63. Interview with Rokid’s 祝铭明: AI Glasses, the Entry Point, and the Ecosystem War with Giants Closing In

Summary

• 祝铭明 sees the essence of the 2025 AI-glasses boom not as a product inflection, but as an inflection in mass awareness: once AI became a global social consensus, “ordinary people at least knew what they could do with a pair of smart glasses.” The key data point is that the share of Rokid users working in broadly defined tech fell from more than 50% to 16%, showing that the category has broken out of its niche; he gave investors a clear warning: “When an industry is growing by more than 2x, stop questioning its growth.”
• With Xiaomi, Baidu, Alibaba, ByteDance and Li Auto all entering the market, Misa says “so far, none of them has made Rokid nervous,” applying Jack Ma’s 12-character framework: “can’t see it, can’t understand it, don’t think much of it, then it’s too late.” Large companies are splitting resources down the middle even on their core battlefields; “the competition is never really between a big company and a small company, but between big companies fighting each other to the death,” leaving very few with bandwidth for edge-market disruption. The eventual contest is “a duel between 2 elite teams”; Rokid has deployed 400-500 top industry talents and believes its investment density is the highest.
• Misa says he “knew on day 1” that Alibaba was working on Quark Glasses—roughly 1 year before the project started—and is not worried about a conflict with angel investor 吴泳铭: if Alibaba refuses to open up its AI strategy and pins its future on Quark Glasses, “Alibaba will die; why should you care about a toe on the body of something that is going to die?” The industry is still racing to stake out territory rather than playing a zero-sum game: Rokid’s sales doubled on the day of Alibaba’s launch event. True zero-sum competition will not begin until industrywide shipments exceed 100M units, likely 3-5 years from now.
• His verdict on the Doubao phone Agent is “the GUI era’s last stand”: the direction is inevitably NUI (Natural User Interface), while a GUI Agent “can be shut down at any time by any major super-app.” His more counterintuitive view is that Alibaba and Tencent stand to benefit the most because they can quickly close the loop on payments, shopping, navigation and food delivery inside their own ecosystems, while ByteDance is “one-dimensional,” lacking payments, logistics and social networking—“just as search engines once could not enter e-commerce.”
• Rokid’s product philosophy is “Say No”: it insists on a single optical engine rather than Quark’s dual-engine design, sacrificing focal-plane adjustment for 1-2 hours of extra battery life and several hundred to RMB1,000 in cost savings, with pricing anchored around the roughly RMB3,000 median smartphone price. Vision Pro is the counterexample: “every metric looks shiny, but as a product it still feels a little over-designed.” The battle among Meta, Google and Apple is “completely irrelevant” to him: China’s must-have user base is 10x America’s, and “even if I give up the market everywhere else, winning China would already make us the world’s largest”; Meta’s AI, meanwhile, is “the worst.”
• His end-state view is that in 3-5 years the phone will become an accessory to the glasses—a computing, storage and communications terminal—while most interaction takes place through the glasses; citing 2.5B pairs of glasses versus 1.24B phones globally in 2024, he expects glasses shipments to exceed phones, at least “1:2 to 1:3,” without specifying the direction of the ratio. “People will inevitably own multiple pairs of glasses, while only a minority own multiple phones.” The market is inherently personalized and fragmented: seeing more than 100 people at a launch event wearing the same glasses looked “like a uniform, very strange”; Rokid plans to announce a solution to the high-SKU problem around next year, along with its strategy for the next 5 years.
• The answer to the moat question is blunt: “Hardware has never had a moat.” The host said a generic AI-glasses product can be assembled in Huaqiangbei for just $30-$50, and Misa said that “could be possible.” Once hardware matures, it gets “crossed off as a common denominator”; the remaining differentiation is ecosystem, OS, design and brand—precisely where Rokid sees its edge, as it says it is still the only domestic player with a complete Full Stack OS. Competition has 3 stages: product and technology converge within 2 years; scale warfare over channels, brand and supply chain, where “most startups actually lose”; then ecosystem warfare. Rokid’s orders are about 300K units this year, versus roughly 200K in shipments according to the host, which Misa answered with “even if we did 200K”; conservative inventory planning was the biggest mistake, costing “at least another RMB200M” in revenue, and the target for next year is 1M units.
• The industry is not yet at iPhone 1—it is “a bit like the BlackBerry era”—and 2026 will bring a feature-phone-style free-for-all, with 5 oligarchs competing for the high ground and a “100-glasses war” at the bottom. Misa’s biggest anxiety is not competing with the giants, but that “even if Rokid manages through its own efforts to become BlackBerry or Nokia, that does not mean it will win the war of the future—will Rokid be the iPhone?” Supply-chain maturity is currently only 40-50 points out of 100 and could reach 60-70 points in 2026.

Deep dive

1. The 2025 Breakout: The Inflection Is in Consensus, Not the Product

• Misa’s explanation of the J-curve is direct: once AI became a global consensus, “ordinary people at least knew what they could do if they bought a pair of smart glasses.” From an industry perspective, he does not see a fundamental product inflection; “the bigger inflection is in acceptance, awareness and consensus.” Before 2025, most people questioned glasses; now the proportions have completely reversed.
• The hard breakout number: this year, only 16% of Rokid’s glasses users work in broadly defined tech, versus more than 50% previously. That shift is amplified by the wow moment created by AI entering the reasoning era: “You put them on and think, oh, it can actually be this powerful, this smart and this useful.”

2. Giants Entering Was Expected: “So Far, None Has Made Rokid Nervous”

• His starting assumption from day 1 was that “major companies will want to do it, can do it, and can do it reasonably well,” and he designed Rokid’s path around that reality. The only open question was “how seriously they will take it, and what level they will actually reach.”
• His operating principle is not to worry about factors Rokid cannot control. Whether a large company enters, how committed it is, and how long it persists are outside his control and not worth spending time on; what Rokid can study is what to learn from the rival’s product and brand marketing, and whether its supply chain creates better terms or an “under-the-table deal.”

3. Quark Glasses: An Angel Investor Entered the Field, and Misa “Knew on Day 1”

• 吴泳铭 is a Rokid angel investor, while the Quark team “was under me” when they were at Alibaba. Misa knew about the project roughly 1 year before it launched, and the two sides discussed the rationale: Alibaba is going all in on AI, and “from the cloud to algorithms to models and finally to every entry point, it will definitely try each one.” Quark Glasses is more of an exploration than a single all-in bet.
• The industry is unusually harmonious for now: “Rokid’s sales doubled on the day of Alibaba’s product launch.” When the category gets busy, everyone’s sales rise. “This industry is about staking out territory; a lively industry matters more than anything. The thing we fear most is a stagnant pool.” ByteDance has also been working on it, but “they have more patience because they are not in a hurry.”

4. The Doubao Phone Agent Is “the GUI Era’s Last Stand”

• Misa posted the verdict on his social feed: it is currently the best solution under the GUI terminal framework, but its weakness is obvious—“you are still trapped inside the entire GUI system framework, and any super-app can shut you down at any time. That cannot be a long-term solution.”
• The direction is inevitably NUI (Natural User Interface): “AI is the means; natural is the goal.” Gestures, eye movements, brain-reading and posture are all forms of natural interaction. Rokid is also working in this direction, but its style is not to showcase intermediate lab results: “Wait until I tell you; then everyone can just wait for the product.”

5. The Ecosystem Ledger: Alibaba and Tencent May Benefit Most From Doubao Agent

• The logic is straightforward: Alibaba can connect payments, shopping, navigation and food delivery into a closed-loop lifestyle ecosystem, while Tencent has social networking, payments, e-commerce and games. “Companies that already have a super-ecosystem are actually in a stronger position; Doubao may have inspired them.” ByteDance’s ecosystem is “too narrow,” skewed toward short- and mid-form content and lacking social networking, payments and logistics. “It’s like search engines not being able to enter e-commerce back then; don’t even talk about a more powerful approach like this.”
• The work recently done by Qwen and Amap is something “we are also participating in.” Misa keeps his conclusion measured: Doubao Agent’s “impact is still some distance away, but the product itself is more or less there.” If glasses eventually become a short-video consumption device, ByteDance’s engagement advantage “could lead to a different conclusion”; it is still too early to call.

6. Rokid Is Already Bypassing the GUI: Full-Platform Payments and Agent Store

• Rokid glasses can already hail cars and make payments. “I can quickly turn them into a super-app on the glasses, but there is no rush.” The priority is to get users accustomed to wearing them and build habit around each function: “Don’t start with grand narratives like a big company. I care more about usage time, wear time, satisfaction and word of mouth.”
• The rollout is specific: full-platform payments should go live this month, allowing every application to call the payment interface and collect money; a major year-end update will launch Agent Store, where third-party AI agents can call on glasses-based scenarios. In a live demo, he pressed with 2 fingers, saw the payment code and paid; “I only needed to say ‘confirm.’ It is already very convenient.”

7. The 3 Factions and Jack Ma’s 12-Character Rule

• The 3 factions are smartphone manufacturers, internet giants and startups. Asked whom he wanted to win, Misa said he hoped a startup would emerge victorious. The host said he wanted startups to win because he wanted the world to have new rules; Misa replied, “As long as someone thinks the way you do, Rokid will not lose.”
• The framework comes from Jack Ma: “can’t see it, can’t understand it, don’t think much of it, then it’s too late.” “Can’t see it” has become “can see it”; whether the giants truly understand it beneath the surface remains unclear. “Don’t think much of it” is certain: “With the little volume we have today, to them it is just a few beggars fighting over a broken mat.” A startup’s job is to move fast enough that big companies are too late—“at least too late to knock you off the table.”

8. The Baggage of Smartphone Makers: OS-Level Access and Internal Cannibalization

• The host noted that smartphone makers control system-level permissions and can determine the power consumption and response speed when glasses call on the phone, while the glasses business could cannibalize the phone business. Misa did not directly endorse that explanation, instead returning to Jack Ma’s “can’t see it, can’t understand it, don’t think much of it, then it’s too late.”
• He added that this kind of scheduling means sacrificing phone performance. “Competition in phones is more brutal… if you lean even slightly toward doing this, you have to give something up elsewhere.” Add cross-department coordination, and the process “at least will not move too quickly.”

9. Competition Is Team Versus Team, Measured by Investment Density

• “I rarely expand product competition into competition between 2 companies. More precisely, it is competition between 2 teams.” The question is whether a large company is putting the same intensity, talent mix and investment density into the team as Rokid.
• The numbers: Rokid has deployed 400-500 people, “all top talent in the industry, probably number one so far.” Alibaba, to his knowledge, has one of the largest teams, also numbering several hundred. Rokid has also mobilized ecosystem partners, working with Alipay, Amap, Tencent, ByteDance, DeepSeek and JD.com. “Who has the greater investment density and the stronger network of allies?”

10. The Opening for Edge-Case Disruption: Big Companies Split Every Yuan in Half on Core Battles

• On why Meituan is not making glasses: “We always assume big companies have unlimited resources. The opposite is true: on core battlefields, they split every yuan into 2. The world is never about a big company competing with a small company; it is big companies fighting one another to the death. Very, very few have the spare energy to take care of edge-market disruption, even Alibaba.”
• The conclusion is symmetrical and unsentimental: “Many people think that diverting just a little attention would let them crush a startup. The opposite is true—they either do it seriously or cannot spare a cent. It is that simple.” Misa believes Rokid can break out because the big companies “frankly are not going all out; this is not their core battlefield.”

11. Why Rokid Is Not Worried: Alibaba Would Die if It Did Not Open Up

• Asked whether Alibaba’s own Quark product made Rokid nervous, Misa offered a counterfactual: “If Alibaba does not open up its AI strategy and blocks its future on Quark Glasses, then Alibaba as a whole will have a problem… I mean Alibaba will die. Why should you care about a toe on the body of something that is going to die?” Alibaba will inevitably open up.
• The host suggested Quark could serve as a benchmark or reference design. Misa did not directly agree, saying only: “I think sooner or later, everyone will understand their own ecosystems more clearly.”
• His position on an acquisition has hardened: “I have no desire to be acquired. Rokid—the team, the capital and the influence—already has the ability to fight this battle. Why do it? Unless they completely knock me down.”

12. Li Auto in Glasses: Friend, Catfish and Beneficiary

• He agrees with Li Auto’s in-car extension logic, such as automatically unlocking the car when approaching it. “Most automakers choose to work with Rokid; it is good for Li Auto to act as a catfish and stimulate demand across the market.” He is close friends with the executive 浩宇 and is himself a Li Auto owner; when he previously raised a partnership, Li Auto candidly said the team wanted to try on its own. Lei Jun’s side gave the same answer. Rokid has just announced a partnership with Dongfeng, and “our automaker partners and scale are far larger than Li Auto’s. We are the beneficiary.”
• The timeline: true zero-sum competition will not arrive until industrywide shipments exceed 100M units, “probably 3-5 years from now.” Within 3 years, shipments of full AI glasses with displays should reach roughly 10M units.

13. The 12-Year Bet: Glasses’ Information Density and the Google Glass Starting Point

• “I said on my social feed that I wanted to make glasses in 2013. It has been 12 years. Who would stick with an idea for 12 years?” His goal is simple: “When people talk about AI glasses, they will think of Rokid, give a thumbs-up and say Rokid makes the best AI glasses.”
• 吴泳铭 put the first Google Glass on him while they were at Alibaba. “This thing will definitely change the future, but they made it so badly—you only had to look at it for a while before becoming cross-eyed.” Rokid’s technical lineage is an extension of the Google Glass school: AI plus AR, rather than the VR or HoloLens schools.
• Asked whether glasses sit in the information triangle, Misa confirmed their high information density: “Your input and output are at their densest and most real-time there.” He is also betting that future AI will be fragmented and available anywhere, anytime, when “even taking out your phone is a burden,” and that AI’s understanding of physical space will become more important. Wearing the device on the head is a good way to understand the surrounding environment—“there is no evidence for this, purely intuition.”

14. The Only Major Misjudgment Was Timing

• “Rokid made one huge misjudgment: the timing. We thought consensus would arrive around 2019 or 2020.” When AlphaGo appeared, he was excited that “it had finally arrived,” but AI unexpectedly slipped back into a cooling period until Transformer-based foundation models returned, “4-5 years later than we thought.” The estimate was later pushed to 2022-2023, while Rokid Glasses, as a full-capability AI product, actually appeared in the first half of 2015. “We kept struggling with and overturning our timing assumptions, but the direction never changed from day 1.”
• The hardest moments came each time the timing was missed: “You have to reorganize and remobilize everything, including fundraising and the team, and adjust the plan again. That is extremely hard.”

15. No Pain, Only Hard Work: An INTP’s Optimistic Methodology

• “When you start a company, you assume from the outset that there will definitely be difficult problems. Why suffer over them?” During difficult fundraising periods, his answer was: “It does not matter; I can put my own money in.” The host saw that as evidence of optimism, and Misa agreed that he is “very sunny.” His MBTI result is INTP: “I rarely engage in ineffective socializing, and my social circle is very small.”
• His inversion of “persistence” is worth noting: “If something requires you to persist, you usually cannot persist. Persistence is painful. What people see as persistence may actually be a founder enjoying a huge challenge.” His view of anxiety follows the same logic: if a problem can be solved through action, there is no need to worry—act; if it cannot, there is no need to worry either. “Just execute the actions you can take to the extreme.”

16. Will People Who Do Not Wear Glasses Wear Them? The Host Catches a Change in Position

• In June, he said, “I never thought about selling smart glasses to people who do not wear glasses; don’t disturb them.” His position now is: “If you cannot get someone to put them on today, it is because the product is not good enough. Do you wear sunglasses? Why? To express yourself. If something truly moves him, he will not mind putting on a pair of glasses.” The path is to first convert existing glasses wearers, then, once the difference in competitiveness between wearers and non-wearers becomes large, “they will be forced or choose to put them on.”
• His timing remains aggressive: in about 3 years, “it is hard to say everyone will have one—that would already mean several billion units—but most people will choose to own one. That is how optimistic I am.”

17. A Battle of the Gods Among Meta, Google and Apple? “Completely Irrelevant”

• The core math: “China’s must-have user base is probably 10x America’s. Even if I give up the market everywhere else, if I win China, I am already the world’s largest.” China’s wearing culture and industrial base reinforce the case. “Meta, Apple and Google all make their supply chains in China. Why should you fear them? Chinese people no longer need to look up to overseas companies. We can independently define a category and even lead one.” WeChat, Pinduoduo, TikTok and Alipay are the precedents. When the host said Apple remains the king of smartphones, he countered: “The products that truly affect thousands of households are Android and Huawei, Xiaomi, OPPO and vivo—not Apple.”
• The host cited Google’s lead in multimodality and Meta Reality Labs’ more than 10,000-person full-stack R&D operation. Misa responded: “If that is your reason, then the most innovative force in the world would be banks, because they have the most money.” Media comparisons between Rokid and Ray-Ban Meta have been almost uniformly favorable to Rokid. “For example, we do AI better than it does. Meta’s AI is the worst.” Meta also launched a display product far later than Rokid.

18. The “Say No” Product Philosophy: Single Optical Engine, RMB3,000 and the Vision Pro Counterexample

• Responding to Quark’s dual optical engine, the host said its display could be adjusted larger or smaller. Misa corrected him: “You can adjust the font size, not the focal distance.” He then said the most interesting trait of large companies is that they always believe they can add more ideas, “which is exactly what they should be careful about. They need to learn to subtract.” Vision Pro is the counterexample: “Every metric looks shiny, but as a product it still feels a little over-designed.”
• The single-engine trade-off is straightforward: sacrifice focal-plane adjustment to save several hundred yuan or RMB1,000 and gain 1-2 hours of battery life. “Why not make that trade?” Yield and capacity also improve with a single optical engine. “Rokid is actually the most decisive company of all when it comes to saying no.” Binocular and stereoscopic vision were completed in the lab long ago; leaving them out reflects a deliberate product judgment.
• The pricing anchor is the roughly RMB3,000 median smartphone price. “Above that, it starts to compete on price with the phone as the primary computing platform, and users will pick at every flaw.” Rokid has tested prices in the RMB2,000s, RMB3,000s and RMB4,000s, and “the differences are still very large.” It does not need a high-priced product to show its muscle: “I do not need to raise money, so why do I need to show muscle?”

19. The Cost of Going Viral: Conservative Inventory Cost RMB200M in Revenue

• He had no idea the teleprompter function had gone viral. “It was not until the next day, when my phone was blown up, that I realized it was hot.” What followed was anxiety rather than joy, with customers pressing for deliveries from June through September. The strategic pace “actually did not accelerate”; the only adjustment was pushing suppliers to add capacity after September. “This was our biggest mistake this year: our forecast for orders was too conservative. This is not false modesty; it was a loss. Otherwise Rokid’s revenue this year would have been at least another RMB200M higher.”
• The host summarized it as “Rokid falls down and the whole industry gets fed.” Misa added, “At least we helped them fall.” The host then noted that Rokid did not actually fall, but the industry did benefit. The explosion in AI interest let Rokid sharpen its positioning: AR remains oriented toward enthusiasts and professional markets—“we have a monopoly” in museums and cultural heritage, and maintain the business even without profit—while consumer products focus on AI with displays. “They will eventually come back together as technology, displays, batteries and compute improve.”

20. The 8-Hour Wear-Time Debate and the Impossible Triangle

• Misa clarified the industry’s skepticism over an average daily wear time approaching 8 hours per unit by distinguishing wearing from active use. Simply wearing the glasses like ordinary glasses counts as “using” them, while taking photos or waking AI counts as active operation. The host noted that ordinary glasses typically weigh 20-35 grams versus Rokid’s 49 grams and feel heavy. Misa acknowledged that they “will be tiring and you will feel them,” but most people still accept them, comparing them with 40-50-gram glasses from the era of glass lenses. The benefits outweigh the cost: “WeChat is directly here; I do not need to take out my phone to reply.”
• For all-display products, “under 40 grams would be ideal. It is extremely difficult, and at least today we cannot see how to get there.” Li Auto’s 38-gram target is a threshold. The impossible triangle of weight, functionality and battery life “will definitely be broken in 3-5 years.” Once companies are willing to make trade-offs, battery life and weight will be the first 2 constraints to improve, with the additional buffer then allocated to capability.

21. Edge, Phone and Cloud at 10/20/70: “Terminals Built for AI Will Only Become More Information-Dense”

• The current compute split is roughly 10/20/70 across the edge—the glasses—the phone and the cloud. At maturity, it could become 20 at the edge, 50-60 at the phone and 30-40 in the cloud. The glasses’ form factor means “it is difficult for them to carry every capability independently; there will inevitably be distributed edge-cloud collaboration.” He corrects himself when a judgment proves wrong: “Most of my judgments will be off, so you need a margin of safety.” He agrees with Zuckerberg’s line on the wall: “Done is better than perfect.”
• The distinction from Plaud is that Plaud is a product-defined “best supporting actor,” while glasses “want to be the lead.” He hopes glasses will become AI’s primary carrier, with voice, images and the ecosystem all completed on the device. The end state is full capability: “People only have so many information channels—ears, mouth, eyes, and the nose has not joined yet. You at least have to satisfy them, otherwise your AI is limping.” He calls the ideal form a “spatial computing platform.” The host summarized it as “a cinema mounted on your nose,” but Misa stressed that it is “still a long way off.”

22. Let Society Handle Privacy; He Is Keeping the Pro-Social Move Under Wraps

• He recounted a conversation 2.5 years ago with the person responsible for Meta’s glasses. In the Google Glass era, people wearing glasses into a restroom could get beaten up; Meta did extensive privacy preparation, then found that users stopped discussing the issue. Over the 10 years before the product launched, phones had already changed how people think about and tolerate privacy. “It is people who change, not technology.”
• An engineer’s duty is to make the system transparent—such as using a recording light, as the whole industry does—and give users the choice. “Technology can always guard against the gentleman, never the villain.” He also reversed the intuitive comparison: a phone recording from inside a pocket or filming while held in the hand can be more covert, while glasses may make the action more visible.
• His concrete example of technology doing good involves a blind child whose parents used a display-free Rokid product to read the same books sighted children read. “The entire family burst into tears.” Previously, the options were human narration or scarce Braille books. “We will have a major announcement very soon. It will be a huge surprise for many people with disabilities. I will keep that under wraps for now.”

23. The Phone Becomes an Accessory to the Glasses: 2.5B Pairs Versus 1.24B Units

• Glasses will not replace phones, but they will change the relationship between the 2. Today, the phone is the AI computing center and the glasses are a peripheral. “In 3-5 years, I hope to see the phone become an accessory to the glasses—a computing terminal, storage terminal and communications terminal—while interaction takes place mainly on the glasses.” Kevin Kelly, Elon Musk and Bill Gates have made similar calls.
• The answer to the shipment debate lies in personal behavior: “How many phones do you own? One. How many pairs of glasses do you own? You cannot count them. That is the answer.” Global 2024 shipments were roughly 1.24B phones and 2.5B pairs of glasses, 30% of them sunglasses, with fashion the primary driver in Europe and America. Misa therefore expects glasses shipments to exceed phones, at “at least 1:2 to 1:3,” though the original statement did not specify the direction of the ratio. He acknowledges that this assumes sufficient consumer education; it is still early.

24. The SKU Fragmentation Problem: 100 People Wearing the Same Pair Looks “Like a Strange Uniform”

• At the November 13 launch event, he saw more than 100 users wearing the same glasses. “My feeling was that it was strange, like a uniform.” Glasses are a style statement: professional wear, casual wear and vacations naturally call for different pairs. The market is inherently highly personalized—Meta paid $3.5B for a 3% stake in EssilorLuxottica, whose portfolio includes more than 150 brands. “How could smart glasses ever be made in a uniform volume? Good question. We will probably announce a solution around next year,” along with Rokid’s strategy for the next 5 years.
• The partnership with Bolon is “basically a copy of Meta’s business model,” but Rokid is exploring opening some SKUs to the broader industry rather than, like Meta, pulling everything tightly into its own closed loop. The interview ran roughly 1 hour late because he was negotiating with 2 fashion companies. Historically, SKUs were limited to a few black models because “before consensus formed, we should not scale up. If you mobilize an industry and your judgment is wrong, the loss is enormous.”

25. Hardware Has Never Had a Moat: Once Common Denominators Are Crossed Off, Ecosystem and OS Remain

• The host said a generic AI-glasses product can cost just $30-$50 to assemble in Huaqiangbei and asked whether that meant hardware had no moat. Misa said it “could be possible” and insisted that “hardware has never had a moat. Hardware moats are always temporary. An industry matures, and hardware inevitably loses its moat; otherwise it would remain a playground for small players forever.” Yet Rokid remains the best-selling product in Huaqiangbei. “If low prices were enough to win the competition of the future, wouldn’t this world be too boring?” Borrowing a Steve Jobs-style formulation: “If you want the exact same experience as Rokid, Rokid is the best-priced product you can get.”
• Hardware accumulation is a platform jump rather than a continuous lead, allowing large companies to “dare to follow rather than lead.” Software is a steadily rising climb. “Do a simple math exercise: cross off the common denominators.” Once hardware disappears as a differentiator, design, brand, ecosystem and content remain—just as the gap between iOS and Android today is ecosystem and software experience, not hardware. The supply chain is entering the early stages of commoditization. On Meta’s decision to develop everything in-house from chips onward, he said mature industries generally have specialist suppliers for the base layer. “The only phenomenon that cannot be explained today is NVIDIA,” although it too faces competition from TPU.

26. Rokid Is the Only Domestic Full-Stack OS Player; Scale Is the Biggest Risk in the 3-Stage War

• Rokid built its own OS because it “does not believe an old operating system can be patched and adapted to a new terminal.” Android’s compute, power and scheduling were all optimized for touchscreens, precisely the component AI-era devices need least. “In China, we are still the only company doing a complete Full Stack OS.” If a giant builds a better OS, Misa’s response is: “If they really make it better and move faster than us, isn’t that the proper outcome? It is never big company versus small company; it is good company versus bad company, good team versus bad team.” AI-era XR will have multiple oligarchic operating systems, not just Apple and Android. “Without geopolitics, I suspect there still would not have been HarmonyOS.”
• His diagnosis of the 3 stages of competition: product and technology will converge “over roughly the next 2 years”; then comes scale warfare over channels, brand and supply chain. “Frankly, Rokid is still building that capability. Against big companies, we definitely have not made it there; Rokid is the weaker party. Most startups actually lose here.” The third stage is ecosystem warfare, where Rokid has a neutrality advantage: it supports Qwen, Doubao, DeepSeek and Zhipu and “has no vested interests, so we can be more open.” The host said Rokid already has 30K developers; Misa emphasized that the company continues to fund developer competitions itself. Agent Store already has more than 100 developers. “At least today, we are one of the few companies already fighting in stage 2, the scale war.”

27. Shawshank’s Bird: Leaving Alibaba, Big Swings and the Truth About Fundraising

• Asked whether he regretted leaving Alibaba, he borrowed from The Shawshank Redemption: “Some birds are not meant to be caged.” He did not believe he could go further at Alibaba; it would have been enough if his achievements at Rokid exceeded what he achieved as a professional manager there. “They definitely have.” He readily credits Alibaba for what it taught him: “Before Alibaba, I was a pure engineer with no ability to mobilize and organize large resources and ecosystems. I still have to thank it.”
• He believes the industry is about to enter “an era of big swings,” in which people who know how—and dare—to deploy large resources will win. Rokid has built around an ecosystem from day 1 and continues to fund developer competitions. His candid assessment of fundraising: “We have not raised much money externally. That does not mean our fundraising ability is strong. True fundraising ability means being able to raise money from strangers; Rokid does not have that ability. It was either a former boss, a good friend or even a customer investing in us.” This year’s funding amount will not be disclosed, but it was “more than many people raise before going public.”

28. The 2026 Forecast: The BlackBerry Era, the 100-Glasses War and Do Not Question the Era

• The industry is “probably a bit like the BlackBerry era,” with BlackBerry, Nokia’s folding and full-keyboard phones, and Motorola’s handwriting devices; it is not yet at iPhone 1. Misa’s iPhone moment is when full-capability glasses reach 10M units. The host offered the emergence of productivity tools and social networking as alternative markers, and Misa replied that “once there is social density, the Matthew effect begins.” In 2026, 5 oligarchs—“not 100, probably 5 at most”—will compete for the high ground, while the bottom sees a feature-phone-style “100-glasses war,” with binocular-style devices, recorder-style devices and “many fun things” emerging. Overall, everyone will remain on an upward trend.
• His real anxiety is not the giants: “Even if Rokid manages through its own efforts to become BlackBerry or Nokia, that does not mean it will win the war of the future. Will Rokid be the iPhone, or will it be another company?” That is why new startups still have an opening.
• The confidence behind the 1M-unit target: this year’s orders are about 300K units. The host put shipments at roughly 200K, and Misa replied, “Even if we did 200K,” noting that mass shipments only began in October and therefore covered just 2 months. He remains confident about next year. Supply-chain maturity is “very immature, around 40-50 points,” but could reach 60-70 points in 2026. The broader macro backdrop is favorable: Chinese hardware is benefiting from the national cycle, and smart wearables are a key part of the 15th Five-Year Plan. “When an industry is growing by more than 2x, stop questioning its growth. You can question Rokid and question the technical direction, but never question that this era has arrived.”