Yunqi Insights | What Does Truly AI-Native Hardware Look Like?
Without AI, the product simply doesn't work.
Over the past two years, AI hardware has re-emerged as a buzzword in primary markets.
From AI glasses, AI voice recorders, and smart rings to companion devices, sleep devices, exoskeletons, and home robots, new products keep appearing and new startup stories keep unfolding.
But this wave of enthusiasm isn't simply "hardware is hot again." Many of these products weren't that no one wanted to build them before — the user experience didn't work, the costs didn't work, the interaction model didn't work. AI has given some long-standing problems new product solutions.
Recently, Hao Liang, newly promoted executive director at Yunqi Capital in 2026, joined a 36Kr WAVES Summit panel titled "From Silence to Roar: Consumer Hardware Worth Getting Excited About" to share our observations on category opportunities, product moats, and investment theses for AI-native hardware. This edition of Yunqi Insights brings you the highlights.
The following is adapted from Hao Liang's remarks at the event.
AI Has Created New Category Opportunities in Consumer Hardware
Before the large model wave, consumer electronics had settled into a relatively stable product cycle.
After smartphones came TWS earbuds, then smartwatches. For a long time after that, no truly major new category emerged. Hardware innovation largely revolved around existing form factors, with relatively predictable upgrade cycles for users, supply chains, and brands.
AI is changing this rhythm. Liang noted that 2023 and 2024 might otherwise have been "down years" for consumer electronics in relative terms. But in reality, both brands and supply chain companies posted solid performance.
A key reason: AI has generated new demand and made previously intractable problems solvable.
This opportunity in AI hardware isn't just about model capabilities themselves. Chinese entrepreneurs have been able to seize this wave quickly thanks to more than a decade of accumulated industrial foundations**. Long-term development in consumer electronics supply chains, battery technology, MEMS, semiconductors, and cloud computing allows today's hardware startups to build product innovation on top of more mature engineering, manufacturing, and delivery capabilities.
The early Apple, Huawei, and DJI supply chains cultivated mature design, manufacturing, supply chain, and delivery capabilities; the maturation of large-scale cloud infrastructure also allows many AI hardware devices to offload inference to the cloud rather than being entirely constrained by on-device compute — the cloud handles the ceiling of capability, while the device handles the floor of user experience.
In other words, this wave of AI hardware opportunity isn't a localized change driven by a single technology variable, but the result of combined advances in model capabilities, supply chain maturity, cloud resources, edge hardware, and consumer use cases.
What's Truly Scarce: Product Definition Capability
As AI hardware heats up again, the market has seen a flood of new products, new funding rounds, and new narratives.
But heat doesn't equal PMF. Liang believes that among this batch of new-category AI hardware, only glasses and rings have achieved relatively clear product-market fit so far. Many other categories remain in early validation stages.
For comparison, mature consumer electronics categories ship in the hundreds of millions annually — over a billion smartphones, several hundred million TWS earbuds, smartwatches at scale. Many of today's well-funded, revenue-generating AI hardware companies are still shipping in the hundreds of thousands, tens of thousands, or even fewer units, with annual revenue mostly in the tens of millions to low hundreds of millions of RMB.
That's a solid start, but still a distance from true large-scale market validation. This is precisely why the core moat for AI-native hardware isn't just supply chain, isn't just brand, and isn't just any single-point technology — it's product definition capability.
Product definition capability doesn't mean stuffing an AI feature into hardware, or adding a large model interface to an existing form factor. What truly matters is whether a team can, based on new technical conditions, re-understand user needs, interaction patterns, and product boundaries, and define a new product that couldn't have worked before.
This is also a key difference between AI hardware and traditional consumer electronics.
Previously, hardware competition centered on industrial design, brand operations, supply chain efficiency, and channel capabilities. In AI-native hardware, the relationship between software, models, data, services, and hardware form factor becomes much tighter. Whether a product works no longer depends solely on leading hardware specs, but on whether it can create a continuously improving user experience.
Two Criteria for Judging AI-Native Hardware
In Liang's view, there are at least two criteria for determining whether a product is truly AI-native hardware.
First: without AI, does the product definition still hold?
If removing AI from a product leaves just ordinary hardware, it's likely closer to "AI-functionalized hardware" rather than true AI-native hardware. Genuine AI-native hardware should be a product whose definition only became possible after AI emerged.
Its core value, interaction model, usage scenarios, and service delivery should all change because of AI's existence.
Second: can it long-term, unobtrusively capture physical-world contextual information that phones cannot?
For over a decade, smartphones have been the most important gateway between people and the digital world. But phones have boundaries. They struggle to continuously, naturally, and unobtrusively perceive emotions, surrounding environments, physical states, action intentions, and richer physical-world context.
The opportunity for AI-native hardware likely lies precisely in these areas phones struggle to cover. When a hardware device can more naturally enter real-life scenarios, continuously obtain information from the physical world, and then use AI to understand, respond, and serve, it has the potential to become a new intelligent entry point.
So the value of AI hardware isn't just "AI inside hardware" — it's that AI gives hardware new perception capabilities, understanding capabilities, and service capabilities.
From Emotional Interaction to Physical Assistance
AI Is Reshaping What Hardware Can Do
In the panel discussion, Liang also shared two recent investment directions at Yunqi Capital.
One is NoonWake's "Good Luck Calendar." (Read more here). Calendars are a long-standing product form with inherent global appeal. Traditionally they've been static time-keeping tools, but with AI they can become hardware entry points for emotional interaction.
Users can communicate their current state to it, and it can generate more personalized responses based on the user's emotions and context — even producing a 3D-printed object related to the user's state through physical output. In other words, AI is giving a traditional "calendar" product the potential for emotional understanding, companion feedback, and physical output.
The other direction applies embodied intelligence-related technologies to exoskeletons and prosthetics. The value of such products isn't just enabling more natural walking, but further building AI perception and active power for complex life scenarios — for example, exercise at certain intensity levels.
What this requires isn't just hardware structure and mechanical capability, but also multi-dimensional perception, motion control, and intent understanding. Here AI isn't simple feature enhancement; it's changing the relationship between assistive devices and people.
From these two cases, we can see that AI-native hardware opportunities may be distributed across very different scenarios: some emerging from emotion, companionship, and daily life; others from body, movement, and real-world interaction.
But they point in the same direction: AI is making hardware into products that understand people more deeply, stay closer to scenarios, and sustain service over time.
Hardware Teams Need to Become More Like AI Teams
Another change with AI-native hardware is that iteration cycles are accelerating.
Hardware has always been more complex engineering than software. It involves design, prototyping, supply chain, production, delivery, after-sales service, and maintenance — iteration speed is inherently slower than pure software.
But AI is changing this pace. Liang mentioned that consumer electronics product lifecycles used to be 8 to 15 months; now some have compressed to 3 or 4 months. If a team needs 3 months just from design and prototyping to production, they'll likely be eliminated by competition.
This means next-generation AI hardware companies need not only product definition capability, but also the ability to truly leverage AI across design, production, marketing, and organizational collaboration to achieve faster feedback and iteration.
AI shouldn't just be a feature in the product; it should become a capability of the organization itself. For this generation of software-hardware teams, a true moat also includes the team's ability to integrate with AI. Not simply using AI as a tool, but whether the entire organization can evolve alongside AI.
In the End, It Comes Back to Product
The heat around AI hardware continues, but whether it lasts ultimately comes back to the product itself.
From our perspective, the key questions in evaluating an AI hardware project are: Does it solve a real need? Without AI, does the product still hold up? Can it capture physical-world information beyond what phones can access? Can the team build stable capabilities across product definition, supply chain delivery, and continuous iteration?
These questions matter more than "is it a hot trend." AI will bring a wave of new hardware products, but what users ultimately keep won't be a device that's better at telling stories — it will be one that's genuinely more useful, more natural, and more embedded in daily life.