/s Sprints, $20B Fabs, and the AGIBOT Era

This week, the "soon" of humanoid robotics felt a lot more like "now." We witnessed a collision of massive capital and record-breaking performance, from Unitree’s H1 reclaiming the speed crown to Elon Musk’s "Terafab" manifesto. The industry is no longer just building cool machines; it’s building the massive semiconductor and data infrastructure required to make them a permanent fixture of our global economy.

The most telling shift, however, isn't just in how fast these robots run, but in how they are being deployed. We’re moving into a "Software 2.0" era where companies like Figure are deleting manual code in favor of end-to-end neural networks. As the hardware becomes a commodity you can buy on AliExpress, the real war is being fought over who has the biggest compute war chest and the most diverse physical datasets.

Unitree Robotics has officially returned to the top of the leaderboard after releasing footage of its H1 model hitting a peak velocity of 10 m/s (approx. 22.4 mph). This 200% increase over previous records was achieved on an outdoor athletic track, matching the benchmark set by MirrorMe’s Bolt.
Read more here

Elon Musk has detailed the "Terafab," a massive semiconductor complex in Austin designed to produce 1 terawatt of annual compute. Partnering with Intel, the facility aims to solve silicon bottlenecks for the Optimus program while exploring space-hardened chips for orbital data centers.
Read more here

Figure CEO Brett Adcock announced the acquisition of an entire data center's worth of NVIDIA Blackwell GPUs to scale robot learning by "orders of magnitude". This massive compute injection supports Figure’s strategy of using end-to-end neural networks to master "pixels-to-torque" autonomy.
Read more here

In a major step toward global retail accessibility, Unitree has officially listed the R1 humanoid on AliExpress. The listing includes the EDU version, which offers open control interfaces and up to 40 degrees of freedom, though it carries a significant "convenience premium" over domestic Chinese pricing.
Read more here

CEO Pete Florence is distancing Generalist AI from industry buzzwords, arguing that their GEN-1 model is a native physical foundation model trained from scratch on 500,000 hours of data. The firm recently demonstrated "intelligent improvisation" through complex tasks like two-handed zipping and ethernet cable plugging.
Read more here

Kia Corporation has committed $36 billion to future business, including a definitive timeline for deploying the production-ready Atlas robot. Full-scale operations are slated for the Hyundai Metaplant in Georgia by 2028, focusing on 16 core manufacturing processes.
Read more here

Palo Alto-based startup Syncere is taking a different approach to domestic robotics with Lume, a sculptural floor lamp that hides a laundry-folding arm in plain sight. Priced at $1,499, the "ambient" robot aims to automate chores without the "Terminator vibe" of a traditional humanoid.
Read more here

Manufacturing giant Luxshare Precision has joined the cap table for EngineAI, leading a funding round that values the company at $1.4 billion. The partnership provides the supply chain expertise needed to hit EngineAI's aggressive target of delivering 5,000 units in 2026.
Read more here

To conclude their ambitious AI Week, AGIBOT announced the world’s first large-scale industrial deployment of embodied AI in consumer electronics. Working with Longcheer Technology, AGIBOT integrated G2 robots into live tablet production lines in Nanchang, achieving a 99% success rate and a throughput of 310 units per hour.

Over the course of the week, AGIBOT unveiled a complete, self-reinforcing stack: the AGIBOT WORLD 2026 open-source dataset, Genie Sim 3.0 for generative simulation, the GO-2 foundation model, the Genie Envisioner 2.0 world simulator, and the zero-code Genie Studio Agent. Together, these technologies form a closed-loop infrastructure designed to move robots from laboratory novelties to essential industrial infrastructure.

Explore the full AGIBOT AI Week Hub