Article by Alpha公社
An AI startup founded at the end of 2025, which has not yet publicly released any products, has raised $700 million in Series A funding, valuing the company at $6 billion. The round was led by Parkway Venture Capital, with participation from NVIDIA, AMD Ventures, Intel Capital, Qualcomm Ventures, and Salesforce Ventures.
It's clear that this company has secured substantial funding in a short time, with backing from the industry's top hardware and software giants.
The company Hark has a clear entry point: they aim to create the next generation of general-purpose human-machine interfaces using a combination of proprietary foundational models and customized hardware.
At its core, this is a new AI interface in the form of AI-native hardware—composed of a series of customized native hardware devices and intelligent computing systems equipped with end-to-end speech models and highly personalized memory capabilities. All these AI systems are multimodal, capable of understanding and interacting in natural ways.
We were not surprised to see Hark raise funds at a $6 billion valuation, with both NVIDIA and Qualcomm joining the round. Since 2024, Alpha公社 has been strategically investing in the “active AI” space—our early investment, Looki, has already sold multimodal AI wearables to users worldwide, becoming the global leader in shipments of multimodal wearable general-purpose intelligent devices; Guangfan Technology has independently developed a native AI operating system for smart hardware and pioneered the category of AI earbuds with visual perception capabilities.
Hark's massive funding further confirms a growing trend: the next decade of AI will not only be on screens, but also in the real world.
AI has become smarter, but it's still using old hardware and interaction methods.
Hark was founded by Brett Adcock at the end of 2025, initially funded with $100 million of his own capital. Brett Adcock previously founded companies such as Archer, Figure, and Vettery.
Archer entered the electric vertical takeoff and landing (eVTOL) aircraft market and successfully went public. Figure is a humanoid robotics company; in 2024, Figure raised $675 million, and in September 2025, it completed a C-round financing exceeding $1 billion, reaching a valuation of $39 billion, with investors including Jeff Bezos, NVIDIA, Microsoft, and OpenAI.
Why would Brett Adcock want to start a company in the direction of "active" AI-native hardware? Because Figure’s approach is inherently a systems engineering project combining AI, hardware, and real-world interaction—essentially aligned with the underlying tech stack of AI-native hardware, and he knows the challenges involved. Recently, Figure demonstrated a live demo of a robot performing extended package sorting tasks, indicating they have already solved some key issues.
Image source: Brett Adcock's official website
In addition to Brett Adcock, Abidur Chowdhury has joined Hark as Head of Design. He previously served as a senior product design executive at Apple, contributing to the design of products such as the iPhone and Air. Hark has also recruited engineers from Apple, Meta, Google, Tesla, and leading AI labs, covering AI research, hardware engineering, and design.
Looking back at the history of personal hardware terminal development, it is essentially a history of alternating advancements in hardware form factors, interaction methods, and applications: as hardware form factors and interaction methods evolve, they give rise to new applications, unlock new capabilities, and reach broader user bases.
For example, once the PC's form factor was finalized and its size became small enough, combined with the maturation of input interfaces like the mouse and GUI, it became easier for average users to adopt. When the internet became widespread, PCs transitioned from being used primarily by business and creative professionals to reaching the general public.
The next breakthrough came with the iPhone—not only did it integrate the capabilities of a computer and a phone into a compact form factor, but it also introduced multi-touch interaction, further lowering the barrier to entry and expanding the user base of smartphones (including tablets) to a level beyond that of PCs.
Moreover, its App Store ecosystem directly became the standard for software in the mobile internet era, facilitating approximately $1.3 trillion in developer revenue and sales globally in 2024.
Currently, the issue with AI is that while it possesses intelligence and strong software capabilities, it primarily operates and interacts through chat interfaces and non-AI-native devices such as computers and smartphones, lacking continuous memory of user identity and hardware specifically designed for intelligent interaction.
An initial industry consensus is that the next phase requires agent systems capable of naturally interacting with people and the real world. These systems should anticipate needs, reduce cognitive load, and operate like collaborative partners rather than waiting for commands like traditional software.
Currently, AI at the software level has already given rise to super startups like OpenAI and Anthropic, each valued close to a trillion dollars. Once AI-native hardware further develops, its impact on the tech industry could be on the scale of the iPhone.
However, maturing "active" AI-native hardware is a complex engineering endeavor. For example, Hark must build this entire system across multiple layers, including models, AI hardware, interaction, and memory.
First, their model will have agent capabilities, multimodal capabilities, and memory, enabling it to remember who the user is, what they have said, and operate across products and services the user has already used.
They will design AI-native hardware and integrate it with Hark’s foundational models. Given their hiring for real-time speech infrastructure roles, their interface is likely to begin with voice.
China-based startups have an advantage in developing "proactive" AI
Current AI, whether chatbots or agents, are merely tools for now, as they are confined to screens and only receive commands from users when needed, after which they deliver results.
Why is "active" AI important compared to these "passive" AIs? Because it transforms AI from a tool into a collaborator. AI can, to some extent, operate independently of humans—to think for them, act on their behalf, and complete tasks.
To build an "active" AI system, you need an AI-native hardware solution that integrates software and hardware, capable of perception and memory, endowed with intelligence, featuring new, lower-barrier interaction methods, and always available beside the user (always on).
During the early exploration phase of AI hardware (such as smart speakers), it had perception capabilities but could only store data, lacked intelligence, and offered rigid interactions.
In this new era of AI explosion, perception capabilities have further improved, and AI's memory and intelligence have made tremendous advances. Interaction is still being explored, but the path forward has already begun to take shape.
True "proactive" AI has taken a significant step forward.
For "active" AI to become increasingly mature, it is not a race of isolated breakthroughs—it requires the coordinated development and advancement of foundational models, agent operating systems, personalized memory, and hardware endpoints. Competition in AI-native hardware is a comprehensive endeavor.
In this field of innovation and exploration, Chinese startups have a greater chance of success, with three unique advantages: first, the manufacturing ecosystem advantage—places like Shenzhen boast the world’s most complete supply chain infrastructure; second, the market scale advantage—China is both the largest manufacturer and the largest application market; and third, the policy support advantage—the state has designated AI as a strategic priority, providing certainty for long-term investment.