The Truth About AI Hardware: Real Demand, False Propositions, and Moats | Unity Ventures AI Hardware Salon Highlights
The core of AI hardware entrepreneurship has always been the product itself.
AI is breaking out of the digital world, using intelligent hardware as its vehicle to reshape the physical one.
China's supply chain advantages provide uniquely fertile ground for AI hardware. Building on the infrastructure laid by large language models, replicating a device is no longer the hard part. What matters is whether you can meet real needs and keep users coming back.
Among all those seemingly sexy new scenarios, which represent genuine demand and which are false propositions? In this era of AI democratization, are the moats around hardware startups getting stronger or more fragile?
At a recent closed-door AI hardware summit hosted by Unity Ventures, Laimo Technology founder Wangshu Gao, AI glasses founder George, Ouropia founder Roger, and FOUNTAIN accelerator founder Tree revealed several overlooked truths about AI hardware entrepreneurship:
- The business model for AI hardware has shifted from pure "product sales" to "subscriptions."
- Building hardware has gotten easier, but building successful hardware has gotten harder — what matters is long-term operations and service capability, with data feedback as the key to product evolution.
- Don't get attached to selling points you think are important; build products users actually love.
- The features users select on surveys and the products they actually pay for are two very different things...
In this discussion, you'll see the industry's most honest logic laid bare. AI hardware entrepreneurship isn't about brute-force miracles, nor is it simply "software-hardware integration." It's a comprehensive battle of real demand insight, supply chain efficiency, and data loops fused with service experience.
Table of Contents:
- Wangshu Gao, Founder of Laimo Technology: Breaking Through in the Cutthroat Overseas Lawn Mower Market
- George, Founder of an AI Glasses Company: The Evolution and Positioning Battle of Smart Glasses
- Roger, Founder of Ouropia: Searching for a Soulful "Body" for Large Language Models
- Roundtable Discussion: AI Hardware's Moat Hasn't Disappeared — It Has Simply Moved
01
Wangshu Gao, Founder of Laimo Technology: Breaking Through in the Cutthroat Overseas Lawn Mower Market
Former co-founder & CTO of Cloud Whale, now founder of Laimo Technology, a smart robotic lawn mower manufacturer. Completed four funding rounds in one year with cumulative capital in the hundreds of millions of RMB, while also crossing 100 million RMB in revenue.
The Four-Part Cognition Framework: Reading Industry Trends, Analyzing Competitive Landscape, Clarifying Resource Advantages, Understanding Competitors
Before starting a company, you need clear cognition of your own strengths, the industry itself, and its future trajectory. Only then can you find your opportunity amid intense competition or narrow gaps.
Laimo chose robotic lawn mowers from day one — a fiercely competitive market, but one where we spotted underlying demand that hadn't yet been unlocked. Many manufacturers and even major brands that later tried to enter this space ultimately failed.
Our earliest four partners all came from technical backgrounds, skilled at R&D, product definition, and identifying selling points that resonate with users. So our chosen path was to build moats through product R&D and software, leading with novel product capabilities or creative selling points, then backfilling channel and supply chain resources.
When choosing your direction, you must be clear about how to maximize your strengths while avoiding competitors' advantage zones. It's nearly impossible to assemble a team without weaknesses from day one — learning as you go is the only way.
Product Philosophy: Build What Users Love, Not What You Love
Product capability always comes first — the hardest and most important thing in entrepreneurship. When you build a truly good product, you'll find that whether it's opening channels or placing ads, your results far outpace competitors.
Product is the 1; everything else is 0.
So how do you build a truly good product? Immerse yourself deeply in learning about your users. Have hours-long conversations with target users. Don't just talk about the product — discuss what sports teams they follow, what car brands they prefer, what lifestyle they aspire to. Only with deep understanding of these people can you hit their needs and preferences when designing products and crafting selling points.
After finding direction through intuition, verify with reason and data. Startups may lack resources for large-scale surveys or industry reports. You can use web scraping and models to capture and analyze existing data, cross-validating whether something represents genuine user demand.
Here's something deeply challenging about building products: you must build what users love, not what you love. Many entrepreneurs get attached to selling points they find cool or important, but deep user interviews may reveal that users actually care about "cheaper" or "easier to operate."
Avoid educating users whenever possible; product logic must follow existing user habits, then expand from there. The success of our first-generation product confirmed that good products let users immediately see the core value and problem being solved. When user perception aligns with your core selling point, marketing and promotion become much smoother.
Another daily fundamental: Use every consumer purchase to exercise your product thinking. When buying electronics, cars, or other consumer goods, ask yourself why you like this product and clarify your decision logic. Pay attention to online reviews — what drives likes and dislikes, economic capability or emotional value. Technical entrepreneurs especially need to step outside their own world and understand human nature.
02
George, Founder of an AI Glasses Company: The Evolution and Positioning Battle of Smart Glasses
Serial hardware entrepreneur with extensive industry background.
Three Waves of Smart Glasses
-
2014–2015 Market Education Period: Google Glass and Microsoft HoloLens were both B2B products. Though they ultimately failed, they educated the market on the concept of AR glasses. China's supply chain hadn't yet developed, so breakout products all came from abroad.
-
2018–2021 Domestic AR Glasses Boom: The Meta-Ray Ban smart glasses collaboration became the only phenomenon-level product with millions in sales. Meanwhile, maturing domestic supply chains gave rise to the "AR Four Dragons," with significantly improved product design over the previous generation.
-
2024–Present "Hundred Glasses War": Large models drove the shift from "AR glasses" to "AI glasses," sparking the "Hundred Glasses War" frenzy. Apple Vision Pro underperformed expectations, the industry lacked breakout hits, and the initial dust is settling.
Positions and Pros/Cons of Smart Glasses Players
-
AR Manufacturers: Industry accumulation and technical depth, pivoting hard toward AI + AR fusion.
-
Phone Makers: Hardware and supply chain advantages, complete developer teams, uniquely favorable position.
-
Internet Giants: Large model capabilities and app development strengths, but lack hardware DNA.
-
3C Accessory Makers: Following trends into the market, lacking AI and hardware technical accumulation.
Display Is the Critical Divide for AI Glasses
I divide smart glasses into two categories: camera-plus-audio glasses, and glasses with displays. Display is the most important factor, directly impacting information access efficiency, interaction experience, and the glasses' intelligence level. The technical difficulty of display is worlds apart from other functions. Balancing display module performance, weight, and cost is critical. Glasses must weigh in the tens of grams while managing display and battery life — a substantial hardware moat.
Users are already accustomed to 2D interaction on phones and other electronics. Glasses represent 3D spatial interaction — the generational leap beyond phones, with massive challenges in both R&D and market education. Most critically, there's PMF: distinguishing true scenarios from false ones. Scenarios that precisely solve real pain points remain scarce.
Only manufacturers that clear these hurdles will ultimately break through. Startup opportunity lies in avoiding giants' territory, going deep through small entry points, and pioneering niche scenarios. I believe smart glasses will end up as everyday-looking, comfortably wearable, affordably priced assistants that provide utility or emotional value in daily life, gradually crossing the chasm to become general-purpose products.
Startup opportunity lies in avoiding giants' territory, going deep through small entry points, and pioneering niche scenarios.
03
Roger, Founder of Ouropia: From "Bottle Spirit" to Real Companionship — Searching for a Soulful Body for Large Language Models
Former DJI product manager, power systems expert, and VP of Product at Cloud Whale — commercial operator behind multiple hit products.
In Goethe's masterpiece Faust, there is a spirit born inside a glass flask named Homunculus. He is omniscient from birth, possessed of pure wisdom, yet he suffers — because he is trapped in that transparent bottle, suspended in midair, unable to touch real dust. His ultimate wish is to shatter the glass that protects him, to plunge into the real world, to endure wind and rain, to possess a real physical body.
Today, AI's soul is urgently seeking a habitat in the physical world — this has become explicit consensus within the industry. But the problem before us is: what should this vessel that carries intelligence actually look like?
In the past we tried to define AI hardware through voice interaction. But before the gravitational pull of the smartphone — this "black hole-class" terminal — and China's extremely mature supply chain, pure feature stacking has proven unable to build real moats. Since we cannot beat the phone on efficiency and universality, we must think in a different dimension, seeking a foothold in "affection" beyond mere "utility."
In the realm of emotion, there is no "standard product" that can please everyone. The extreme personalization and delicacy of companionship demand means it absolutely cannot be built like a phone — a "bucket machine" that does everything adequately. Surveying the current market spectrum, from tool-heavy "smart butlers" to emotionally overloaded "virtual lovers," we discovered a paradox: too shallow a connection and you're easily forgotten; too deep and it feels burdensome.
To break this binary opposition, our long-running focus has been on deconstructing the essence of "companionship," finding a balance point between light and heavy that can carry high emotional value — what we define as "presence-based companionship."
Technology has finally handed us the key to break species' creative constraints. The explosion of the AI technology cycle lets us surpass carbon-based biological limits, making penguins, pandas, even fantasy creatures possible companions, massively expanding category boundaries.
This breakthrough isn't merely visual richness — it's about deep relationship restructuring. The core moat of presence-based AI pets lies not in what they look like, but in how they let users cross species boundaries to generate genuine emotional projection and resonance. This is also the core problem Ouropia is committed to cracking in this niche track.
To inject soul into machines, we must grant them "cyber subjectivity." This means true companionship cannot stop at a machine that merely responds to commands — it must evolve into a lifeform with independent will. We seek to strip away mechanical interaction, establishing this "sense of life" through flowing gazes, subtle body language, and externalized unique personality, letting users deeply feel that what sits across from them isn't cold code, but a vivid, opinionated "other."
Perfection belongs to tools; imperfection is life's privilege. Based on this pursuit of life texture, I've poured enormous effort into crafting AI pets' "eye contact" interaction, because the most intuitive evidence of life's existence often occurs in an unguarded moment of mutual gaze.
We also insist that good companion AI must have "personality" — it shouldn't be an obedient, perfect assistant. It should be "imperfect." It's these unpredictable small flaws that give it the granularity of real life.
The future of AI companionship is an exploration of the breadth and depth of human emotion. It will respond to our softest inner needs with imperfect warmth, letting every lonely soul find echo in the digital world.
We set out searching for a "body" for large language models, and ultimately discovered we weren't merely building a shell — we were constructing a mirror that reflects the human heart. The future of AI companionship is an exploration of the breadth and depth of human emotion; it will respond to our softest inner needs with imperfect warmth, letting every lonely soul find echo in the digital world.
In Faust, Homunculus, to embrace reality, actively crashes into the sea — the shattering of that glass flask is his birth cry.
Everything we do today is merely the first crack in that glass. Where this new species will take human emotion, I'm uncertain. But I'm certain of this — when it first looks at you, you'll hear the answer.
04
Roundtable Discussion: AI Hardware's Moat Hasn't Disappeared — It Has Simply Moved
In the era of AI democratization, are hardware startups' moats getting stronger or more fragile?
Wangshu Gao: Hardware catches users' eyes, but software and service keeps them. Software is the long-term moat. The market gives entrepreneurs less and less time to iterate — you have to get every link right in a short window.
George: Supply chain is relatively easy; the hard part is subsequent service experience. Many startups build a hammer they personally like from an industry and technology perspective, then go looking for nails — rather than starting from delivering good service experience.
Roger: Open-source models have unleashed software capabilities, so small companies with product capability can also create new species. From this angle, some hardware moats are disappearing. In the software-hardware integration space, spatial intelligence remains a moat.
Tree: If we view large models as supply chain, this supply chain has broken the clear boundary between digital and physical worlds. AI hardware puts a "hard shell" on large models; AI software puts a "soft shell" on them. From this logic, the boundaries between past hardware categories are disappearing. The core logic of hardware development has become using large model companies as core supply chain, delivering new interfaces for intelligence at the front end.
As the physical interface for large models, AI hardware and large models mutually reinforce each other. The stronger the model, the greater the hardware demand — it won't be eaten by large models.
Contrary to many people's perception, first-mover advantage in AI hardware creates obvious moats. Software subscription revenue enables sustained income, allowing you to compress hardware margins and leave latecomers no profit space. You can also use user experience feedback to more accurately find direction and pull away from competitors. This model optimization based on accumulated user data is the true "data flywheel" of the AI hardware era.
How do you judge whether a new scenario is opportunity or trap?
Wangshu Gao: Cross-verify every question with multiple data sources, analyzing more rationally. Some products have low penetration ceilings because the pain point itself isn't actually painful. Scenarios with clear pain points succeed more easily. Some categories merely have insufficient product design constraining demand — once you break through user perception or demand barriers, volume ramps extremely fast.
George: Distinguish strong demand from weak demand across many scenarios. Don't fully believe user interview questionnaires or open-ended answers — interest and actual payment are two different things. Don't just test a few prototypes; mass production delivery is what matters. The only path to validating demand: from product definition to mass production, then user feedback and product iteration. Without going through this closed loop, scenario judgment is just armchair theorizing.
Roger: Comprehensively observe users' life trajectories, plus psychological and social mechanisms, making commonsense inferences. If your product interacts and resonates with user behavior, genuinely changing their lives, it's more likely "real demand."
Tree: Beware of products that sound especially "big and comprehensive." Though entrepreneurship should seek high-ceiling directions, real and massive demand is extremely scarce. If you find a small-market need, don't brainwash yourself or manufacture illusions.
The United States is the world's largest AI software and consumer market, but consumer electronics in the US has been broken for over a decade. US VCs don't invest in hardware, and there's no local hardware talent.
From product definition and supply chain capability perspectives, China-background teams have advantages. But watch for data compliance risks — future compliance requirements may tighten, with data-related scrutiny. This is a "minefield" Chinese teams going overseas must avoid in advance. US-incorporated company, China-based supply chain, localized compliance and scenarios — this is a strategy many teams have already validated.
A Closed-Door Discussion on PMF Validation and Overseas Growth | Unity Ventures AI Salon
