XPeng's robotics unit raises $900M to chase Tesla in humanoid robots
XPeng's robotics unit raised $900 million in its first round — a shot across Tesla's bow in the humanoid race. The Chinese EV maker said its robotics business pulled in the money from IDG and Gaorong Ventures at a valuation of more than $6.3 billion, with Tencent and Alibaba on the cap table. It's one of the largest funding rounds for an embodied-AI startup in China to date, and it lands days after Unitree, another local robotics maker, priced a humanoid IPO on the mainland.
Why it matters: XPeng is betting that the same playbook that built its EV brand — aggressive capital, tight in-house integration of hardware and software — can crack physical AI. The unit's IRON humanoid is positioned to go after Tesla's Optimus, and the round gives XPeng a war chest that most rivals simply don't have at this stage. That Tencent and Alibaba joined signals China's biggest platforms see humanoid robots as a strategic layer worth owning, not a side experiment. For the broader industry, the scale of a first-round raise at this valuation is the clearest market signal yet that investors are treating the humanoid race as a race worth backing, not a science project.
What to watch: whether XPeng starts shipping IRON in volume before Tesla, and how the company prices its robots against a crowded field of Chinese challengers.
Nvidia is moving to bring CUDA to RISC-V server chips — a quiet step that could widen who gets to feed Nvidia's GPUs. At Hot Chips this week, the company laid out the requirements RISC-V CPUs would need to meet to run CUDA workloads, from a server-grade platform and ACPI support to PCIe coherency and peer-to-peer transfer. Nvidia is partnering with SiFive, which plans to demo a CUDA-capable system at the conference, and it's also opening its NVLink Fusion program so partners can license Nvidia's interconnect IP for custom accelerator designs.
Why it matters: CUDA's lock-in is a big part of Nvidia's moat, so opening a path onto RISC-V — the open-standard architecture that China and other export-sensitive markets are pushing hard — is strategically loaded. But the fine print matters: the requirements are steep, and Chips and Cheese notes most existing RISC-V hardware won't qualify, meaning early support will land in server systems rather than hobbyist boards. If Nvidia follows through, it blunts the main argument for RISC-V accelerators by taking away their software disadvantage; if it stays demanding, it keeps the ecosystem at arm's length.
Do you think XPeng's cash pile gives it a real shot at Tesla in humanoid robots, or is embodied AI still too early to crown a leader? Tell us in the comments.
