While everyone's watching his humanoids pack boxes, Adcock's quietly building the device that makes the robot obsolete.
The Summary
- Figure AI CEO Brett Adcock's new startup Hark raised $700M at a $6B valuation to build personalized AI models and hardware devices, funded by Nvidia, AMD, and Salesforce Ventures
- Adcock seeded the company with $100M of his own capital late last year, then ran a Series A before most founders finish their pitch deck
- The play is personal AI that "knows you" plus hardware to run it, rolling out models this summer with devices on an unspecified timeline
The Signal
Brett Adcock runs Figure AI, the humanoid robotics company that just put machines on BMW assembly lines. But he founded Hark late last year with his own money while Figure was still proving its robots could sort packages. That's not hedging. That's someone who sees the curve bending.
The thesis: personalized AI needs its own hardware. Not a chatbot in your browser. Not a voice assistant piggyback on your phone. A device purpose-built for an agent that actually knows your context, your preferences, your patterns. Figure's robots taught Adcock how to make AI interact with the physical world. Hark is the consumer version of that lesson.
"The AI that everyone deserves but no one has built yet — one that actually knows you, speaks your language, is highly personalized, and lives on hardware made for you."
Adcock hired Abidur Chowdhury, who designed the iPhone Air at Apple, plus hardware and AI engineers from Google, Meta, and Amazon. That's not a research team. That's a shipping team. The models launch this summer. The hardware timeline is vague, which probably means it's closer than they're saying. You don't raise $700M to iterate on whitepapers.
The competition is obvious and brutal. OpenAI is building devices with Jony Ive. Apple, Google, and Meta are all chasing the same territory. But none of them bootstrapped with $100M of founder capital, then closed a $700M Series A six months later. That kind of velocity means either the demos are very good or the pitch is very clear. Probably both.
Here's the hidden edge: Figure's robots already use Hark's tech. That's not mentioned prominently, but it's buried in the source material. Adcock isn't starting from zero. He's productizing infrastructure that's already running in commercial robotics deployments. The personalized AI models Hark is building have been trained on real-world interaction data from Figure's systems. That's a dataset advantage no pure consumer AI startup has.
Key dynamics to watch:
- How fast Hark's hardware ships after the model launch this summer
- Whether the device is standalone or needs a phone to function (that choice reveals everything about the strategy)
- What "personalized" means in practice — local data processing, federated learning, or just better fine-tuning on user behavior
The Implication
If Hark's hardware lands well, the AI agent race shifts from software to form factor. The winner won't be the best model. It'll be the model that lives where you actually are, running on silicon designed for it. Adcock's building the opposite of the cloud-first AI consensus. He's betting that Web4 agents need bodies, not just for factories but for people.
Watch who buys the first devices. If it's developers and power users, this is another toy. If it's non-technical people who just want their AI to work without thinking about it, Adcock just skipped three product cycles.