104. A Conversation with Rokid’s 祝铭明: Wu Ma, Alibaba, and Year 11 in Hardware’s Black Forest
Summary
祝铭明’s core call is that AI has already “filled the reservoir,” and AI+AR glasses are on the eve of the transition from BlackBerry to the iPhone generation. He believes the inflection point could come “this year, or perhaps next year”; people willing to wear glasses may broadly adopt AI within three to five years, with the most conservative estimate at five to ten years. Phones may not disappear, but fragmented, lightweight tasks will migrate to glasses first.
Rokid is not betting on a pair of glasses with AI features, but on a new entry point built around always on, display, and an OS. AR Lite users now spend more than two hours a day on average; Rokid says it has more than 10,000 developers, including over 3,000 enterprises, while 60% of product sales come from repeat purchases and referrals. On that basis, 祝铭明 argues that glasses are beginning to show platform characteristics, while rings, watches, and speakers remain constrained by narrow input and output.
Display is the fundamental divide between Rokid and the Ray-Ban Meta route: the former is an information-consumption and AI entry point, while the latter began closer to a social-content creation tool. Without a display, translation has to wait for the model to build sufficient confidence before reading the result aloud, leaving five to ten seconds of dead air in a conversation; text display can offer a rough answer first and correct it as context accumulates. “From day one, they were different species.”
Rokid’s most consequential capital-allocation decision came in 2019: it terminated its AI consumer-electronics line within a week, went all in on AR, and cut roughly 60% of its workforce. The company reserved N+1 severance and ten months of cash for validation; five days after its first glasses招商会 on January 15, 2020, the pandemic erupted, and demand for infrared thermometers took the product to 80 countries. Looking back, 祝铭明 says that if the launch had been five days later, “Rokid would not exist.”
The industry window is short, but the real bottleneck is not orders—it is mass production: Rokid is targeting several hundred thousand units this year and expects major-company competition in complete-display products to begin in the first half of next year. The sudden attention means the first products can no longer be handed to early users at 70 points for co-creation; they must be closer to 90. Rokid is working with Lens Technology to validate processes for which no mature, repeatable solution previously existed, while trying to preserve a 12–18-month lead in product definition. “Make one mistake and your advantage is gone.”
祝铭明 believes hardware leads are quickly erased once the industry reaches a platform phase; sustainable compounding can come only from the OS, experience, and ecosystem. His comparison is roughly three to four hours of battery life for Meta versus four to six for Rokid, with translation feedback delivered within 0.5 seconds. Facing giants, he uses the framework “can’t see it, can’t understand it, don’t think it’s worth pursuing, and run out of time”; startups may have only two or three years to turn “not worth pursuing” into “too late.”
Rokid’s governance advantage is not a founder who is always right, but fast decisions, fast admission of error, and the ability to bear the funding consequences of long-term R&D. 祝铭明 admits that “70% to 80% of my decisions are wrong.” He rejected AR Lite 3 times, yet the team built it behind his back and turned it into one of the company’s most profitable products; he has invested roughly RMB100M–200M cumulatively. His financing principle is to let the team “focus on doing the right thing,” rather than distort the business in pursuit of scale.
Deep dive
1. The iPhone generation pushed 祝铭明 toward a full-stack mobile OS
After the first iPhone launched in 2007, Nokia, BlackBerry, and the Windows Mobile camp were still watching from the sidelines. 祝铭明, by contrast, concluded that smartphones would inevitably move in that direction and that mobile internet needed a software foundation spanning hardware platforms.
His first company tried to port an iPhone-like software architecture across MTK, Android, and Windows Mobile, rather than merely building a customized Android system. Projects that “others can think of and dare to do, but may not be able to stick with” fit his preferred selection criteria.
The team had just over 20 people and operated out of a “bicycle garage” resembling a basement in a Hangzhou residential building. 祝铭明 stressed that when it came to the kernel and operating system, “having more people doesn’t solve the problem,” but at the time there were almost no teams in China building a standard full-stack Mobile OS.
2. Alibaba saved the company in its final 2 weeks, with less than RMB4,000 left
In May 2009, the company’s product had just reached demo stage, with no functioning commercial loop, while less than RMB4,000 remained in the bank and payday was 2 weeks away. 祝铭明 calls it “the first time in my life I felt despair.”
He gave employees 2 choices: stay and take stock surrendered by the founder, though the shares “might be nothing more than a piece of worthless paper”; or take the computers off the desks as payment and walk away with no further obligations on either side.
That weekend he was too afraid to go to the office, and he did not lock the door. When he returned on Monday, “not a single person had left.” Alibaba had already begun making contact, but he refused to promise anything to the team until the money actually arrived.
In less than 2 weeks, Alibaba invested about $1M, and the company never missed a day’s wages. 祝铭明 attributes the outcome of his first startup to “luck,” because the rescue arrived exactly at the point where cash was about to run out.
3. The entire $10M acquisition payment was taken in Alibaba stock
A year later, as the company prepared for another financing round, Alibaba decided to acquire it outright in a transaction worth about $10M. 祝铭明 asked for no cash and all stock: “Even if I took the money, I would invest it anyway. So I might as well invest it back into Alibaba.”
The initial lead came from Jack Ma, after which Joe Tsai（蔡崇信）took over the process and arranged for the team to speak with Wang Jian. 祝铭明 says he had never imagined that an obscure small company like his would suddenly become an Alibaba acquisition target.
Jack Ma offered no view on the technology, leaving only the comment that “Misa and this team are a pretty sunny bunch.” Wang Jian was more focused on the architecture’s forward-looking design. The team later discovered that its architecture had independently evolved to almost the same place as Andy Rubin’s camp: Linux, a virtual machine or runtime environment, and then the Application layer.
4. Alibaba’s Mobile OS got off to a decent start but could not match Android’s open-route velocity
祝铭明 believes they may have been Alibaba’s first true mobile-internet investment; the earlier PHPWind was a community product and did not belong to the same category. Alibaba’s investments in 2009 still looked like early-stage VC: “invest early, invest small.”
After joining Alibaba, he initially ran YunOS. Alibaba chose relatively closed end-to-end control from hardware through software, later trying to work with handset makers and investing in Meizu, but its own influence was insufficient and its expansion lagged open Android.
His summary of that history is: “It was a decent start, but the outcome wasn’t that good.” As far as he understands, the system later shifted mainly into automotive applications, while he moved internally to more frontier work on AI and entry points.
5. M Lab placed an early bet on cameras, microphones, and AIoT
With support from Wu Yongming, 祝铭明 formed M Lab. Its mission was not to build a single App, but to find “new entry points”: cameras, microphones, and AIoT.
He says Alibaba’s earliest QR-code and barcode scanning came from M Lab, with later teams from Alipay and DingTalk taking over parts of the work. Taobao’s image search and Taobao Live also grew out of experiments using the Camera as an entry point.
Around 2012 or 2013, the team began buying GPUs in volume for speech recognition and image search. 祝铭明 acknowledges that Baidu was earlier, calling Alibaba’s team “the second,” while most people at the time did not even know why GPUs could be used for Deep Learning.
These projects reinforced his view of interaction entry points: from terminals, keyboards, and monitors to GUI and touchscreens, every major shift appears to be a UI change, but in substance it is a migration of “the interface through which people interact with technology.”
6. Leaving Alibaba was not about rejecting employment; the uncertainty could not be priced internally
祝铭明 spent 4 years at Alibaba and left in 2014. He was not fixated on entrepreneurship versus employment; the core question was simply “where is the right place to do this?”
His rough estimate at the time was that AI+AR would require at least RMB1B and 5 years of commitment. Looking back, “RMB1B wouldn’t be enough.” If he quietly let the project “grow in the shadows” inside Alibaba, it could not attract first-rate talent; if he openly took the best people and huge resources, it would be unfair to sibling teams.
He told Jack Ma and Wu Yongming directly that staying inside meant either he would be unhappy or the organization would be unhappy—there would be no “everybody happy.” Jack Ma flew back specifically to speak with him for more than 3 hours and ultimately accepted that an independent startup might be better for both sides.
7. Wu Yongming’s way of supporting young people was to let them try before demanding certainty
祝铭明 calls Wu Yongming “Wu Ma” and says his openness toward young people and new things is “far beyond mine.” Google Glass was also something Wu Yongming obtained first and then handed to the team to experiment with.
M Lab started with just an Idea, yet Wu Yongming still supported it “with everything he had.” When the team began buying GPUs at scale, working on scanning, image search, and AI entry points, the business returns were still unclear.
Jack Ma’s only warning about the second startup was: “This time, there’s a high probability you’ll go all in on yourself. You need to think it through.” 祝铭明 replied that even if he lost through hard work all the money he had earned through luck, it would still be worth it.
8. Rokid’s capital base has always been Family and Friends
In the early days, 祝铭明 invested $2.3M personally. Wu Yongming and IDG invested the same amount, while friends including Hu Zemin and Wang Huai of Linear Capital also joined, forming a financing structure of “family and friends, plus former bosses.”
During the difficult period around 2018, Joe Tsai’s Family Office invested again. He also mentioned a Malaysian investor who entered in the same batch. Rokid rarely raised money publicly with a BP, yet continued to announce financings.
This structure was not a deliberate rejection of external capital. 祝铭明 knew that Rokid had made many mistakes and taken many detours over 10 years; unfamiliar investors who did not trust the team would find it difficult to put real money behind such an uncertain hardware route.
Most of the founding team remains, and the company has reached a C+ round and is approaching profitability. 祝铭明 attributes its long survival to “this group’s ability to stick with it and carry the load,” rather than to a product path that never went wrong.
9. AI and AR were never a before-and-after transformation; they were the same founding thesis
祝铭明 wrote in his social feed in 2013 that he wanted to build something like Google Glass. His view when leaving Alibaba was: “AI and AR will change the world, and they will become one thing.”
Rokid established AR Lab in 2015 and formally set up Reality Lab in Silicon Valley in 2016. The team expected AR to mature only around 2019–2020, so it brought AI products to market first while the AR team accumulated expertise in waveguides, microdisplays, and imaging inside the lab.
The supposed “change of track” in 2019 was actually the termination of the AI consumer-electronics line one year ahead of schedule, fully releasing the AR line that had been running in parallel. “We were running on two legs from the beginning.” What changed was the allocation of capital and talent, not the ultimate direction.
10. Alien the robot made the software team pay its first hardware tuition
Rokid’s earliest consumer product was not a speaker but the companion robot Alien. It had a camera, a face, and micro-expressions, and could perform audio, visual, and lighting interactions, but it could not walk. “Even looking at that design today, it’s excellent,” but it was too far ahead of its time.
The team gradually simplified Alien into a smart speaker, partly to build capabilities in supply chain, production, sales, and hardware definition. 祝铭明’s earlier experience was concentrated in software, systems, platforms, and operations; the first 2 years were essentially about “paying every bit of tuition.”
A1 was a refined product priced above RMB4,000. 祝铭明 also used the example of paying RMB5,999 in one shot for a non-essential product to illustrate the purchase pressure faced by an ordinary white-collar worker. Later speakers still mostly cost more than RMB1,000, with the cheapest at about RMB800.
He summarizes that period as 2 extremes existing at once: “Everyone was applauding you, but nobody was paying for you.” It was not a successful product, but it trained the OS, algorithm, supply-chain, and production teams.
11. From day one, speakers lacked the conditions to become platform entry points
祝铭明’s view is that an entry point must have sufficiently broad Input, Output, and use cases. Speakers met none of the 3 conditions and could carry only music, audiobooks, and simple Q&A.
His scenario math is straightforward: a person is awake for roughly 10 hours a day, but may spend less than 2 hours actually in the bedroom or living room. Speakers cannot cover on-the-go, fragmented needs, so they will not become the next-generation entry point.
American users have a long audio tradition, from phonographs and tape recorders to home Parties, making Alexa more likely to become a daily necessity. Chinese users lean more toward headphones and private consumption; playing music aloud is not even a universal habit.
When Alibaba, Xiaomi, and other giants treated speakers as platforms and subsidized the hardware, Rokid did not want to burn money alongside them. “Burning money to fight a war I don’t believe is the future” made less sense than directing the cash on hand toward AR.
12. The 2019 pivot was a capital-allocation decision calculated around the cash runway
Rokid moved from AI speakers fully into AR in roughly 1 week. Investors did not strongly object because the company had explained the destination from day one; the original plan was simply for AI products to “fund the war with the war” and cultivate AR while competition remained healthy.
祝铭明 imposed 2 hard constraints: departing employees had to receive N+1 and could not be mistreated because of the transformation; after the adjustment, the company had to retain 10 months of cash to validate the AR direction.
The strategy had long been clear. The real delay came from not wanting to let go of people. He waited until the cash-planning Deadline to act and later acknowledged that this exposed the company to unnecessary survival risk.
13. Cutting roughly 60% of the workforce was more painful than the direction call itself
Before the adjustment, the company occupied 2 buildings: one for speakers and one for AR. After execution, almost nobody moved in the AR building, while the other building was largely emptied. Roughly 60% of employees left within 2 weeks.
What 祝铭明 found hardest to accept was that these people “had done nothing wrong,” and many were outstanding employees he had personally hired. The employment environment had deteriorated in 2018–2019, and many employees already had families to support, making the pressure far greater than during his first startup.
He spoke one-on-one with every affected employee, explained the reason for the pivot, and promised to reopen related roles first once AR worked. “This time, nobody is to blame.” In fact, many employees later returned to Rokid.
14. January 15, 2020 was Rokid’s narrowest escape from death
Rokid held its first glasses招商会 on January 15, 2020. Five days later, on January 20, the pandemic erupted, movement and offline events rapidly shut down. Looking back, 祝铭明 says that if the launch had been 5 days later, “Rokid would not exist.”
Paired with an infrared sensor, the new glasses could measure temperature in batches without contact. Pandemic demand quickly took the product to 80 countries and gave Rokid its first opportunity for industrial-scale production.
He refuses to say “we are grateful to the pandemic,” but acknowledges that the objective result was an opportunity for the product to mature and enter mass production. Rokid says it became the first company to mass-produce waveguide products in a glasses form factor; the Glass 2 in its showroom was a product of that period.
Financial performance then accelerated: revenue changed by an order of magnitude in 2021, roughly doubled in 2022, and roughly doubled again in 2023. Financing did not stop during the difficult years, which was also related to the growth numbers.
15. To B, B to C, and To C formed a route for progressively lowering product risk
The 2018 glasses prototype was still heavy and suited only to specific To B settings such as industrial and government applications. Rokid did not directly challenge daily consumer wear; it first validated the technology and clarified demand.
Museums and cultural institutions created the B to C stage: institutions defined the scenario and service, while ordinary consumers did the actual wearing. The team used this to observe acceptance, comfort, and real needs, then launched a To C product 2 years later.
祝铭明 still sees himself first as an engineer, not an “excellent product person” who can define the right product through instinct in one shot. He would rather let the market and continuous refinement tell the team “what kind of product is a good product.”
16. Usage time and developer count are beginning to prove AR’s platform characteristics
The core number 祝铭明 cites is that AR Lite users now spend more than 2 hours per day on average. Among electronic products other than phones, he has not seen another product reach that level.
Main use cases include movies, games, and social interaction. Rokid’s community has weekly active usage above 70%, while roughly 60% of sales come from repeat purchases and referrals. He sees these as signs of user stickiness rather than one-off trial.
Rokid says its ecosystem now includes more than 10,000 developers and over 3,000 enterprises, and on that basis calls itself the world’s largest AR ecosystem company—explicitly excluding VR.
This platform status currently belongs mainly to AR products. AI glasses are still more tool-like, with thinner content and ecosystems. But 祝铭明 believes the 2 product categories will eventually converge.
17. The 2 product lines are moving toward the same destination from opposite directions
The AR category emphasizes rich content, ecosystems, fun interactions, and Portable use. It is “worn with you,” but not yet全天Wearable. Its direction is to become lighter, All in One, and eventually eliminate the connecting cable.
The AI-glasses priority is “a good pair of glasses, a good pair of headphones, a good camera, and only then a good assistant.” Even when switched off or out of battery, users should still want to wear them.
The AI line must make content richer while preserving the form of everyday glasses; the AR line must reduce weight while preserving the ecosystem. 祝铭明 is confident the 2 will meet, but admits he still cannot see “when that will happen or what the technical path will be.”
18. VR is like a game console; AR may become a portable computing platform
The underlying reason Rokid “doesn’t believe in VR” is not content quality but the fact that VR cannot be Always On. Having to take a device out of a bag, put it on, and turn it on is worse than a phone for fragmented tasks such as translating a menu.
AI becomes genuinely valuable when it is available anywhere, anytime, for fragmented needs and immediate invocation. Glasses naturally sit on the face, making AR a potential AI carrier, while VR is better suited to dedicated scenarios.
祝铭明’s analogy is: “VR is a PS; AR is more like a phone.” The former can become an excellent gaming and content device; the latter may become a personal tool and portable companion.
19. Glasses first eliminate the System Cost created by GUI paths
Scanning a QR code on a phone requires unlocking it, finding WeChat or Alipay, entering the scanning function, and pointing at the code. With glasses, the user can simply say “scan,” then confirm the payment amount with a voiceprint.
祝铭明 calls the operations required before reaching a function “System Cost.” GUI forces users to follow a path step by step, while AI may bypass the menu entirely and deliver “what you think is what you get, or what you need is what you get.”
His most common small action is lightly tapping the temple to check the time. He no longer takes out his phone or raises his wrist during a conversation; WeChat and DingTalk notifications can also appear there, with unimportant ones ignored directly.
张小珺 countered that phones can also remove App paths through AI tools such as Siri. 祝铭明 conceded that “it’s possible,” but argued that taking a phone out of a pocket and unlocking it still costs another 2 or 3 seconds. A device already on the face is more direct.
20. Phones will not suddenly disappear, but the frequency of taking them out will keep falling
祝铭明 does not claim that phones will inevitably be eliminated. He only says that as AR capabilities improve, “the probability of taking out the phone keeps falling.” Short queries, notifications, and fragmented interactions will move first; complex displays and heavy operations will remain on phones or computers.
He once compared screen time with a friend in Tencent’s investment division: about 8 hours for himself versus about 11 hours for the friend. On closer inspection, most of it consisted of fragmented sessions lasting less than 5 minutes.
His boundary is that tasks lasting more than 30 minutes remain difficult for glasses to replace. In the future, phones may become communication and heavy-task terminals, while glasses handle lightweight tasks, with both developing together.
21. The first phase should not force people who do not want glasses to change their habits
张小珺 said she is nearsighted but dislikes wearing frame glasses, and even feels dizzy when trying them on. 祝铭明’s answer was notably restrained: “Don’t make life difficult for people who don’t want to wear glasses.” There is no need to persuade them at this stage.
Some people feel dizzy even wearing frames without lenses because the frames in their field of vision can continually trigger a sense of lost focus. Long-term wearers’ brains filter out the frames, while people unused to them may remain tense. His advice is simply not to wear them.
He calls those who refuse to come ashore the final cohort of “AR residents” or “XR residents.” When others gain a “generational information advantage” through glasses, they can still choose again, but the decision should remain with the individual.
22. China’s myopia rate provides an early user pool for AI glasses
The figure 祝铭明 cites is that roughly 92% to 93% of Chinese high-school graduates are nearsighted. Even if half choose contact lenses or reject frames, around 50% of the population could still become AI-glasses natives.
For people who already wear glasses, he expects AI capabilities to become widespread within 3 to 5 years—“not any farther away.” The most conservative estimate he has heard is 5 to 10 years.
For people who do not currently wear glasses and do not want to, another 5 to 10 years may be required. The transitional form need not be worn all day; like a tool, it can stay in a pocket and be “worn and used immediately” when needed.
23. Display determines whether AI is a real-time entry point or merely the phone’s auxiliary camera
With no display, translation has to accumulate context and build sufficient confidence before reading the complete spoken result aloud. Each round can involve 5 to 10 seconds of waiting, making the exchange awkward.
With a display, the model can provide an imperfect text result first and correct it as later context arrives. 祝铭明 stresses that the underlying AI capability may be the same, but the output medium makes the experience completely different.
Ray-Ban Meta initially targeted photography, video, and sharing as a social-content production tool; actual content consumption still returns to the phone. Rokid defined itself as an AI entry point from day one and therefore treats display as mandatory.
24. The US Sunglasses logic is not the same as China’s Always-on logic
祝铭明 believes the premise for most Americans is closer to Sunglasses: they wear them while driving or outdoors and take them off indoors, so the product does not need to exist all day.
That logic can support fashion, capture, and social use, but it is inherently not Always On. Meta’s AI was also not Ready at the beginning of the product-definition process, so omitting display was not a mistake—it served different scenarios.
He has heard that Meta may begin shifting toward display and AI from year-end, but still expects no-display social glasses to retain substantial room. “For its scenario, it is definitely right.” That does not mean only one of the 2 routes can survive.
25. AI’s upstream reservoir is full; downstream hardware is taking over
祝铭明 compares the current state of AI to a reservoir that has already been filled. The next step is not to keep proving that the water exists, but to convert the energy into “irrigating rice fields and generating electricity.”
DeepSeek made previously unattainable capabilities inexpensive and Open Source, while helping large companies lower operating costs. AI applications and hardware therefore have a more realistic economic foundation.
He divides AI hardware into “embodied intelligence” and “portable intelligence.” Robots and future cars belong to the first category; phones and glasses belong to the second. Rings and watches can be good products, but their display capabilities are too narrow for them to become platform entry points.
26. Smart glasses are at the inflection point from BlackBerry to the iPhone generation
祝铭明 believes the most accurate analogy today is “the point of transition from BlackBerry to the iPhone generation,” with the industry potentially reaching its true iPhone-generation moment “this year, or perhaps next year.”
Rokid’s sudden attention is not merely the result of company promotion. He places it against the backdrop of Chinese technology confidence rebuilt by DeepSeek and the market’s search for downstream vehicles for AI.
Glasses fit the combined imagination of “portable intelligence,” display, and AI applications, so traffic has gathered quickly. But he still calls the industry a black forest and does not claim that the current route is the only answer.
27. Major-company competition in complete-display glasses is expected to arrive one beat late
祝铭明 believes most major companies were still following the no-display Ray-Ban Meta route, which sits at a different product level from Rokid. The true entry point for complete-display products may come in the first half of next year.
He set Rokid’s sales target for this year at “several hundred thousand units.” The biggest difficulty is not demand, but whether production and industry capacity can meet it.
As for Xiaomi, he believes it has a clearly defined user base and does not need to confront everyone head-on. Glasses are less convergent than phones and more like fashion-inflected consumer products, capable of accommodating different designs and audiences.
His guess for the market structure is roughly 10 players in a “multi-oligarch competition,” but he emphasized that this was just a back-of-the-envelope guess without data. Rokid’s immediate goal is to earn the qualification to “sit at the same table as the adults.”
28. Viral attention raised the product launch threshold from 70 points to 90
When the company was little known, Rokid could use F-codes and Early Adopters to give a 70-point product to a small circle for co-creation and refine it toward 90. The first mass-produced products under the spotlight receive no such tolerance.
祝铭明 says peak traffic reached roughly “300 million people, not impressions.” Even if only 1 person in 10,000 complains, amplification could produce tens of thousands of public negative comments, while satisfied users tend to remain silent.
The company therefore pushed the launch into the second quarter to continue improving hardware, stability, and onboarding. 祝铭明 wanted another 2 or 3 months so that an ordinary user could put on the glasses and start using them without an engineer standing by.
29. The mass-production challenge is that there was no mature process to copy
Complete products with displays had never previously been mass-produced at scale. Production lines, processes, and yields all had to be validated first. Once validated, they might no longer be difficult, but the first company still has to bear the exploration cost.
Rokid worked with Lens Technology on mass production because a startup could not independently absorb the process-validation burden. 祝铭明 admits that a company of Rokid’s size should not have had to do it, but the giants had not yet mass-produced along the same route.
He believes Apple and Meta had previously invested more heavily in the Vision Pro direction before shifting back toward lightweight AI+AR glasses. Rokid unexpectedly ran ahead, gaining first-mover position while also being forced to shoulder more risk, responsibility, and cost.
30. Overly conservative demand forecasts turned underfunding into a capacity bottleneck
In the middle of last year, Rokid made a weak assessment of the environment and shifted its supply chain to “produce to order”: check orders first, then release the corresponding capacity. That looked rational at year-end, but once traffic exploded, the company could not accept orders quickly enough.
祝铭明 admits that with a fuller balance sheet, the company could have bought out capacity in advance and held 2 to 3 months of inventory. The gradual ramp controlled cash, but sacrificed elasticity when demand accelerated.
This also corrected the external impression that “Rokid has raised a lot of money.” For a project simultaneously building hardware, an OS, AI, and an ecosystem, he does not consider the funding abundant. The capacity shortage is evidence.
31. The real Know-how is concentrated in the OS, not the hardware specifications
祝铭明’s battery comparison is roughly 3 to 4 hours for Meta versus 4 to 6 hours for Rokid. The extra one-third comes mainly from continuous power optimization at the OS layer.
Real-time translation requires a result within 0.5 seconds. Stable Bluetooth, smooth data flow, and reliable connections are also system capabilities; if Bluetooth itself is unstable, later AI experience and power optimization are impossible.
Hardware progress is not linear. Once an industry reaches its boundary, it enters a platform phase, and more effort cannot produce a sudden leap, while competitors can catch up through mature supply chains. Cars and glasses both allow giants to “dare to come from behind.”
Software, experience, and ecosystems can compound continuously or even exponentially, creating Matthew effects. Rokid’s defense is therefore not permanent ownership of exclusive components, but an OS and user experience that latecomers cannot quickly replicate.
32. A 12–18-month lead balances funding capacity and industry cost
Rokid wants to maintain a 12–18-month lead in product definition. A longer lead would require massive investment starting with materials, basic research, and upstream capacity, which does not fit its funding scale.
Extending the lead to 18–24 months or longer would impose supply-chain capex similar to Apple’s or Meta’s. A shorter lead would quickly be erased by mature processes.
“Twelve to 18 months—make one mistake and your advantage is gone.” 祝铭明 does not describe this window as a completed moat, but as a race that must be won again with every product generation.
33. The giants’ “four don’ts” may leave startups only 2 or 3 years
Jack Ma once summarized the opportunity for startups as “can’t see it, can’t understand it, don’t think it’s worth pursuing, and run out of time.” Giants first overlook a marginal opportunity, then fail to understand it, dismiss its small scale, and finally find themselves unable to catch up once the market breaks out.
AR glasses failed the first 2 conditions from day one—giants could certainly see them and certainly understand them. The startup opportunity may be limited to the 2 or 3 years in which giants temporarily “don’t think it’s worth pursuing.”
Rokid’s task is to build an equal competitive position inside that window, not fantasize about investing on equal terms with giants such as Tencent. “While everyone thinks it isn’t worth pursuing, can we make it so good that they run out of time?” is the shared objective across financing, product, and ecosystem strategy.
34. Building the ecosystem alone is not Rokid’s optimal answer
祝铭明 wants to find more “allies.” Tencent has social, content, and scenario capabilities and may not need to build hardware itself. If they build the ecosystem together, Rokid can focus on the platform, technology, and product experience.
At the AI-model layer, users can choose Doubao, Yuanbao, Tongyi Qianwen, or DeepSeek. Even if ByteDance and Alibaba are experimenting with hardware internally, Rokid still wants to remain Open.
He believes that if a large company truly commits, accumulating 10,000 developers might take less than 6 months. Rokid’s developer base, built over many years, has value, but it cannot carry all content production alone.
The company’s strategy therefore has “2 legs”: build up provisions and fortify the walls to withstand exploration and preserve the lead, while making more friends and amplifying the platform through Ecosystem Building rather than closed control.
35. The purpose of financing is to keep the team from taking detours for cash
祝铭明 defines financing as ensuring that the company can “focus on doing the right thing,” without compromising for survival or launching distorted businesses in pursuit of short-term scale.
He initially invested $2.3M and later invested roughly RMB100M–200M cumulatively. So investment from others is welcome, but even without it he would continue investing. He says at least he is “raising money while sitting down, not raising money on his knees.”
Rokid has roughly 400 employees and the team still firmly controls the company. The next financing round has “no schedule and no ceiling,” but he also stresses that “financing is responsibility” and that the company should not take more money than it genuinely needs.
Against the giants, the company must at least withstand the first 2 years of “carpet bombing and scorched-earth containment.” A professional manager who fails to win for 2 years may change strategy; a startup has no fallback. Cash therefore first buys time to stay on the right route.
36. A “playful spirit” keeps the team patient in hardware’s black forest
Rokid comes from Robot Kid: the products and industry are young, and the company should remind itself that it is still like a child, while users should retain the youthful willingness to try new things.
The first value, “playful spirit,” is not carefree amusement. It means finding the pleasure of continued exploration amid the unknown, failure, and difficulty. 祝铭明 says it is closer to an endorphin-like sense of achievement than to immediate dopamine.
From product definition and supply chain to PMF, capacity, and competition with giants, every step produces a more intense problem. “Only by constantly solving these difficulties and finding the fun inside them can you keep growing.”
37. The CEO’s value is not making fewer mistakes, but taking responsibility for their consequences
祝铭明 estimates that “70% to 80% of my decisions are wrong.” He does not see himself as a genius product person who can foresee the future; his strength is making a quick call when others are afraid to decide.
He explains that a founder is not necessarily smarter, but is the last person in the company able to bear the consequences. If he hesitates, the team may lose its window to validate the idea regardless of who is right.
He also describes himself as “very forceful, but extremely quick to admit mistakes.” Once the product experience or numbers are in front of him, he “won’t protect my own ego for even one second.”
38. AR Lite was rejected 3 times; the team overruled the boss with a finished product
Both of Rokid’s most profitable products were once rejected by 祝铭明. AR Lite was the clearest example: after he killed it 3 times and the product and business teams failed to persuade him, they built it behind his back.
Once the finished product was “shoved in my face,” he used it and admitted that it was good. The 3A gaming experience especially moved him because he could carry it to a hotel and play Black Myth: Wukong.
The episode proved that Rokid does not define products through a one-way founder process. 祝铭明 believes a product company accumulates results through “tens of thousands of decisions,” and that in many details he is actually a Trouble Maker.
39. His biggest regret is not choosing AR, but executing 6 to 12 months late
Looking back at 2018–2019, 祝铭明 believes the switch should have happened 6 months to 1 year earlier. Not wanting to let people go did not change the final result; it merely delayed the pain and pushed the company closer to the cash cliff.
If the organization had been reorganized earlier, the company might have retained more outstanding talent and avoided a one-shot cut within 2 weeks. The most dangerous risk was that a launch 1 week later could have allowed pandemic lockdowns to cause a “fatal blow.”
AR Lite was also delayed by several months because of his judgment. He does not believe the company ever made a fundamental strategic mistake, but admits that many small mistakes happen every day. Continuous reflection matters more than defending personal correctness.
40. The hardest call is not the direction, but when the paths converge and which alternative route wins
祝铭明 is firmly convinced that AI+AR will eventually merge. The real uncertainty is when AR Lite and Rokid Glasses will become one product and which technical path will get them there.
He is not most worried about giants copying Rokid, because large companies’ routes and resources are relatively visible. More dangerous is a handful of people in a garage taking “a completely different road from yours” and overturning the existing definition.
He is in no hurry to declare lighter solutions such as Halliday and Even Realities right or wrong. Some retain the Ray-Ban trade-off; others take subtraction to the extreme. These differences are more worth studying than simple copies of Rokid.
His attitude is to encourage startups to keep exploring. If someone else has a better idea, “perhaps one day I’ll become the person applauding everyone.” Investment, acquisition, or letting the other side complete its mission are all acceptable outcomes.
41. Whether to sell depends only on which structure can make the product real
Asked whether Rokid might be sold like his previous company, 祝铭明 replied, “Why not?” Starting a company, being the boss, or joining a larger company are merely organizational forms. “Doing the work matters more.”
He also explains that most sales are not good choices. After entering a large company, a team faces organizational pressure, limits on resource mobilization, and the burden of existing businesses, and may be forced away from the industrial foundation.
If a large company could provide sufficient resources while preserving the direction of the business and creating synergies, he would not reject a transaction. Refusing to sell is not the goal; preserving the product mission is the constraint.
42. Success should be defined internally; the brand should eventually eclipse the founder’s IP
For his second startup, success is not making money, going public, or receiving external praise. It is continuing down the path he chose and having users genuinely like and wear the product over the long term. Making money while being forced off course would still count as failure.
祝铭明’s ideal state is: “People really like our products, but nobody knows this person Misa.” Rokid is currently merely “seen by the eyes and heard by the ears,” and has not yet entered users’ hearts the way Huami, OV, and other brands have.
He does not want to reproduce a Steve Jobs-, Jack Ma-, or Elon Musk-style founder IP, because public figures must sacrifice personality and freedom. Jack Ma is strong enough inside; 祝铭明 says he is an I-personality and “doesn’t know what to say if I’m not talking about the product.”
In reality, he has already become well known, so he has lowered the bar: the company brand and product attention must be far greater than his personal attention. Perhaps users will eventually stop thinking about the founder’s face.
43. 梁文锋 and Hangzhou provide examples of technological confidence and long-termism
祝铭明 describes 梁文锋 as extremely introverted and soft-spoken, but an idealist. Under the pressure of DeepSeek’s traffic, 梁文锋 even said that users seeking a smoother experience could use Doubao, because the satisfaction comes from seeing technology used rather than locking traffic inside one’s own platform.
He sees DeepSeek and 梁文锋 as symbols of rebuilt technological confidence, and Feng Ji and Black Myth: Wukong as symbols of rebuilt cultural confidence. Both phenomena happened to emerge in Hangzhou, bringing concentrated attention to the city.
His personal comparison of the cities is that Hangzhou is more pragmatic than Beijing and more patient than Shenzhen. Rokid and Game Science both took roughly 10 years; a rhythm of “invest today, with no guarantee of output next year” is easier for Hangzhou to accommodate.
44. The final bet remains that AR will change the phone, not that the founder will exit
祝铭明’s ultimate product goal is simple: first get every person who already wears glasses to switch to Rokid and wear it every day. The ecosystem and grand narrative must come after that fact.
If Rokid succeeds and he starts another company, he wants to build a deep-sea submersible and “use new technology to take people to places they have never been.” Many people are already exploring space; the deep sea still preserves the unknown.
But his biggest Bet for the next 3 to 5 years has not changed: “I’m betting that AI will replace the phone—AI glasses, yes, AR will replace the phone.” More precisely, glasses will first change how often people use phones, then compete for the next personal-computing entry point.