HSG's Li Yannan: How to define smart hardware in the next decade?

What will the next decade of smart hardware look like?

During the 2026 WAIC, a roundtable forum themed "The 'Demise' of Smartphones and the 'Rebirth' of Robots" garnered widespread attention. Li Yannan, Managing Director of HSG, engaged in a captivating dialogue with multiple guests focused on the smart hardware sector. He shared his judgments and insights on topics including the future form of smartphones, the essential differences of AI-native hardware, and visions for new smart products.

The following is an organized transcript of his remarks:

Q: Over the past year, we have seen a plethora of non-smartphone form factor smart hardware. Is the smartphone as a product category being replaced, or is it being extended? Will smartphones still dominate smart devices in the AI era? What will the smart hardware ecosystem look like in the next decade?

I do not believe smartphones will be simply replaced. Smartphones will remain highly important, as they possess several irreplaceable advantages. They have the largest user base, the most mature supply chain, the strongest payment and identity systems, and the most complete developer ecosystem. Any new hardware that attempts to completely bypass smartphones will incur extremely high costs.

However, smartphones are evolving from "the sole entry point" to "a core node".

Over the past year, we have witnessed many non-smartphone form factors: AI glasses, AI earphones, companion robots, humanoid robots, and in-vehicle intelligent agents. On the surface, they all seem to challenge smartphones; but at a deeper level, they are essentially breaking down the functions previously undertaken by smartphones and deploying them in more natural scenarios.

The transformation in the AI era is that interaction will shift from "people seeking apps" to "agents understanding people". The past decade was defined by "all scenarios converging back to the smartphone"; the next decade may see "all devices collaborating around the user's agent". The smartphone will still serve as the hub, but not necessarily the only interface.

Therefore, new hardware opportunities do not necessarily lie in building a more powerful smartphone, but rather in creating an entry point that aligns more closely with human natural behaviors: glasses handle visual and first-person perspective tasks, earphones take charge of voice interaction and companionship, robots execute tasks in the physical world, vehicles provide mobile spaces, and smartphones continue to undertake identity verification, payment, privacy protection, computing power collaboration, and complex operations.

Specifically, there are three categories of opportunities:

The first category is "seamless interaction" hardware, such as AI glasses, earphones, and wearable devices. They do not replace smartphones, but rather take over a portion of smartphone usage time in high-frequency, fragmented, and instant scenarios.

The second category is "emotional and companion-focused hardware", such as companion robots and AI terminals for children and elderly care scenarios. In the past, smartphones connected people to information; AI hardware may connect people to emotions, trust, and long-term relationships.

The third category is "physical execution hardware", namely embodied intelligence and robots. Smartphones and glasses solve problems in the information world, while robots solve problems in the physical world. Truly significant changes will occur after AI evolves from "answering questions" to "completing tasks".

Thus, I believe the smart hardware ecosystem in the next decade will not be about one device replacing another, but rather form a multi-terminal, multi-agent, multi-scenario network.

Q: In the past two years, many new hardware products have emerged under the name of "AI native". From an investment perspective, when we discuss AI-native smart devices, what are their essential differences from the smart devices of the previous mobile internet era?

Previous-generation smart devices are "connected terminals", while AI-native devices are "intelligent organs". The mission of the former is to connect people to the network, while the mission of the latter is to endow intelligence with eyes, ears, hands, and feet. This positioning difference gives rise to four essential distinctions.

First, the interaction logic has changed: from "people adapting to machines" to "machines understanding people". The interaction of mobile internet devices is command-based — users have to learn the operation paths of icons, menus, and apps, which essentially means people adapt to the GUI. AI-native devices are intent-based — you do not even need to voice your intention, as the device can actively understand you through voice, vision, and environmental perception. This leads to a key corollary: previous-generation devices are passively responsive tools, while AI-native devices are actively perceptive agents. To determine whether a piece of hardware is truly AI-native, check if it can still function properly after removing the screen and apps.

Second, the capability curve has changed: from "peak performance at factory release" to "starting point at factory release". The capabilities of traditional consumer electronics are locked in by hardware specifications at the moment of manufacture, and they only depreciate afterward. The core capabilities of AI-native devices reside in the cloud and models, and they continue to evolve via OTA updates as models iterate — the AI glasses you bought last year may be twice as "smart" this year as they were when you first purchased them. This directly subverts the valuation logic: consumer electronics are evaluated based on shipment volume and gross profit, while AI-native hardware is measured by retention rate, usage duration, and subscription revenue. For the first time, hardware companies need to be assessed in the same way as software companies.

Third, the nature of data has changed: from behavioral data to perceptual data. The mobile era accumulated behavioral data such as clicks and browsing histories, which served recommendations and advertising. AI-native devices collect perceptual data including first-person vision, voice, and environment, which feeds back to models and forms the user's personal context. The moat of the previous generation was network effects and traffic entry points, while the moat of this generation is context — the more the device understands you, the higher the switching cost. This is why I say the value of the entry point will shift from that glass screen to the Agent behind it.

Fourth, the business model has changed: from "selling entry points" to "delivering outcomes". The business model of the mobile era was hardware profits plus traffic distribution — the app economy, advertising, and commission collection. The business model of the AI era is shifting toward the "razor-and-blade" model: the hardware itself may have slim profits or even be subsidized, and the real revenue comes from intelligent subscriptions and pay-per-task or pay-for-outcome services. This sets new requirements for investors' evaluation criteria: when assessing mobile hardware, we need to judge three dimensions — supply chain, cost, and channels; when evaluating AI-native hardware, we need to add one more dimension — intelligent supply, that is, whether the model capabilities have crossed the threshold of scenario availability and whether the token cost curve is declining fast enough.

Q: For hardware devices in the next decade, can you propose your most audacious vision?

Ten years from now, the very act of "buying hardware" will disappear. You will no longer need to purchase a device; instead, you will "employ" an intelligent agent, and the hardware will merely be its physical avatar.

Many great innovations originate from ideas that no one had previously envisioned. Just as we saw numerous robot products at WAIC, which seemed peculiar or even bizarre when they first appeared, they are now already working in many application scenarios.

Q: Who will hold the defining power for smart hardware in the next decade?

This is truly a challenging question. I do not have the answer myself. But I believe AI may unlock greater imagination than the mobile internet. At the same time, it may take as long a cycle as the mobile internet to fully mature. From the advent of electricity, through the transformation of steam engines, to the establishment of assembly line factories, nearly 40 years elapsed. The progress brought by the mobile internet has been rapid, and AI may advance faster than people expect — but it has only been three years since its widespread adoption, and there is still enormous room for future development.

Q: Regarding hardware devices in the next decade, what is a question that you are personally curious about but do not yet have an answer to?

We still do not know whether users will accept a device that can see, hear, and remember them for a long time; nor do we know whether this "AI that knows me best" will eventually emerge from a large enterprise that already exists today, or from a startup that has not yet been founded.

This article is from the WeChat official account "HSG Collective" (ID: Sequoiacap), author: Hong Shan, and published with authorization from 36Kr.