Brett Adcock, the CEO of $39 billion humanoid robotics company Figure AI, publicly launched Hark on Tuesday, an AI lab he has been quietly funding with $100 million of his own money since late 2025. The San Jose company has 45 engineers and designers pulled from Apple, Meta, Google, and Tesla, and it plans to ship its first AI models this summer.
The pitch is one Silicon Valley has heard before: build a personal AI that actually feels intelligent. Persistent memory, multimodal understanding, real-time interaction. But two details make Hark worth watching beyond the press release.
The designer who left iPhone behind
Hark's most conspicuous hire is Abidur Chowdhury, the London-born industrial designer who led Apple's team on the iPhone Air and presented the device during Apple's keynote. Chowdhury left Apple last fall to join Adcock's lab as Director of Design, which is the kind of move that either signals deep conviction or a very persuasive pitch over dinner.
In an interview with TechCrunch, Chowdhury was careful not to reveal specifics but did volunteer some pointed skepticism about current AI hardware. He said he's "not the biggest believer in a lot of the wearable AI platforms," and called out pins and cameras strapped to faces as uncomfortable. That's a swipe at Meta's Ray-Ban glasses and the wreckage left by Humane's AI Pin, and it raises a question: if not a wearable, then what?
Chowdhury wouldn't say. He framed Hark's design philosophy around removing barriers between intelligence and users rather than adding new gadgets to carry around. The comparison to Jony Ive's work at OpenAI is obvious and unavoidable. Both are former Apple designers. Both are building AI-native hardware. Hark's spokesperson declined to discuss Ive, which tells you they've thought about the comparison plenty.
The NVIDIA compute bet
According to the company's press release, Hark has signed a deal for thousands of NVIDIA B200 GPUs, with the cluster coming online in April. Jensen Huang offered a quote calling personal AI the next frontier, which is the kind of endorsement NVIDIA hands out selectively these days.
"Bringing that vision to life requires enormous compute to build powerful multimodal foundation models," Huang said, and sure, that's boilerplate from NVIDIA's CEO, but the B200 allocation is not. GPU access remains the bottleneck for every AI startup, and getting a cluster of that scale suggests either strong connections or a compelling roadmap. Probably both.
The plan, as Hark describes it: build foundation models, software, and hardware together from scratch. Vertically integrated, in the way that only a company spending someone else's (or in this case, their own founder's) $100 million can attempt before having revenue or users.
So what's the catch?
A few things are worth noting about what Hark hasn't said. There's no product demo. No model benchmarks. No hardware renders. The summer timeline is for AI model access, with hardware to follow at an unspecified date. That's a lot of vision and not much artifact.
Adcock is also running two companies simultaneously. He remains CEO of Figure AI, the humanoid robotics startup valued at $39 billion after its Series C last year. Hark's models are already training on Figure's robots, according to TechCrunch, though the companies say there are no plans to merge. Both operate on the same campus. Whether one person can effectively run a humanoid robotics company and a personal AI startup at the same time is a bet Adcock is making with his own capital, which at least aligns his incentives.
The $100 million seed, all from Adcock's pocket, is significant but not enormous by 2026 AI standards. The Information first reported the lab's existence back in December. The team has grown from around 30 engineers in January to 45 today, with a target of 100 in the first half of this year.
Where this fits
Hark joins a growing and increasingly crowded field of companies trying to build the AI device that people actually want to use. OpenAI is developing hardware with Jony Ive's team, expected to be unveiled later this year. Meta keeps iterating on Ray-Ban smart glasses. Humane and Rabbit proved that getting AI hardware wrong is expensive and public.
Chowdhury told TechCrunch the opportunity reminds him of when the iPhone first arrived. That's a big swing from someone who would know. Whether Hark can deliver anything close to that kind of shift with 45 people and no shipped product is the question that matters, and it won't have an answer until summer at the earliest.
