Setting aside sentiment and returning to industrial fundamentals. There are only three core criteria for judging whether an industry has entered its later stages: solidified technology roadmaps, closed-loop business models, and a locked-in leading player landscape.
In recent months, KNOWLEDGE ATLAS and MiniMax saw their stock prices skyrocket immediately upon listing, leading the market to immediately celebrate and declare: China's large model race has been decided, and the industry has directly stepped into its later stages. This statement simply does not hold water.
Unlisted players like DeepSeek and Kimi possess technical prowess that is completely on par with the first two; in some specific technical metrics, they are even more competitive. On another front, giants like Baidu, Alibaba, Tencent, and ByteDance hold hundreds of millions of C-end user traffic, and their AI assistant and agent products are thriving. However, when it comes to Web Coding or tackling underlying technological challenges, they immediately show weakness, forced to rely on defending their core traffic base.
The industry's human resources landscape appears even more turbulent: Xiaomi successfully recruited Luo Fuli, strongly shoring up the core短板 (weakness) in its self-developed large model capabilities. Rumors of the CEO of Qianli Technology and Tencent's AI head Yao Shunyu leaving their posts spread like wildfire across the internet, only for both parties to swiftly issue denials, stating their teams remain stable as usual. The cycle of rumors and denials playing out precisely indicates this track has never been calm.
With capital clustering, talent undercurrents, technological iteration, and a shifting competitive landscape, the so-called "industry later stage" is purely a false proposition. China's large model sector is still in the most chaotic, intense, and variable early-stage breakout battle. What seems like a stable landscape today could be completely rewritten tomorrow.
01 Listing ≠ Deification: Newcomers Lead the Pack with Dual Buffs of Tech + Capital Are the soaring stock prices of KNOWLEDGE ATLAS AI and MiniMax a sign of achieving peak strength or simply a capital bubble? The answer is straightforward: the first-mover listing红利 (dividend) far outweighs the valuation support from the technology itself.
In the 2026 A-share market, pure large model investment targets are extremely scarce. Most AI companies are either involved in computing hardware, algorithm outsourcing, or industry application integration. Players who genuinely started from the underlying large model and focus on native technology are few and far between. KNOWLEDGE ATLAS and MiniMax precisely caught this稀缺性 (scarcity)风口 (trend), becoming the unique pure large model targets in the capital market, leading to a flood of capital and顺势 (following the trend) surging stock prices.
However, this does not mean the two rely solely on capital luck. The core reason for attracting heavy capital investment is stepping onto the right key track—coding capability. While the vast majority of models in the market are still focused on conversational interaction, copywriting generation, and general Q&A, KNOWLEDGE ATLAS and MiniMax had long ago heavily invested in Coding, complex reasoning, and API commercialization, accurately targeting the core needs of developers, enterprise services, and software development, and were the first to establish a closed loop of "technology output - commercial monetization."
A technology foundation,叠加 (coupled with) capital助推 (boost), created their强势 (strong) performance in the secondary market. To conclude from this that listing is the industry's endgame is far too naive.
Unlisted players DeepSeek and Kimi are underestimated隐形巨头 (hidden giants). Without the光环 (halo) of the secondary market or the溢价 (premium)加持 (support) of capital, these two have been relentlessly focusing on underlying technologies. Their code generation efficiency, reasoning accuracy, long-text processing, and low-cost optimization capabilities firmly place them in the industry's first tier. Compared to the listed KNOWLEDGE ATLAS and MiniMax, they show no gap; in fact, their technological purity holds an advantage. Their recent consecutive signals of large financing and refinancing rounds are not due to lack of funds but a deliberate strategy to avoid the束缚 (constraints) of listing,全力蓄力 (fully preparing) for a long-term war.
For pure-technology AI companies, listing too early equates to being held hostage by financial reports, stock prices, and short-term performance, forcing them to sacrifice long-term R&D to cater to the short-sighted expectations of the capital market. DeepSeek and Kimi staying in the primary market is a refusal of capital interference, channeling all resources into underlying technology, open-source ecosystems, and global布局 (layout), trading time for absolute technological barriers.
These four represent the true first tier of China's large models. Unburdened by traditional business baggage or organizational内耗 (infighting)拖累 (drag), they focus entirely on native innovation. They are not流量附属品 (traffic appendages), not业务边角料 (business scraps), and do not rely on母公司输血续命 (parent company输血 (blood transfusions) to survive). They compete globally凭硬核技术 (relying on hardcore technology). For China's large models to gain a firm foothold globally, it will depend not on the traffic of giants but on the technological底气 (confidence) of these four newcomers.
02 Giants Only Guard Their Traffic Base: Big but Not Strong, Destined to Be AI Supporting Actors The situation for internet giants' large models is extremely尴尬 (awkward). Baidu's ERNIE Bot, Alibaba's Tongyi Qianwen, Tencent's Hunyuan, ByteDance's Doubao—each boasts满 (full)知名度 (awareness) and顶 (top) user scale. Backed by super ecosystems in search, e-commerce, social, and content, giants naturally possess流量 (traffic),场景 (scenarios),资金 (funds), and生态 (ecosystem) advantages that startups cannot match. Their C-end product experience, AI Agents落地 (implementation) capabilities, and user触达 (reach) efficiency堪称 (can be called) industry顶级 (top-tier).
So why do they始终 (consistently) fail to surpass startup newcomers in专业市场 (professional markets) and技术圈层 (technical circles)? The reason is残酷又直白 (cruel and straightforward): for giants, large models have never been a core主业 (main business), merely增值配件 (value-added accessories) for their traffic ecosystems.
Baidu's core盘 (base) remains search and cloud services; Alibaba's重心 (focus) is always e-commerce, local services, and cloud computing; Tencent's根基 (foundation)离不开 (cannot be separated from) social, gaming, and investment; all of ByteDance's businesses ultimately serve content traffic and commercialization. Within giant corporate structures, large model departments are always创新业务 (innovation units),辅助业务 (supporting units),赋能业务 (empowerment units),排不进 (unable to rank into) the group's first strategic priority.
This注定 (dooms) the R&D logic of giants' large models to be off-track from the start. Startup newcomers can All in底层技术 (underlying tech),允许试错 (allow for trial and error),允许长期无盈利 (allow for long-term lack of profitability),允许攻坚无人区 (allow for tackling uncharted territory). Giants' large models must快速落地 (land quickly),复用现有场景 (reuse existing scenarios),反哺主业营收 (feed back into main business revenue). They must pursue technological innovation while背负严苛KPI (bearing strict KPIs), and还要周旋 (also navigate)跨部门争夺 (cross-departmental争夺 (competition)),冗长决策 (lengthy decision-making), and战略妥协 (strategic compromises).
The final result is: giants are adept at turning AI into useful C-end products but cannot produce industry-defining underlying technologies. In areas普通用户看不见 (ordinary users don't see)—Web Coding, complex logical reasoning, underlying code optimization, open-source technology output, global API commercialization—giants' models show no亮眼突破 (remarkable breakthroughs). They can only rely on存量用户 (existing users),生态捆绑 (ecosystem bundling), and场景复用 (scenario reuse) to占据 (occupy)大众市场份额 (mass market share).
They may seem体量庞大 (huge in scale),声势浩大 (grand in声势 (momentum)), but in reality, they only停留在应用层 (remain at the application layer),触碰不到 (cannot touch) core technical barriers. Put simply, giants are important players in China's AI but are unlikely to become technology definers. They can普及 (popularize) AI to hundreds of millions of users, can靠生态赚走存量利润 (use ecosystems to earn存量 (existing) profits), but cannot主导 (lead) underlying technology directions, let alone抢占全球技术话语权 (seize global technological discourse power).
In the future industry landscape, giants will be AI生态运营商 (ecosystem operators), while the four startup newcomers will be the破局者 (game-changers)掌握核心技术 (mastering core technology).
03 Rumors Reveal More Than Truth: The Frenzied Talent Grab is the Real Barometer Recent industry personnel震荡 (turbulence) is格外值得玩味 (particularly noteworthy). Xiaomi's successful poaching of Luo Fuli is an undisputedly重磅信号 (major signal). This core AI expert's joining is not a simple executive入职 (onboarding); it signifies Xiaomi彻底跳出 (completely跳出 (breaking out of)) its "终端应用AI (terminal application AI)"浅层布局 (shallow布局 (layout)),正式杀入 (formally entering) the underlying self-developed large model赛道 (track). With手机 (phones), AIoT,汽车全终端 (cars全终端 (all terminals)) +自研大模型 (self-developed large models), Xiaomi's AI全局版图 (overall blueprint) has now补齐最后一块短板 (filled its final短板 (weakness)).
Even more耐人寻味 (intriguing) is the循环戏码 (recurring cycle) of "rumors - denials." News of the CEO of Qianli Technology and Tencent's AI head Yao Shunyu leaving引爆业内 (exploded within the industry), with rumors of job-hopping,战略分歧 (strategic分歧 (differences)),股权矛盾 (equity矛盾 (conflicts)), etc.,甚嚣尘上 (running rampant). Yet很快 (soon after), the relevant parties接连公开辟谣 (successively issued public denials),核心管理层在岗履职 (core management remains in their posts),团队业务稳步推进 (team operations稳步 (steadily)推进 (advancing)). A potential震荡风波 (turbulent storm)最终落地为虚惊一场 (ultimately turned out to be a false alarm).
But this is绝不是 (absolutely not)无关紧要的行业八卦 (irrelevant industry gossip). The proliferation of谣言 (rumors) itself is proof of the industry's极度躁动 (extreme躁动 (restlessness)). In a成熟稳定 (mature and stable) industry,核心高管变动 (changes in core executives) would never trigger a闹剧 (farce) of "疯传-辟谣 (wild spreading - denial)." Only in a高速爆发 (rapidly爆发 (exploding)),格局未定 (landscape-unsettled)赛道 (track) are顶尖人才 (top talent) the核心资源 (core resource)各方疯抢 (everyone is scrambling for), where every一点风吹草动 (slight stir) is无限放大 (infinitely放大 (amplified)).
Top technical talent and core managers in the large model industry remain必争之物 (must-have assets) for all tech players. Startup newcomers抢人才 (grab talent) to筑壁垒 (build barriers);跨界厂商 (cross-border manufacturers)抢人才入AI (grab talent to enter AI); internet giants抢人才守阵地 (grab talent to defend their positions). With三方激烈争抢 (three parties fiercely competing),人才溢价水涨船高 (talent溢价 (premium)水涨船高 (rises)), the market becomes极度敏感 (extremely sensitive) to any personnel stir, even without实质变动 (actual changes).
This恰恰印证 (precisely confirms) a核心判断 (core judgment): the industry is far from固化 (solidified),人才还在流动 (talent is still流动 (flowing)),机会就从未消失 (opportunities have never disappeared). A true industry later stage必然 (inevitably) features核心团队稳定 (stable core teams),人才流动停滞 (stagnant talent flow),中小企业出清 (SMEs cleared out),头部格局锁死 (leading player landscape locked in).
Currently, in the large model赛道 (track),顶尖人才跨平台流动 (top talent flows跨平台 (cross-platform)),跨界玩家持续进场 (cross-border players持续 (continuously)进场 (enter)),中小团队仍有突围可能 (SME teams still have breakout potential). Talent, capital, and industry三方裹挟 (three forces intertwined),不断推高竞争门槛 (continuously raising the competitive门槛 (threshold)).人才流向 (The flow of talent) is the未来走向 (future direction).今日暗流涌动 (Today's undercurrents) are all明日格局洗牌的前兆 (precursors to tomorrow's landscape reshuffle).
04 Capital Logic Completely Reshuffled: Storytelling无效 (Ineffective), Hardcore造血 (Profitability) is the Passport The上市暴涨 (listing surge) of KNOWLEDGE ATLAS and MiniMax has再次误导市场 (once again misled the market): listing equals success,市值即实力 (market cap equals strength). This is the most dangerous industry错觉 (illusion).
After the 2024 "百模大战 (Hundred Models War)"彻底退烧 (completely cooled down), the capital logic for large models underwent颠覆性洗牌 (disruptive reshuffling). In previous years, simply贴上 (sticking) the "large model"标签 (label) could轻松融资 (easily secure financing); the larger the参数 (parameters), the louder the口号 (slogans), the more资本追捧 (capital追捧 (chased)). Now资本全面清醒 (capital is fully清醒 (sober)). Models without核心技术 (core tech),无落地场景 (no implementation scenarios),无变现能力 (no monetization ability) find无人问津 (no one interested), regardless of hype.
Industry financing极度向头部集中 (extremely concentrated towards the头部 (top)). Over ninety percent of funds flow to the four leading newcomers: KNOWLEDGE ATLAS, MiniMax, DeepSeek, and Kimi, while尾部中小玩家 (tail-end SMEs) are快速出清 (quickly cleared out). The industry has彻底从 (completely shifted from) "粗放扩容 (extensive expansion)"转向精英淘汰赛 (to an elite淘汰赛 (elimination tournament)).
Capital评判标准 (evaluation criteria) have become极度现实且苛刻 (extremely realistic and苛刻 (demanding)): First,是否存在不可替代的底层技术壁垒 (Is there an irreplaceable underlying technological barrier?). Second,能否跑通可持续的商业化闭环 (Can a sustainable commercial closed loop be established?). Third,能否实现低成本、规模化技术输出 (Can low-cost, scalable technology output be achieved?). Fourth,有无全球市场长期扩张潜力 (Is there long-term global market expansion potential?).
Listing is merely one融资方式 (financing method), never the成功终点 (success终点 (endpoint)). KNOWLEDGE ATLAS and MiniMax's high market caps离不开 (cannot be separated from) the情绪溢价 (sentiment溢价 (premium)) of scarce A-share targets. Whether their市值能否站稳 (market caps can stand firm) ultimately回归 (returns to) technological iteration and商业变现 (commercial monetization). DeepSeek and Kimi暂缓上市 (postponing listing) and securing大额再融资 (large refinancing) is not due to资本不认可 (capital disapproval); it is放弃短期泡沫 (abandoning short-term泡沫 (bubbles)) to蓄力长期技术霸权 (accumulate strength for long-term technological霸权 (hegemony)).
Future large model players that can survive will be of only two types: either those掌握底层核心技术 (mastering core underlying technology), becoming行业基础设施 (industry infrastructure); or those深度绑定垂直场景 (deeply绑定 (bound to) vertical scenarios),实现稳定盈利造血 (achieving stable profitability and造血 (self-sufficiency)). Players relying on概念 (concepts),流量 (traffic), or母公司输血 (parent company输血 (transfusions)) will终将被彻底淘汰 (ultimately be completely淘汰 (eliminated)).
05 Don't Deceive Yourselves: China's Large Models Are Far From the Later Stage There are always those in the industry急于 (eager) to declare the大模型定终局 (large model endgame): newcomers listed,格局初定 (landscape初步 (initially) defined),巨头成型 (giants formed),下半场开启 (later stage开启 (begun)). Setting aside sentiment and回归产业本质 (returning to industrial fundamentals). Judging whether the industry has entered its later stages has only three core standards:技术路线固化 (solidified technology roadmaps),商业模式闭环 (closed-loop business models),头部格局锁死 (locked-in leading player landscape).
Comparing against these three points, China's large models haven't even completed the上半场关键期 (key period of the early stage).
First,技术路线远未定型 (technology roadmaps are far from定型 (set)). Currently, Coding capability becoming the core competitive indicator is only because code generation is最易落地 (easiest to implement),最快变现 (fastest to monetize); it is绝非 (absolutely not) the大模型终极形态 (ultimate form of large models). Autonomous agents,端云协同 (end-cloud synergy),多模态深度融合 (deep multimodal integration),具身智能 (embodied intelligence),通用推理进化 (general reasoning evolution)—future technological directions are充满变数 (full of variables). Today's code advantage may be颠覆 (overturned) by new approaches tomorrow.
Second,商业化远未跑通 (commercialization is far from established). C-end traffic, AI assistants, and agents are only the商业化初级形态 (initial forms of commercialization). The industry's真正的终极市场 (true ultimate market) lies in企业数字化 (enterprise digitalization),产业智能化 (industrial intelligence),全球开发者服务 (global developer services),软件开发范式革新 (software development paradigm innovation). Currently, the vast majority of domestic players have not achieved可持续、规模化盈利 (sustainable, scaled profitability); the商业终局 (commercial endgame) is远未到来 (far from arrived).
Third,全球竞争才刚刚开始 (global competition has only just begun).国产大模型 (Domestic large models) achieving breakthroughs in global API call volumes is only出海第一步 (the first step in going global).海外市场 (Overseas markets')合规门槛 (compliance thresholds),技术竞争 (technological competition),生态壁垒 (ecosystem barriers),品牌认知 (brand recognition) are远比国内残酷 (far more残酷 (brutal) than domestically). The真正的全球硬仗 (real global硬仗 (hard battle)) has尚未全面打响 (not yet fully打响 (commenced)).
Fourth,行业格局随时重构 (the industry landscape can be重构 (restructured) at any time).跨界玩家持续进场 (Cross-border players持续 (continuously)进场 (enter)),核心人才暗流涌动 (core talent undercurrents涌动 (churn)),资本仍在狂热加注 (capital remains狂热 (feverishly)加注 (piling in)).没有任何一家玩家 (No single player) can锁定永久优势 (lock in permanent advantage). Today's头部选手 (leading players),稍有松懈 (if they slacken slightly),就会被超车 (will be overtaken); today's大厂巨头 (giant corporations),不突破技术瓶颈 (if they fail to突破 (break through) technological瓶颈 (bottlenecks)),就会被边缘化 (will be marginalized).
Comments