Tencent's Tang Daosheng Comments on Yao Shunyu, Hunyuan 3, and Yuanbao 腾讯汤道生评价姚顺雨、混元 3和元宝

Yao Shunyu didn’t directly address “fast or slow”; instead, he offered two game-theoretic perspectives: long-term and multi-dimensional. This felt like a carefully crafted “disclaimer” yet also a strategic recalibration. His subtext was clear: don’t measure AI with the internet playbook of “rapid territory expansion and DAU chasing.” AI is a heavy-weight battle involving intelligence pricing—demanding significant assets, computing power, and real-world implementation. Consequently, we witnessed Tencent’s sharp pivot: the consumer-facing app “Yuanbao,” which had been heavily marketed at the start of the year, nearly vanished, replaced by a series of hardcore-sounding, enterprise-oriented efficiency tools like CodeBuddy and WorkBuddy. Martin Lau was blunt during the earnings call: high-value use cases matter more than DAU; the value of intelligence lies in what people are willing to pay for it. In other words, stop fixating on users who only want free chatbots and start looking at how much enterprises are willing to spend for “efficiency.”

This strategic shift is pragmatic to the point of being ruthless. Tencent has finally acknowledged that in the “attention race” for consumer-facing large language model applications, it may not rival players like Doubao. User engagement has fallen behind, and those seemingly “sexy” chatbots burn astronomical computing power while generating meager ad revenue—or perhaps even becoming a cost black hole. Thus, Tencent has redirected its gun towards the enterprise sector. Business clients have clearer willingness to pay and more straightforward commercial models, placing them closer to revenue. This is a classic maneuver of avoiding strength and targeting weakness, and also the most prudent “return to value” for a giant amid uncertain times.

But questions arise. Not to mention formidable competitors on this path (Microsoft, Google, Alibaba Cloud), even the “shovel problem” Tencent itself raised is worrying. Computing power is tight, and token costs remain high. Tang Daosheng admitted that under current inference costs, it’s difficult to cover consumer-facing services through advertising alone. This implies that even for enterprise clients, they might face the dilemma of “efficiency improved, but the bill is terrifying.” Tencent has found the “gold mine,” but the “shovels” to mine it are not only expensive but may also be insufficient. More subtly, while Tang Daosheng spoke at length about strategy backstage, he also casually remarked, “Commercialization is not our current priority,” and “We haven’t set commercialization targets for the Buddy team.” This echoes the saying that when someone claims “I’m not in a rush to make money,” they’re often doing the most meticulous calculations. Is this strategic patience, or a “strategic cover-up” when costs are too high and the profit model is too hard to justify?

The most paradoxical avoidance remains the elephant in the room—WeChat. Throughout the dialogue, Tang Daosheng never mentioned the possibilities of AI in WeChat. When asked, he deflected with polished rhetoric like “Tencent’s ecosystem is diverse, and it’s hard to guarantee every segment will lead.” This exposes the deepest contradiction in Tencent’s AI strategy: it possesses China’s largest social graph and content ecosystem (WeChat), which should be the most fertile and imaginative soil for AI implementation. Yet, for the sake of “safety,” “compliance,” and avoiding disruption to existing businesses, it chose to experiment first in the enterprise sector, temporarily sealing WeChat—the nuclear weapon. Is this prudence, or fear of disruptive innovation? As competitors thrive outside the WeChat ecosystem, Tencent’s approach of “internal horse racing” while daring not to let its most core product compete is akin to tying its own hands.

Yao Shunyu’s theory of a “long-term game” provides a theoretical basis for Tencent’s “slowness.” But business competition never gives you time to adjust slowly. The consumer battlefield changes rapidly, and the enterprise landscape is far from solidified. Tencent’s bet on “high-value use cases” and efficiency improvements is undoubtedly a profound self-correction—a return to commercial fundamentals. However, once this train departs, it means facing the brutal computing arms race, the minutiae of vertical scenario implementation, and a long investment period before “intelligence” is truly priced. When Tang Daosheng spoke of “having walked through highs and lows,” did he anticipate that in this new AI cycle, the trough might be longer than expected, and the cost of “persisting through the entire cycle” could far exceed a lucrative enterprise contract? The silence of WeChat hangs like a sword above, reminding everyone: Tencent’s AI story is far from over. But at least today, it has chosen to temporarily close the most tempting yet dangerous door and instead delve into a deeper, tougher street fight.

汤道生把问题抛给姚顺雨时，场面有一瞬间的尴尬，但很快被一种更深层的张力所取代——“很多人说腾讯在AI上慢了，你觉得我们真的慢了吗？” 这句话像一根针，刺破了腾讯云AI大会精心布置的、满是Agent和B端场景的华丽帷幕。台下心照不宣：大家来看的不是你的效率工具，而是微信里那个呼之欲出却始终沉默的AI身影。但镜头前，一个27岁的天才科学家，用近乎哲思的方式，接住了这道来自高层的灵魂拷问。

姚顺雨没有直接回答“快慢”，而是抛出了两个游戏论断：长期的，多元的。这像是一场精心设计的“免责声明”，又像是一次战略校准。他的潜台词很清晰：别用互联网那套“快速跑马圈地、追求DAU”的尺子来量AI。AI是场关于“智能”定价的重资产、重算力、重落地的硬仗。于是，我们看到了腾讯的急转弯——年初还在营销上大撒币的C端应用“元宝”几乎隐身，取而代之的是CodeBuddy、WorkBuddy这一系列听起来很硬核、很To B的“提效工具”。刘炽平在财报会上的话说得很直白：高价值用例，比DAU重要；智能的价值，在于人愿意为它付多少钱。翻译一下：别整天盯着那些只愿白嫖的聊天用户，去看看企业愿意为“效率”掏多少真金白银。

这战略转向，务实得近乎冷酷。腾讯终于承认，在C端大模型应用的“注意力竞赛”中，自己可能不是豆包们的对手。用户活跃度被甩在身后，而那些看似“性感”的聊天机器人，烧的是天文数字的算力，赚的是微薄的广告费，甚至可能是个成本黑洞。于是，腾讯调转枪口，杀向B端。企业客户，付费意愿明确，商业模式清晰，确实离钱更近。这是一个经典的避实击虚，也是巨头在不确定时代最稳妥的“价值回归”。

但问题也随之而来。且不说这条路上强敌环伺（微软、谷歌、阿里云），单是腾讯自己喊出的“铲子问题”就足以让人捏一把汗。算力紧张，token成本居高不下。汤道生亲口承认，当前推理成本下，很难靠广告模式覆盖C端服务。这意味着，即便是B端，客户也可能陷入“效率提上去了，账单也吓人”的困境。腾讯找到了“金矿”，但挖矿的“铲子”不仅贵，而且可能不够用。更微妙的是，汤道生在后台一边大谈战略，一边却轻描淡写地表示：“当前商业化不是我们的重点”，“没有给Buddy团队设商业化目标”。这像极了一个人在说“我不急着赚钱”时，往往正是算盘打得最响的时候。这到底是战略定力，还是成本太高、盈利模式太难跑通时的“战略性遮掩”？

最吊诡的回避，依然是那个房间里的大象——微信。整场对话，汤道生对微信AI的可能性只字未提。当被问及，便以“腾讯业态多元，很难保证每个板块都领先”这样滴水不漏的话术搪塞过去。这暴露了腾讯AI战略中最深刻的矛盾：它拥有全中国最庞大的社交关系和内容生态（微信），这本应是AI落地最肥沃、最具想象力的土壤。但为了“安全”、“合规”以及避免对现有业务的冲击，它选择了在B端先行先试，把微信这个核武器暂时封印。这到底是稳健，还是对颠覆性创新的恐惧？当竞争对手在微信生态外已经风生水起，腾讯这种“内部赛马”却不敢让最核心的产品参赛的做法，无异于自缚手脚。

姚顺雨的“长期游戏”论，为腾讯的“慢”提供了理论依据。但商业竞争从来不给你慢慢调整的时间。C端战场瞬息万变，B端格局也远未固化。腾讯将宝押在“高价值用例”和效率提升上，无疑是一次深刻的自我纠偏，是一次向商业本质的回归。但这趟列车一旦开动，便意味着要直面算力军备竞赛的残酷、垂直场景落地的琐碎，以及在“智能”被真正标价之前漫长的投入期。当汤道生说出“走过高潮走过低谷”时，他是否预见到，在AI这个新周期里，低谷可能比想象得更长，而“坚持走过整个周期”所付出的代价，可能远超一份漂亮的B端合同？微信的沉默，像一柄悬顶之剑，提醒着所有人：腾讯的AI故事，远未到盖棺定论时。但至少今天，它选择暂时关上那扇最诱人、也最危险的门，转而扎进了更深、更硬的巷战里。