Hark Raises $700M at $6B Valuation to Build the 'Universal AI Interface'

In late 2025, Brett Adcock put $100 million of his own money into a problem he thought no one was solving correctly: building a truly personal AI. On Thursday, a constellation of some of the biggest names in chip manufacturing confirmed they agree with him. Hark, Adcock’s eight-month-old AI lab, announced it has raised more than $700 million in a Series A round at a $6 billion post-money valuation—one of the largest first institutional rounds in tech history.

The investors backing the round include Parkway Venture Capital as lead, alongside Nvidia, AMD Ventures, ARK Invest, Align Ventures, Brookfield, Greycroft, Intel Capital, Prime Movers Lab, Qualcomm Ventures, Salesforce Ventures, and Tamarack Global. The presence of Nvidia, AMD, Intel, and Qualcomm simultaneously in a single startup’s cap table is unusual and underscores how seriously the semiconductor industry views the race for AI-native hardware.

What Hark Is Actually Building

Hark’s official positioning is deliberately vague—the company describes its mission as building “personalized intelligence” through a “universal AI interface.” But the strategy, pieced together from Adcock’s statements, the company’s March 2026 launch announcement, and subsequent hiring signals, is more legible than the marketing language suggests.

The company is developing a full-stack AI system from scratch: foundation models, an application layer, and dedicated hardware, all designed to work together rather than being assembled from off-the-shelf components. This vertical integration is the company’s central architectural bet. Rather than building a wearable that runs GPT-4 or Gemini, or wrapping Claude around a consumer device, Hark is training its own models specifically optimized for the way it expects users to interact with AI—continuously, multimodally, and with persistent memory across sessions.

The AI system Hark is developing is intended to be proactive rather than reactive. Rather than waiting for prompts, the platform is designed to surface relevant information, make suggestions, and take actions based on accumulated context about the user’s life, preferences, and routines. It can process speech, vision, and text simultaneously.

This summer, Hark expects to release its first multimodal AI models. These will initially power a software platform that integrates with existing services and devices—phones, computers, smart home systems. Hardware purpose-built for Hark’s models is planned to follow, though the company has not disclosed specific form factors or a release timeline.

The Brett Adcock Pattern

Adcock’s track record makes the valuation less surprising than it might otherwise appear. He co-founded Vettery, a hiring marketplace that sold to Adecco for roughly $100 million in 2018. He founded Archer Aviation, an electric aircraft startup that went public via SPAC and became one of the more credible eVTOL manufacturers. He then founded Figure AI, a humanoid robotics company that raised $675 million from investors including Microsoft, Nvidia, and Jeff Bezos, and secured a manufacturing agreement with BMW.

The pattern is consistent: large capital raises, hardware ambitions, and silicon-level partnerships. Hark follows the same formula, but with a more software-centric starting point. Adcock’s thesis is that hardware without models is inert, and models without the right hardware fail to deliver the experience users actually want. Building both simultaneously is expensive—hence the scale of the Series A—but also defensible in a way that software-only AI companies are not.

The company has built its team to roughly 70 employees in under a year and secured computing infrastructure including a new Nvidia B200 data center for model training. The B200 is Nvidia’s current flagship data center GPU, used primarily by the largest AI labs and hyperscalers. Its presence in Hark’s stack signals that the company is training frontier-scale models, not adapting existing ones.

The Personal AI Race

Hark enters a crowded but still unsettled market for personal AI. The most direct precedent—and cautionary tale—is Humane’s AI Pin, which launched in 2024 to scathing reviews and was ultimately acquired for parts. The lesson Humane taught the industry was that AI-native hardware cannot succeed when the AI itself is mediocre. The hardware experience and the model quality are inseparable.

Several well-funded companies are working on related problems. Rabbit has shipped hardware; the results have been mixed. Meta’s Ray-Ban AI glasses have demonstrated genuine consumer traction, establishing that AI-augmented eyewear has a market. Apple’s ongoing work on camera-equipped AirPods—currently in advanced testing—suggests that the ambient, always-available AI form factor has mainstream interest.

What distinguishes Hark’s approach is the decision to own the full stack from model to device, avoiding the dependency on third-party AI APIs that hamstrung most first-generation AI hardware companies. The question is whether Hark can actually train models that are competitive with Claude, Gemini, and GPT-5 at a company of 70 people—and whether its summer 2026 release will demonstrate enough differentiated capability to justify the $6 billion bet.

The chip consortium backing the round suggests the semiconductor industry believes hardware-AI integration is the next frontier worth owning. Nvidia, Intel, AMD, and Qualcomm are not passive observers—they are betting that a new hardware category will emerge from the AI transition, and they want their silicon at the center of it.

Whether Hark becomes the company to define that category remains to be seen. But Thursday’s $700 million confirms it will at least get a serious attempt.