--- title: "We Got to the Bottom of “What’s Next for AI” | Unique Research’s CES Las Vegas Event Recap" author: "Unique Research" sourcePublication: "Unique Research Substack" originalPublishedAt: "2026-01-13T13:00:40+00:00" canonical: "https://ffcap.cn/en/research/src-20260113-01html" source: "https://uniqueresearch.substack.com/p/src-20260113-01html" language: "en" --- # We Got to the Bottom of “What’s Next for AI” | Unique Research’s CES Las Vegas Event Recap _Original · Unique Research · 2026-01-13_ _Historical edition: This complete January 13, 2026 text reports the January 8 Las Vegas gathering. All product capabilities, financial expectations, benchmarks, customer results, staffing claims, partner credentials and predictions remain the named speakers’ or original report’s statements, not new tests, current guarantees or independently verified legal conclusions. The “11 founders” introduction refers to the eleven panel guests, whose listed roles also include non-founders; the moderators and eight open-mic speakers are listed separately. The source writes “FMM” while expanding it as Founder Market Fit; that lettering is retained. It calls VLA a visual model and places NPU beside Snapdragon X Elite; these are source labels, not complete technical definitions or an assertion that the whole chip is only an NPU. Rokid’s seven-to-eight-hour usage figure has no time window specified. Currency is unspecified for the valuation in the billions, the 100 million profit example and the 30 million investment; US-dollar amounts explicitly identified in the source remain US dollars. Relative dates, the planned book and the organizer profiles describe the original event period._ A Late-Night Conversation Among 11 Founders Beyond CES The main CES venue never lacks excitement: robots are beginning to "take a seat at the table," wearables are trying to look more like everyday accessories, and smart homes continue to be reinterpreted by AI. More importantly, however, a clearer theme is emerging—AI is no longer confined to screens but is entering the "physical world." From autonomous driving and simulation training to robotics platforms and edge computing, vendors are telling us that "the battlefield for the next generation of AI is not the chat box." Against this backdrop, on January 8, 2026, Unique Research joined FOSHO and Solvea (Shulex) to hold an offline gathering in Las Vegas during CES, the 2026 CES AI MeetUp. From 4:30 p.m. to 10:00 p.m., more than 100 AI founders, corporate executives, and leading innovators gathered for two in-depth panels and one open-mic session. In a smaller venue with denser conversation, they unpacked several issues that would genuinely shape 2026: business models for AI hardware, how brands and efficiency are repriced under the AI wave, and how long-termism can be put into practice. What follows is a detailed recap intended to reconstruct the scene as faithfully as possible. It identifies the participants in each panel and organizes the most valuable takeaways and quotations into reusable frameworks for judgment. 01 An Afternoon and Evening of In-Depth Conversation Time: January 8, 2026, 16:30-22:00 Location: 2210 Red Rock St., Las Vegas, 89146 Organizers: Unique Research, FOSHO, and Solvea (Shulex) Strategic partner: Master Concept Event Schedule 16:30-17:00 Registration, reception, and pre-dinner drinks 17:00-18:00 Opening remarks - FOSHO founder & CEO Charley - Solvea (Shulex) co-founder Steven - Unique Research founder & CEO Delta - Master Concept General Manager Jason 18:00-19:00 Buffet dinner 19:00-19:40 Evening Conversation One: AI Hardware Panel 19:40-20:20 Evening Conversation Two: AI Software Panel 20:20-21:00 Open-mic session 21:00-22:00 Open networking Within the CES narrative, "AI" has moved from a standalone exhibition area to the default backdrop for almost every product launch. Yet for entrepreneurs and companies, the truly difficult question is often not "can it be built?" but three things: 1\. Can it form a healthy business model? Is hardware the door opener? How should subscriptions be priced? Where is the gross-margin floor? 2\. Can it establish long-term barriers? Which of Memory, Agent, brand, and intellectual property are true moats, and which are merely temporary gimmicks? 3\. Can efficiency and brand remain balanced under the AI wave? How should budgets be allocated? How should talent be restructured? These were also the shared themes of the evening's two panels: translating "trends" into "playbooks" and "excitement" into "economics." 02 AI Hardware PMF and the Evolution of Future Trends Moderator: Tony Sun, co-founder of FOSHO Participants, 6 in total: 1 moderator + 5 guests Guests: 1\. Chen Bin, founder and CEO of Lockin Technology 2\. Shi Yi, founder and CEO of FlashLabs 3\. Shen Junxiao, founder and CEO of Memories.ai 4\. Gao Jia, CBO of Mobvoi 5\. Guan Liang, head of overseas ecosystem at Rokid Opening: Five Companies, Five Directions, One Shared Proposition Tony asked each guest to introduce their company and latest product in 30 seconds. The opening round already outlined the diversity of the AI hardware landscape: Chen Bin of Lockin Technology presented an AI smart lock just launched at CES. Its central breakthrough is Aura charge wireless power technology, which can power a door lock from 4 meters away using an emitted light source, enabling the lock to carry high-compute chips and foundation models. "We added two touchscreens and three cameras to the door lock, hoping to use AI to upgrade it into a 'smart household manager.' For example, it could warn parents when a 3-year-old child goes outside or remind an older adult with Alzheimer's disease to take an umbrella and pick up a grandchild." Shi Yi of FlashLabs comes from a completely different field: helping B2B companies find commercial opportunities through Agent automation. "We have an AI-native architecture and will release an end-to-end model this month specifically for the Phone Call scenario, because many sales opportunities require call-center agents to communicate by telephone." Gao Jia of Mobvoi introduced new products in the TicNote ecosystem, including TicNote Pods and TicNote Watch, both launched at this CES. "We have built a comprehensive Agent-driven product-collaboration ecosystem dedicated to AI Native intelligent experiences." Shen Junxiao of Memories.ai has a team drawn from Meta and Google Brain, focused on video understanding and AI's "long-duration memory." "We have also produced a hardware reference design showing how future wearable hardware can use our technology." Guan Liang of Rokid presented glasses weighing only 38 grams. "Although they are Display-less, all the AI functions demonstrate highly robust performance." Core Topic One: The Relationship Between Software and Hardware—Is Hardware a Door Opener or the Core Source of Profit? This was the most contentious and valuable part of the entire discussion. Shen Junxiao first raised an honest uncertainty: "Although we demonstrated the hardware last year, we have never formally released it. I personally still cannot convince myself why users would use a camera every day and pay for software. In my definition, AI hardware requires users to pay for AI input and output, for example ten-odd to twenty US dollars each month. I do not want to become an iPad or Humane AI Pin that merely gathers dust. We will wait until we identify a specific Use Case before releasing it." Shi Yi broke the issue down more clearly: "First, distinguish B2B from B2C. In B2B, the goal is to secure several thousand high-ticket paying customers, each worth more than US$20,000. A few thousand customers can support a valuation in the billions. In B2C, software easily accumulates large numbers of free users, while paid conversion is low. If a hardware company also competes on price, the brand is ruined." His key emphasis was that one must find a way to move from One-off sales to Recurring subscriptions. "If you only sell hardware, the valuation may be a fraction of revenue; under a subscription model, the valuation may be dozens of times revenue." Chen Bin gave a more direct answer based on painful lessons: "I believe AI hardware is the best direction for the future. In China, software payment rates for cameras and door locks are extremely low, at 5%-10%. In the United States, however, the trend of offering basic functions free and charging for value-added functions, such as video editing and cloud notifications, is strong. But experience has taught me that the hardware itself must make money! Never expect software fees to offset losses on hardware. The old smart-speaker and television wars showed that competition at negative hardware gross margins is unsustainable. Hardware gross margins must exceed 30%-40% in China and 50%-60% in the United States for the business to work." Gao Jia offered a different perspective: "I partly agree with Mr. Chen. I do not believe the core question is 'which is more important'; together they constitute a living organism: software is the soul, and hardware is the body. A body without a soul is a walking corpse; a soul without a body is a wandering ghost. Hardware is the gateway, software is the moat, and the ecosystem is the ultimate battlefield. In the future, hardware may even be provided free as an entry point, because the real value lies in users' willingness to pay for continuously evolving Agent services within it. As model capabilities stabilize, AI will be wherever the data is." Guan Liang returned to user experience: "When the iPhone first appeared, iOS was not sold separately either. For smart glasses, strong software support is essential. Rokid's average user time is seven or eight hours, and translation and teleprompter functions are essential needs. Good software creates stickiness and prevents a device from being bought only to 'gather dust.'" Core Topic Two: A Reasonable Revenue Mix Tony followed up: in the long term, what is a reasonable split between hardware and software revenue? The answers revealed an interesting divergence: - Shi Yi: 7:3, meaning 70% hardware and 30% software, is more reasonable. - Gao Jia: The reverse—it should be 3:7, with hardware : software subscriptions : ecosystem services = 3:4:3. Hardware is the carrier; software and services create differentiation. - Chen Bin: In the current door-lock category it may be 95:5 or even 99:1, but he hopes to move toward 7:3 or 6:4. - Shen Junxiao: Software and AI have different cost structures. "AI's marginal cost, namely Token consumption, will not fall to zero as it does for traditional software. People are willing to pay for results, and the revenue potential tied to that Token consumption can expand without limit." The value of this debate is that it reminds everyone building hardware: you cannot discuss only the "product form"; you must design the "profit structure" as part of the product. Core Topic Three: A Guide to Avoiding Pitfalls While Finding PMF Chen Bin offered three specific suggestions: 1\. Founders must personally conduct user research: "Do not make decisions from your desk, and do not rely only on data analyzed by AI. To build the AI lock, we hired Apple's former user-research leader as an adviser, and the founder personally joined in-depth interviews with American users." 2\. Avoid the "faith in technology" trap: "People from technical backgrounds easily 'carry a hammer in search of a nail,' forcing future technology into current products and overengineering their functionality." 3\. Find extreme users to identify needs, and professional users to validate them: "For example, people living in extreme cold or facing very heavy traffic outside their doors are 'extreme users' of door locks." Shi Yi's advice was more concise: "Before PMF works, do not build a large team. Start with a Service: before you have a product, use a service to solve the pain point. If people are willing to pay for your service, then determine how a product can change the cost structure of that service." Gao Jia offered three criteria: 1\. The Builder should personally be a deep user; first solve a problem that hurts one person three times a day. 2\. The standard for judging PMF: users complain about a price increase yet continue to pay. 3\. Build an AI-native workflow: deeply integrate AI into the organization, greatly shorten the paths for market screening and product R&D, and use AI to reconstruct the very process of finding PMF. Shen Junxiao brought a Silicon Valley perspective: "PMF is dynamic. Silicon Valley discusses FMM, or Founder Market Fit. Revenue does not necessarily indicate PMF. You have found PMF only when revenue is Predictable, defensible, and growing at a clear rate." Guan Liang stressed the importance of community: "You must build your own Community. No matter how niche the product, attract the 10 or 100 core users on Reddit or social media. Rokid is the Chinese manufacturer most willing to 'listen to advice'; every week we organize users' P0/P1 requirements. Continuous iteration is itself part of PMF." Core Topic Four: CES Observations—What Is a Fool's Tax, and What Is a Real Opportunity? The guests used specific products rather than speaking in generalities: Chen Bin: - Bullish on exoskeleton products: "They empower people and solve real pain points such as mountain climbing and mobility difficulties among older adults, making them easier to commercialize than humanoid robots." - Bearish on humanoid robots within 5 years: "Dexterous hands and visual models, or VLA, are far from mature. Entrepreneurship should begin with what 'can be built'; do not start by trying to change the world." Shi Yi: - Bullish on the xTool laser printer: highly innovative. - Bearish on the homogeneous "three robotic musketeers": "robot vacuums, lawn mowers, and pool-cleaning robots. Chinese companies everywhere are making exactly the same things. Without disruptive innovation, the contest ultimately comes down to supply chains and channels, where new startups have no advantage." Gao Jia discussed a bird feeder that uses a camera to capture birds' feeding habits and patterns and trigger a response mechanism. She believes the dividing line at today's CES is the "scope through which a product is defined": are you making a device with stronger functions, or designing a system that can continuously sense, learn, and evolve? She believes this reflects a leap to the mindset of a "systems designer," moving from solving isolated problems to designing underlying rules that drive ecosystem evolution. This system of thought is also one of the core propositions of her forthcoming book, Super Organization, scheduled for publication in the first quarter, Q1, of this year. Shen Junxiao is bullish on Qualcomm's Snapdragon X Elite (the source adds “NPU”): "On-device AI capabilities can solve privacy and connectivity problems." Guan Liang is bullish on Shokz: "Although bone-conduction sound quality is average, it precisely addresses extreme scenarios such as sports and swimming. That is value." He proposed a reflective test: "If you returned to Day 1, would you still do this? Consider a robot vacuum that can climb stairs: paying an extra US$1,000 for stair climbing is inferior to placing one on each floor. Returning to basic needs enables the right decisions." Core Topic Five: Evolution Over the Next 3-5 Years Guan Liang: Specialized on-device chips—"Like autonomous driving, the AR/ wearables industry must solve power-consumption problems. There are major opportunities in 'small but beautiful' vertical chips that large companies disdain." Shen Junxiao offered three trends: 1\. From Coding to Agent: from Cursor and Lovable to today's Manus. 2\. The video-understanding wave: visual understanding is at the core of Personal Intelligence. 3\. Computer Use: in the future, it will not be Agent invoking Agent, but AI directly controlling existing Webapp and software. Gao Jia: 1\. AI-native collaboration: AI evolves from a "tool" into an "environment and partner"; future hardware will no longer be a container of functions, but a perceptual interface into an intelligent environment. 2\. Return to human value: we are not people visiting an exhibition at the Louvre, but the people deciding which painting the Louvre will hang next. Shi Yi: 1\. Organizations of full-stack talent: the traditional UI/ backend/testing structure will be replaced by full-stack talent made more efficient by tools such as Cursor. 2\. Do not obsess over Token cost: cost falls 90% every 12 months; the focus now should be how to increase usage through innovative growth methods. Chen Bin: 1\. Rebuild smart homes with AI: visual devices such as doorbells and cat feeders are all worth rebuilding. 2\. Find inner drive: "Building hardware requires persistence for 8-10 years. If you only want to sell the company and cash out in 3 years, that is fine; but if you want to build a lasting business, you must find the inner conviction connecting you to the family and protecting its members." 03 Brand, Efficiency, and Long-Termism Under the AI Wave Moderator: Delta, Wu Wei, founder & CEO of Unique Research Participants, 7 in total: 1 moderator + 6 guests Guests: 1\. Jan Harling, CEO of VIRTUS ASIA and former global media director at Huawei and OPPO 2\. Han Yunyun, COO of EverMind, Shanda Group 3\. Hunter, founder & CEO of Solvea/Shulex 4\. Henry Du, CEO of Huski.ai 5\. Chen Zhiwu, founder & CEO of SellerSprite 6\. Charley, founder & CEO of FOSHO This six-person panel had the most guests of any session. Delta had to restore order several times at the opening: "Friends over there, please stop talking. If you want to talk, go outdoors and do not disturb those who are listening." That conveys how lively the scene was. Opening: Introductions from the Six Guests Jan Harling introduced himself in English: "I am German and have lived in China for nearly ten years; I now live in Thailand. I formerly served as global media director at Huawei and OPPO. For the past three years, I have run my own media consultancy, helping Chinese companies improve efficiency and building advertising technology systems for their US offices." Han Yunyun: "I come from Shanda Group. Beginning two years ago, Shanda Group returned comprehensively to AI, with the entire group 'All in AI.' My principal work now is Shanda's long-term-memory project, whose team brand is EverMind. On November 11, we released the open-source memory system EverMemOS and achieved the industry's SOTA level. Our field is called AI Memory." Hunter: "I am a co-founder of Solvea. Shanda Group is also one of our investors. We specialize in AI-powered customer service. We came to CES because we already serve more than 100 large brands of this kind. We firmly believe that in the future, AI can amplify an organization's capabilities without limit and bring an AI Employee for every role to the entire industry." Henry Du: "We operate in a relatively niche market: using AI to protect and manage brand content. We principally monitor market intelligence and identify counterfeits, infringements, and compliance issues. Our current clients are mainly luxury brands such as Chanel, LV, and Gucci; sports brands such as Adidas and Converse; and consumer-electronics brands such as Apple. Against this background, the only thing that cannot be replicated is Intellectual Property. Therefore, to compete on the world stage, you must think clearly about brand and positioning from day one." Chen Zhiwu: "I am the founder of SellerSprite. Put simply, it is an auxiliary tool for Amazon product selection and operations; fundamentally, it helps sellers discover opportunities and analyze markets. If upgraded with today's AI, what we want to build is a 'decision brain' for cross-border e-commerce sellers. We currently have more than 1.6 million users." Charley: "I am Charley from FOSHO. We focus on Marketing AI and currently have two principal solutions: Affiliate Marketing, which helps sellers build overseas channels, and Media AI, which helps brands achieve overseas growth. We have now served more than 2,000 brands worldwide." Proposition One: Changes in Media Strategy and Budget Allocation Delta asked Jan a key question: how will AI change future brand-budget allocation? "Budget Optimization is a major subject. The reality today is that if you are not doing it internally, your Agency is doing it. Agencies are generally relatively slow; even large agencies provide monthly and quarterly reports, so optimization is not fast enough." He shared a specific case: "We recently ran a campaign for a cosmetics company across 11 markets and improved efficiency by about 32%. This kind of automated budget optimization across countries and channels remains rare in the market." More important is the evolution of KPI: "Many brands, including Huawei, which I have served, still use outdated KPI such as looking only at Impressions. It is easy for agencies to generate impressions, but are they high quality? Is the audience right? Is the brand safe? This requires an evolving perspective." When asked whether agencies would still be needed in the future, Jan gave a pragmatic answer: "It depends on the size of the company. I believe you should have your own experts in the future rather than rely entirely on agencies. But the market is changing too quickly, and you will still need partners such as FOSHO. You cannot maintain an agency-sized team internally, but you must have enough In-house capability to manage agencies, or you will face major challenges." Proposition Two: Memory—the Next Barrier in LLM Competition Delta asked Han Yunyun: why is Memory so important? "If you have followed Sam Altman's recent speeches, he has repeatedly emphasized that Memory is where the barrier in LLM competition lies. By developing long-term memory, Shanda not only wants to become an enterprise AI company but also wants to define the field of Memory clearly." She explained the core value of Memory: "Current LLM systems have Context limitations. Even with RAG, or retrieval-augmented generation, one must confront problems involving recall, latency, and workflow fit. We hope to use Memory as a one-stop solution to infinite context, personalization, and other issues for Agent systems in personal and enterprise settings. When our Memory is good enough, you will no longer need to study RAG or be constrained by context, and you will gain personalization and proactivity. When you switch among different LLM models, your Agent experience will remain continuous." She also shared technical progress: "We have already integrated the EverMem OS memory system into Tanka, an enterprise-collaboration platform invested in by Shanda, allowing the underlying GPT and Gemini models to switch seamlessly. We open-sourced EverMemOS and related Benchmark evaluations on GitHub, demonstrating that our memory system is more accurate than existing memory systems and GPT-4.1 while consuming only 1/10 as many tokens." Regarding its future positioning: "We want to build the industry's best memory system. Our near-term priority is achieving original breakthroughs in core technology and defining scenarios together with developers. We will also organize various competitions and welcome everyone to participate in all kinds of Plugin development. We are also eager for the best people to join us. We hope to become the Oracle of the AI-native era. We will not build at the application layer, but we will empower the entire application ecosystem." Proposition Three: The End State of Agent—Replacing People or Amplifying Them? Delta asked Hunter: to what extent can AI Agent actually solve problems? Hunter answered with a specific case: "Our first customer, Mr. Yang of Anker, set me a clear objective: the company had 200 customer-service agents; could we eliminate all of them? At the time, I thought a foundation model should make that possible, but so far we have helped him eliminate 80%, and there is still distance to cover on the remaining 20%." More important was the change in direction: "I believe we should reverse the question. Are we replacing people, or amplifying them? Our lesson this year is that we want to amplify people's capabilities. For example, one customer-service agent previously served 100 customers; with AI, one outstanding employee can serve 100,000 people simultaneously. We no longer seek to replace everyone, but to replace those who 'do not use AI' and amplify without limit the capabilities of those who 'use AI well.'" In their business model, they made a bold commitment: "We are the only company in the industry willing to put the 'resolution rate' into the contract. For example, promising a 30% resolution rate means saving you 30% of the workforce. Our best-performing customers currently achieve a 95% resolution rate." Proposition Four: Brand and IP—the Final Variable in Overseas Competition Delta asked Henry: how can AIGC content compliance be ensured? Henry redefined Copy: "For a startup in the United States, intellectual property has two sides: continually generate and protect your own IP, and avoid malicious infringement. In the AIGC era, the definition of Copy has changed. Comparison is no longer pixel-level, but at the Context level. Consider short dramas going overseas: before the authorized version is available, pirated copies are everywhere, and the legal boundaries remain unclear." His central advice was: "Have IP awareness, including trademarks and designs, from the first day of building an overseas startup. The cost of breaking the law in the United States is extremely high." To illustrate the importance of brand premium, he shared a vivid case: "I want to share the example of a friend who opened a restaurant in Arizona. Chinese food in the United States is often seen as cheap and unhealthy. But he built a brand concept around 'Healthy Asian Food,' enabling him not only to charge premium prices but also to collaborate with the NBA's Phoenix Suns. With a good story and a good brand, many people are willing to pay. If you depend only on low prices and Copy, when technology and supply-chain costs fall further, you will have no road left except a price war." Proposition Five: The Essence of an AI-Native Enterprise Is Not "Using AI," but "Yielding to AI" Delta asked Chen Zhiwu: will people still use traditional "tools" or "software" in the future? "Let me offer an analogy. Venezuela has the world's largest oil reserves yet fares poorly, while the United Arab Emirates is healthy. Why? Because the UAE understands that oil is a global asset and should be used by the world. The same applies to companies. When a company earns 100 million in profit, the money no longer belongs to the owner personally; it belongs to the social organization." His definition of "AI-native" was especially illuminating: "For our industry, the true barrier is not 'how to use AI,' but 'how to become an AI-native enterprise.' Being AI-native means abandoning the idea of 'oneself as the primary subject' and yielding to AI. If this cognitive issue can be resolved, everything else becomes manageable. In the internet era, information was democratized; in the AI era, capabilities are democratized. But that does not mean outcomes are democratized. Owning a car does not mean you can enjoy life; the true barrier is having 'money and leisure' to enter that way of life. For us, the core remains understanding needs, not the tool itself." At the implementation level: "AI-native means organizational restructuring. Previously, 1 product manager worked with 4 developers; now it may be 2 product managers with 1 developer, or even a product manager leading the entire R&D process and allowing AI to guide collaboration." 2026 Trend Forecasts: Six Non-Consensus Views Jan Harling: The Differentiation of Advertising Creative "On today's social media, attention is declining and prices are rising. If brands rely excessively on AI to produce high-frequency content, every brand will look the same, like today's automobile advertisements. How to use AI to create content that is both efficient and distinctive and memorable is a major challenge." Han Yunyun: Young Founders Emerging in Large Numbers "I foresee many new companies becoming famous overnight, with very young founders, because AI infrastructure is rapidly leveling the capability curve for entrepreneurship. For example, an Agent developed with our Memory can have the best continuous experience. Original AI-native thought and the capacity for action are particularly critical and especially favorable to capable young people with ideas. One of our interns has not yet graduated, but his project already has thousands of stars on GitHub. Mr. Chen, Chen Tianqiao, invested 30 million for him to independently own a project. After AI resources are democratized, more original thinking will grow from young minds. Many of our core team members were born after 1995, and some even after 2000." Hunter: A Wave of M&A and the Rise of Chinese Models "First, there will be more mergers and acquisitions, especially among AI startups. Second, I believe one of OpenAI's hidden 'landmines' may explode and trigger change. I am more optimistic about Chinese foundation models because our Benchmark shows that Chinese models have a higher ROI level. Open source belongs to China." Henry Du: Fundamental Changes in How Humans Learn "I believe the way humans acquire knowledge will change fundamentally. Information noise is increasing while our memory is declining. We need AI to help convert all the raw data we see and hear directly into knowledge and store it. Knowledge workers will not be replaced by AI but empowered by AI." Chen Zhiwu: 2026 Is the True Inaugural Year of "Implementation" "From 2023 to 2025, AI was still at the 'flexing its muscles' stage. The year 2026 is the true inaugural year of 'implementation.' But as a veteran software professional, I must say that there is no 'silver bullet' in software. Productivity cannot easily improve by 10 times, and the penetration of business value is extremely slow. Users do not even know how to analyze data. So do not move too quickly; return to the business itself." Charley: The Explosion of Manus for X and ToB "First, products in the form of Manus, a general-purpose AI Agent, will enter vertical sectors. A marketing plan that previously took 1 month to complete will take only minutes in the future. Second, application-layer ToB products will explode. As for people, I believe humans will always retain the final Decision Maker role. Decisions to spend hundreds of thousands or millions of US dollars cannot be handed entirely to machines. AI improves efficiency; people make decisions." Delta summarized it in one sentence: "One is Cursor for X, and the other is Manus for X." 04 Open-Mic Session: More Voices Join In After the two in-depth panels, the event moved into an open-mic session. Speakers, 8 in total: 1\. Bolbi Liu, CEO of AdsGency AI 2\. Ada Liu, CEO of Share Creators 3\. Wu Jiabing, General Manager of Strategic Investment at Wondershare 4\. Gao Shouzhi, CEO of EntGroup 5\. Ziqi Wang, Founder & CEO of Biuty.ai 6\. Cheng Zhenhua, CEO of Lingxi Future, Lingxi AR 7\. Gao Chitao, Co-Founder of Cloudwise 8\. Chen Peilin, CEO of Vika Moderator: Delta, founder & CEO of Unique Research This session gave more entrepreneurs and practitioners an opportunity to share their observations and reflections, further enriching the dimensions of that evening's discussion. Voices from advertising technology, the creator economy, enterprise software, AR/VR, data analysis, and strategic investment collectively formed a diverse picture of the AI era. 05 Three Conclusions to Take Away As night fell over Las Vegas and this five-and-a-half-hour in-depth conversation neared its end, the evening's most valuable insights could be summarized in three points: 1) The business model for hardware is fundamentally productized design of the "profit structure." This is not a new proposition, but it has been redefined in the AI era. From Chen Bin's "the hardware itself must make money" to Shi Yi's "move from one-time sales to recurring subscriptions," and from Gao Jia's 3:4:3 to Shen Junxiao's "Token consumption will not fall to zero," the consensus behind these debates is that you cannot discuss only product form; the profit structure must be designed as part of the product. The gross-margin floor—30%-40% in China and 50%-60% in the United States—the subscription share, whether 7:3 or 3:7, and how essential the Use Case is: these seemingly "trivial" figures determine whether a hardware company can survive for 8-10 years. 2) Memory, Agent, and Brand/IP Are Competing for the Same Thing: Becoming the User's "Default Gateway" Han Yunyun said Memory is the barrier in LLM competition; Hunter said Agent should amplify rather than replace people; Henry said brand and IP are the only assets that cannot be replicated. On the surface they were discussing different things, but fundamentally they were answering the same question: in an era when AI capabilities converge, what constitutes a true moat? The answer: whoever can become the user's long-term default gateway. Only the form of the gateway differs: some live in the data layer, Memory; some within workflows, Agent; and some within cognition and trust, Brand/IP. 3) The Core of an AI-Native Enterprise Is Not "Using AI," but "Yielding to AI" Chen Zhiwu's statement may have been the evening's most profound insight. In an era of democratized capabilities, tools are no longer the barrier; organizational restructuring is. From the ratio of product managers to developers and founders' willingness to "yield to AI," to the evolution of KPI and the construction of Community, these organizational changes determine whether you are "using AI" or "becoming an AI-native enterprise." And 2026, as Chen Zhiwu said, is the inaugural year in which AI moves from "flexing its muscles" to "real implementation." Do not move too quickly, but think clearly about the direction. 06 Why Are These Conversations Worth Recording? In CES's enormous exhibition halls, you can see countless dazzling product launches and technology demonstrations. But whether these technologies can be implemented and these products can succeed is often determined not by those moments on a launch stage, but by the seemingly "trivial" details in panels like those held this evening: Where is the floor for hardware gross margin, at 30%-60%? How can you tell whether genuine PMF has been found—predictable, defensible, and growing at a clear rate? Which KPI should define good advertising placement—not Impressions? How does Memory overcome Context limits, while consuming only 1/10 as many tokens? How high can Agent's resolution rate go, with a best result of 95%? How can a brand premium be built—with a good story + good positioning? How should an organization be restructured in the AI era—the product-manager ratio? These questions did not appear in any CES keynote, but they may determine a company's survival more than any technical parameter. That is also why we brought together founders, investors, and operators fighting on the front lines during CES: to translate "trends" into "playbooks" and "excitement" into "economics." If 2026 truly is the inaugural year of AI "implementation," success will be determined not by whose model is stronger, but by who can convert these seemingly simple business principles into executable strategies sooner. When the event ended at 10 p.m. and the last guests said goodbye beneath the night sky on Red Rock St., these conversations left behind not only quotations and viewpoints, but also a portable and reusable framework for judgment. That may have been the evening's greatest value. 07 About the Organizers and Partners The offline gathering held during CES, the 2026 CES AI MeetUp, was jointly initiated by Unique Research, FOSHO, and Solvea (Shulex), with support from strategic partner Master Concept. Unique Research is an authoritative third-party institution in AI. Guided by the principles of "open source, neutrality, and reproducibility," it regularly publishes global AI application rankings and in-depth reports through data-driven insight and ecosystem-level resource integration, providing companies and investors with critical decision-making evidence. FOSHO TECH is a marketing-technology company driven by big data and AI. Its two intelligent marketing solutions, "AI affiliate" and "AI media," promote dual growth in global brand traffic and sales. The FOSHO AFF AI affiliate-marketing cloud platform is the world's first intelligent growth platform to use AI to empower brand-partner marketing, providing comprehensive insight into refined marketing data for more than 1 million brands and 10 million channels worldwide. It has now served more than 2,000 brand clients globally. Solvea (Shulex) is a globally leading provider of AI customer-service solutions for e-commerce, helping brands efficiently handle presale, in-sale, and after-sale issues through intelligent AI Agents. Unlike traditional question-and-answer robots, Solvea's core advantage is that it genuinely "solves problems." Drawing on industry knowledge bases and process templates accumulated from extensive real business scenarios, Solvea can be deployed rapidly across multiple channels and deliver significant results within weeks. Master Concept was founded in 2003 and is one of Google's partners with the most comprehensive product portfolio, as well as a top-tier platinum partner. It specializes in Google enterprise solutions, cloud solutions, and enterprise-content-management solutions, and is recognized as one of the most experienced cloud-computing solution providers. --- Original publication: https://uniqueresearch.substack.com/p/src-20260113-01html On-site reading page: https://ffcap.cn/en/research/src-20260113-01html