沈帅波

庄明浩’s central judgment for mid-2026 is that models have cleared the intelligence threshold for most white-collar work, making Agent adoption unstoppable. The competitive center is shifting from pure Chat products such as 豆包, 千问, 元宝 and ChatGPT to OpenClaw, Coding Agent and general-purpose Agent products. Products such as 腾讯云’s WorkBuddy Agent are already reaching beyond technical users to HR teams, finance professionals and small-shop owners. By year-end, he expects China’s task-oriented AI to move from millions of tech enthusiasts and coders to tens of millions or more. “The Chatbot battle is basically no longer worth watching.”

Agents are turning Token demand from linear growth into geometric growth and pushing the raw materials fed to the silicon-based world toward physical limits. A single task may run 100 agents in parallel, while advanced models can work continuously for 10-plus or 20 hours. Compared with the 20-30-second conversations of the past, concurrency and duration alone could produce several orders of magnitude more demand.庄明浩 believes that, on a linear extrapolation, this “may only just be getting started.”

The market’s question for AI has moved from when model companies will turn profitable to what ROI and GDP Token consumption is actually creating. As leading Chinese and US model companies scale geometrically, their path to profitability may already be measurable. The next question is what applications, implementation work and To B SaaS products deliver after consuming vast quantities of Token. 沈帅波 says silicon investment has yet to visibly benefit the carbon-based world. 庄明浩 concedes there is still no perfect answer, but says that each layer of progress lifts “everyone with the tide.”

AI assets remain in an uptrend with expanding volatility, but 庄明浩 rejects equating “unstoppable” with a recommendation to chase prices. Nvidia spent last year rangebound around $180-$200 despite beating expectations for 3 straight quarters. After a Q1 pullback, new factors lifted it to a higher plateau, with the 2026 average price now holding around $220-$240. 沈帅波 says only “buy and never leave” avoids being thrown off the train. 庄明浩 says he is both lazy and fearful: “It’s like surfing—you’ve already been thrown off several times, and if you chase it again, you’ll get killed.”

Hong Kong-listed AI companies face a three-way squeeze from lockup expiries, valuation and global liquidity; positive product milestones may offset the risks but cannot eliminate them. The key dates are the first wave of cornerstone-investor lockups expiring in early July and full circulation in January of the following year. More than 50%, and possibly around 60%, of Hong Kong-listed companies have market caps below HK$500M, while new listings are creating another concentrated wave of unlocks. By one comparison cited on the show, OpenAI may already be a trillion-dollar company with ARR of $40B-$50B or more, implying roughly 20x sales; some Chinese Hong Kong-listed model companies trade at triple-digit ARR multiples—“a 90% drop would only bring them back to double digits.”

The real risk is debt and capital siphoning, not simply high equity valuations. 庄明浩 cites roughly $80B of financing for SpaceX, potentially several tens of billions to $100B absorbed by OpenAI, and debt issuance by Google, Nvidia and Meta. Those giants alone could pull “several hundred billion dollars” from the market. Nvidia’s first bond issue in years was around $20B, while Google also issued equity for the first time in years. His conclusion is blunt: “The stock may still be fine…but the debt is a problem,” because debt carries hard obligations and transmits stress through the system.

The technology agenda for the second half has narrowed to “autonomous evolution” and “world models,” while the hardware agenda remains a push to break through supply and physical constraints. Harness is like putting a chassis, steering wheel and drivetrain on an F1 engine. Loop goes further, laying roads, building gas stations, traffic lights and traffic rules so models can assign tasks, call tools, evaluate results and replan on their own. World models should in turn be split into renderers for human-facing output, planners that follow physical rules and simulators that understand composite world knowledge—three layers corresponding to different paths across video, games, robotics and autonomous driving.

AI may give this generation a new “Engels Pause”: productivity and GDP can rise without a synchronized increase in income, occupational identity or a sense of meaning. 庄明浩 uses the roughly 40-year stagnation in textile workers’ wages during the steam-engine era as an analogy, and argues that professional skills in presentations, analysis and content production are rapidly becoming commodities. Handcrafted essays, PPTs and even code may eventually become niche arts, much like vinyl records. What remains valuable will not be stronger standardized expertise but emotion, taste and courage. A friend’s answer to him was: “AI isn’t as interesting as you are.”

The core competition in video models in 2026 will take place mainly among Chinese players, with Google potentially the only US contender still able to compete head-on. 庄明浩 sees Seedance 2.0, Kling 3.0 (adding “if I remember correctly”) and MiniMax Hailuo as evidence that China is building a cluster of leaders; ByteDance and Kuaishou have no strategic reason to exit, while OpenAI has shut down Sora, Anthropic never entered, and Runway has retreated into a multi-model aggregation platform that “sells shovels.” Training, inference, deployment and usage all burn more cash, making near-term profitability unlikely for the leaders; the real barrier is investment scale and strategic commitment without hesitation.

OpenClaw has ignited not a short-lived product cycle but a systemic shift in AI from L1 chat and L2 reasoning to L3 Agents. An Agent must read context, understand delivery standards, absorb feedback and call databases, permissions, files and even the computer itself; the base model is the engine, the reasoning model is the F1 engine, skills, harnesses and the safety architecture make up the car, and tokens are the fuel. 庄明浩 believes the transition “from being very good at chatting to being very good at getting work done” could run through the end of 2026, while model iteration cycles may compress from 6 months to 3 months, 2 months or even 1 month.

The trade in China’s pure-play model companies is initially a scarcity trade, not a financial one. 庄明浩’s “1%-2% comp” simply applies a percentage of US leaders’ valuations: NVIDIA’s $4.5T implies roughly RMB300B-RMB600B for Chinese GPU companies, while OpenAI’s $840B implies roughly HK$60B to more than HK$100B for model companies; Zhipu and MiniMax have already been pushed to 5%-6% of US leaders, or roughly HK$300B-HK$400B. Kimi’s reported valuation jumped from $6B to $10B on the night Kimi 2.5 launched, and Kimi’s PR materials said revenue in the first 20 days exceeded the full-year figure for the prior year; but once the “left foot stepping on the right foot” mutual bid-up stops working, the 6-month unlock around midyear and the 12-month unlock at year-end or early next year become hard risk points.

The market is indulging “new AI” names such as MiniMax and Zhipu while punishing old money such as Tencent, Alibaba and Microsoft, because AI’s direct revenue contribution at the latter may not even reach 1% of total revenue. The S&P information-technology sector’s PE has fallen from roughly 40x back to 20x, reconverging with non-tech companies; NVIDIA trades at about 17x, while Walmart and Costco trade at about 40x, and Microsoft fell roughly one-third in a quarter even though it is difficult to identify what it did wrong. Alibaba is still burning cash in e-commerce and food delivery, while Tencent can use DeepSeek and OpenClaw to showcase its product and ecosystem capabilities—but that is only a tactical positive, whereas the market wants a more aggressive strategic posture.

Anthropic’s most disruptive advantage is not any single benchmark but “coding plus everything,” which connects general-purpose Agents directly to enterprise budgets. 沈帅波 says Claude-related products may have shipped more than 70 features in 3 months; the boundary between coding Agents such as Claude Code, Codex and Trae and general-purpose Agents has already blurred, while a CIO could cut an existing SaaS budget from 100 to 40 and then keep pushing it toward zero. After a model uncovered security vulnerabilities that had persisted for decades, the cybersecurity ETF fell roughly 7% in a day; the reaction was excessive, but Anthropic’s To B revenue curve is forcing OpenAI—more To C-oriented and dependent on memory and personalization for retention—to rethink its strategy.

China’s model cost curve is shifting from “cheap but nobody pays” to “80-85 points and beginning to monetize for real.” 沈帅波 estimates that Chinese models or services may cost only one-seventh as much as US equivalents, and one-tenth at scale; 庄明浩 says the spending required for Zhipu and MiniMax to reach their current level was also roughly 1% of what leading US model companies spent. Users are starting to buy coding plans, tokens and other compute services, while Seedance 2.0 raised prices 3 times within 1-2 months of launch and still could not meet demand; ecosystem companies such as Tencent and Google can also bundle AI with documents, cloud storage, email, sharing and enterprise accounts, making the payment case broader than a single model feature.

AI hardware is more likely to enhance phones than replace them in the near term, while industry diffusion is running into a two-sided mismatch between physical supply and real demand. Hardware reaching consumers today was often completed in the second half of 2025 and initiated in 2024, leaving the underlying intelligence structurally 1 generation behind; return rates for some AI toys may exceed 50%, Chinese glasses makers may still be shipping at the tens-of-thousands level versus Meta’s million-unit scale, and the category remains far from the “early majority.” On the other side, the show says storage capacity is sold through 2027-2028, more than half of US data-center projects in 2026 have been delayed, and copper transmission is beginning to require a shift to fiber and optical communications—while more than 80% of the world has never spoken a single sentence to AI. One side says the story has just begun; the other has already hit hard limits. That gap is the stress test for the current narrative.

This round of “Chinese asset re-rating” has already delivered many leading Chinese stocks’ full-year return targets of 30%-40% in just over a month, with some up roughly 60% or more. Anyone chasing the rally should first guard against becoming the bagholder. 庄明浩 observes that Chinese stocks often get a window in February and March; once “East rises, West falls” becomes a consensus view across the internet, fund managers may decide, “We can call it a day.” He does not know whether the pullback will last days or weeks, and says he personally has no stock account.

DeepSeek’s breakout was driven by more than open source and low cost: the Lunar New Year mood, product experience and distribution timing combined to turn it into a “national-fortune” event. R1 launched on January 20, U.S. mainstream media picked it up on January 24, and 冯骥 opened a post with “national fortune” just before New Year’s Eve on January 26. Many users were also seeing a visible “thinking process” for the first time while being able to turn on deep thinking and web search together. “For users, this experience was entirely new. It was magic.”

The window for actionable AI entrepreneurship is reopening: the technology is still advancing, but its capability frontier has become more defined than it was over the past 2 years. From late 2022 through the second half of 2024, a team could spend 3 months building a product only for the underlying model to swallow the original plan. Now founders can work through “technology—use case—product—delivery—iteration” again. Robot vacuums with robotic arms, companion toys and AI glasses all suggest that the real moat is often scenario definition, hardware cycles and delivery in messy real-world environments—not simply plugging into a model.

Integrating DeepSeek does not mean the model war is over; the big platforms are using it to “borrow the fake to cultivate the real.” After WeChat added DeepSeek search, some users were routed to Yuanbao, which nevertheless defaults to Hunyuan. Tongyi Qianwen also launched its own reasoning model, while as of the recording Kimi, Doubao and Alibaba’s Tongyi had yet to integrate DeepSeek into their main apps. 庄明浩 likens reasoning models to Level 2 autonomous driving: “If you cannot get here, do not think about what comes next.” Once past that threshold, however, each company will retain proprietary models for its business, data and security needs.

DeepSeek has split the AI capex narrative into competing paths. The guests cited 2025 spending plans of $80B for Microsoft, $75B for Google, $60B for Meta and $100B for Amazon, plus the $100B proposed by Trump—roughly $420B in total versus about $220B the prior year. But Microsoft canceled 2 data centers and Satya Nadella questioned the actual economic growth, prompting the market to trade a “downward spiral.” Palantir’s 10% one-day drop was a sign of high-consensus assets losing support.

For now, AI is more likely to cut costs and raise productivity in existing businesses than to create broad-based new revenue and jobs; getting there may require a restructuring of organizations, business models and the way people connect. 沈帅波 worries that if the gains accrue mainly to chip sellers, data-center builders and a handful of model companies, AI could further shrink middle-class employment and purchasing power. 庄明浩 invokes the electricity revolution and a “compute—connectivity” framework to argue that the productivity dividend may take much longer to arrive.

The primary market still offers episodic profits, but compressed valuations have reduced the margin for error, and exits are harder than sourcing deals. Chinese teams have already built plenty of wrapper products to around $10M ARR; “if there is a shell, fill it” is not shameful. But going from obscurity to a multibillion-dollar valuation in 6 months leaves later-round investors, buyers of old Unitree shares and MiniMax investors exposed to the risk that 5 years of expectations have been priced in at once. RMB funds favor humanoid robots with physical delivery, while dollar funds lean toward applications; whatever the investment, the constraints around IPOs, M&A and buybacks cannot be ignored.