Brett Adcock has done this before. He co-founded Vettery, sold it to Adecco. He founded Archer Aviation and took it public. He founded Figure AI and built one of the best-funded humanoid robotics companies in the world. In late 2025, he seeded a new company called Hark with $100 million of his own money and kept quiet about it for roughly two months. On Thursday, Hark emerged with a $700 million Series A at a $6 billion post-money valuation — one of the largest early-stage financings in AI this year — and a thesis that the entire consumer AI stack needs to be rebuilt from the ground up.
The round was led by Parkway Venture Capital and included NVIDIA, AMD Ventures, Intel Capital, Qualcomm Ventures, Salesforce Ventures, ARK Invest, Brookfield, Greycroft, Prime Movers Lab, Align Ventures, and Tamarack Global. The presence of all three major GPU vendors — NVIDIA, AMD, and Intel Capital — on a single cap table is notable. It reads less like a typical investor syndicate and more like a supply chain insurance policy, suggesting Hark has already had serious conversations about allocation and silicon access at scale.
The Argument Against Chat
Hark’s founding premise is that the dominant AI delivery mechanism — a chat window on a smartphone or laptop — is a transitional artifact, not a destination. The company argues that hardware built before AI’s emergence cannot serve as its native interface, and that models without persistent memory cannot function as genuine personal intelligence. Both observations are defensible. What distinguishes Hark’s response from prior attempts is the decision to address both problems simultaneously through a vertically integrated stack: foundation models, software infrastructure, and purpose-built hardware developed in parallel rather than in sequence.
This is the Apple thesis applied to AI. Build the chip and the operating system and the device and the application layer together, so each layer can be optimized in relation to the others. The argument for vertical integration is that it produces a better product. The argument against it is that it produces a slower company.
A Category With a Body Count
The AI hardware space has produced more cautionary tales than successes. Humane’s AI Pin launched in 2024 to near-universal critical rejection and the company sold its assets to HP by mid-year. The Rabbit R1 shipped to early backers and delivered a software experience that failed to justify the device. Neither product died because the underlying vision was incoherent — the vision of ambient, wearable AI with persistent context is coherent — they failed because the model capability and the hardware form factor arrived out of sync, and because neither company had enough capital or time to close the gap.
Hark has more capital than either predecessor and a founder with a track record of shipping physical products. Adcock built and flew an electric aircraft at Archer. He put a humanoid robot into a BMW factory at Figure. The argument for Hark is not primarily the pitch deck. It is the operational history.
What Remains Undisclosed
Hark has not disclosed its hardware form factor, target price, headcount beyond approximately 70 employees, its launch market, or a customer pipeline. The summer model release will be multimodal, operating on existing consumer devices before the proprietary hardware ships. The sequencing — models first, hardware later — mirrors what nothing in the prior wave of AI hardware attempted. Whether software traction translates into hardware adoption is an open question.
What Adcock has noted publicly is the positioning logic: Anthropic is prioritizing coding tools, OpenAI is moving toward enterprise and IPO readiness, and neither is focused on the consumer personal intelligence layer the way Hark intends to be. That gap is real. It is also contested by every major consumer hardware company on earth, including Apple, Google, Meta, and Samsung, all of which have announced or shipped AI-native hardware in various stages of development.
The $700 million buys time to figure out what no one else has yet figured out: how to build a device that captures intimate contextual data at scale without making every person in the user’s immediate environment a nonconsenting data subject. That is not an engineering problem. It is a social one, and it will require a product answer.
Hark has the capital and the founder to attempt it. The field is littered with companies that had one or the other.
Leave a Reply