Manulife Hong Kong and Alibaba Cloud Sign Strategic Partnership to Accelerate Large-Scale AI Implementation in Insurance Industry 宏利香港与阿里云达成战略合作，加速保险业AI规模化落地

Another "Strategic Cooperation Memorandum," this time between an insurance company and a cloud provider. Manulife Hong Kong and Alibaba Cloud are teaming up, claiming to accelerate the large-scale application of AI across all insurance scenarios and explore new paths for intelligent transformation. The phrasing is so precise it feels copy-pasted directly from slide 20 of a PPT. Can the insurance industry—an ancient, conservative sector that has struggled with paperwork and processes for centuries—really be "accelerated" with a single memorandum? I’m highly skeptical. True transformation never lies in the red cloth at signing ceremonies but in the details of the terms: How will data be integrated? On whose cloud will the models be trained? Will AI be used to ease the workload of underwriters, or to replace them entirely? More crucially, who will bear the cost when the model makes a mistake? When an AI error leads to a claims dispute, can this memorandum serve as a legal basis? Most likely not. The cooperation is real, but the phrase "large-scale implementation" is as light and insubstantial as a bubble.

On one side, the insurance industry’s "intelligence" remains largely theoretical; on the other, XPeng Motors CEO He Xiaopeng revealed in a livestream that the GX model has already secured 1,000 "blind orders" in the Middle East without even announcing a price. The numbers are eye-catching, but I have to ask: How high is the barrier for these "blind orders"? Is the deposit a refundable few hundred yuan or a locked-in tens of thousands? The marketing number games in the automotive industry have always been more complex than code. Still, this at least demonstrates two things: first, the brand halo of Chinese smart cars does hold sway in certain overseas markets; second, He Xiaopeng understands traffic and knows how to turn a factory livestream into a global pre-sale launch. This kind of "hot before selling" operation is far more compelling—and far more substantive—than the insurance company’s paper agreement.

The real climax is overseas. ChatGPT and Codex have officially merged, aiming to bring a "super Agent" to 1 billion people. This is undoubtedly a major integration in the AI application layer, combining conversation, coding, and task execution into one cohesive force. But my concern is this: When "Agent" becomes a standard feature for everyone, will it become a handy "super assistant" for all, or will it turn into a "super app" that everyone is forced to learn, thereby intensifying the rat race in digital skills? Don’t rush to celebrate just yet. Meanwhile, Microsoft Windows is also turning itself into an "Agent workstation." Giants are eagerly stuffing AI into every crevice of operating systems. But here’s the question: What if I just want to quietly write a document, rather than have my computer constantly ready to "intelligently" suggest what I should do next? This kind of omnipresent "empowerment" often feels more like the imposition of an "empowerment anxiety."

The battle in the chip sector is also heating up. Intel has unveiled a "big move," claiming it will end NVIDIA’s computing monopoly. The courage is commendable, but the market response may be underwhelming. In the AI training chip arena, NVIDIA has built a formidable moat with its CUDA ecosystem and extreme performance optimization. Intel’s entry now feels more like defensive maneuvering than disruption. Unless it can deliver a generational, ecosystem-compatible killer product, the "big move" might just end up being a "big letdown."

The domestic tech landscape presents a different picture. Tencent’s stock price has surged to a three-year high. The market is clearly voting for its earnings and AI narrative, but don’t forget—its core moat in social media and gaming has never been fundamentally redefined by AI. Another headline, "An estimated 110 billion yuan in childcare subsidies will be arranged this year," may seem unrelated to AI, but it reveals the current macro sentiment: Under economic growth pressures, livelihood issues like childbirth are receiving unprecedented resource allocation. This may matter more to the social fundamentals than any increase in AI model parameters. Meanwhile, "Volcano Engine" has raised its MaaS revenue target to 15 billion yuan, bluntly indicating that the competition in large models has moved from technical showboating to a white-hot phase of commercial harvesting. The winner will be whoever can faster turn "computing power" into "revenue."

Finally, it’s worth turning our gaze to DeepSeek. A discussion about whether it can save China $1 trillion in computing expenses tackles one of the core pain points in current AI development: cost. While everyone chases larger models and higher parameters, finding more efficient and economical algorithms to achieve equal or better results is the key to making AI truly accessible. Compared to flashy strategic cooperation agreements and unverified overseas blind orders, this pragmatic pursuit of efficiency and cost-effectiveness may well be the foundation for how far a technological revolution can go. Giants are busy signing deals, hosting livestreams, and making announcements, but the real determinant of victory may lie in these dry but critical numbers.

又一份“战略合作备忘录”，这次是保险公司和云厂商。宏利香港与阿里云携手，宣称要加速AI在保险全场景的规模化应用，探索智能化升级新路径。这话术精准得像从PPT第20页直接复制粘贴而来。保险业这个古老、保守、与文件和流程搏斗了几个世纪的行业，真的能靠一纸备忘录就被“加速”吗？我表示严重怀疑。真正的变革从来不在签约仪式的红布上，而在具体条款的细节里：数据怎么打通？模型训在谁的云上？是用AI给核保员减负，还是直接替掉他们的工作？更关键的是，谁来为模型出的错买单？当AI错误导致一笔理赔纠纷时，这份备忘录能变成法律依据吗？大概率不能。合作是真的，但“规模化落地”这五个字，轻飘飘得如同泡沫。

这边厢，保险业的“智能化”还在务虚；那边厢，小鹏汽车CEO何小鹏的直播透露，GX车型在海外连价格都没公布，就拿到了中东市场的1000笔“盲订”。数字很亮眼，但我得问一句：这“盲订”的门槛有多高？订金是可退的几百块，还是锁死的几万元？汽车行业的营销数字游戏，向来比代码还复杂。不过，这至少说明了两点：一是中国智能汽车的品牌光环，在某些海外市场确实管用；二是何小鹏很懂流量，知道怎么把一场工厂直播变成全球预售发布会。这种“未售先热”的操作，比保险公司那纸协议性感多了，也实在多了。

真正的高潮在海外。ChatGPT和Codex官宣合体，号称要让10亿人喜提“超级Agent”。这无疑是AI应用层的一次重磅整合，把对话、编码、任务执行拧成一股绳。但我的担忧是，当“Agent”成为人人皆可拥有的标配时，它是会成为人人用得顺手的“超级助手”，还是会沦为人人不得不学的“超级应用”，从而加剧数字技能的内卷？别急着欢呼。与此同时，微软Windows也要把自己变成“Agent工位”。巨头们都在急不可耐地把AI塞进操作系统的每一个缝隙，问题是，我是不是只是想安静地写个文档，而不是让我的电脑时刻准备着“智能”地建议我下一步该干嘛？这种无所不在的“赋能”，有时更像是一种“赋能焦虑”的强加。

芯片领域的战事也在升温。英特尔放出“大招”，号称要终结英伟达的算力垄断。勇气可嘉，但市场反应恐怕会很骨感。在AI训练芯片这个赛道，英伟达靠着CUDA生态和极致的性能优化，已经筑起了深不见底的护城河。英特尔此时入场，更像是在应战，而不是在颠覆。除非它能拿出跨代际的、且生态兼容的杀手级产品，否则，“大招”很可能只是“大嘘声”。

国内科技圈则呈现另一幅图景。腾讯股价暴涨，创三年多新高。市场显然在为它的业绩和AI叙事投票，但别忘了，它的核心护城河——社交和游戏——从未被AI完全重新定义过。另一条新闻，“预计今年安排约1100亿育儿补贴”，这看似与AI无关，却揭示了当下的宏观情绪：在经济增长压力下，生育等民生问题获得了前所未有的资源倾斜。这或许比任何AI模型参数的增长，都更关乎社会的基本面。而“火山引擎”提升MaaS营收目标至150亿，则赤裸裸地显示，大模型竞争已从技术炫技进入白热化的商业收割阶段。谁能更快地把“算力”变成“收入”，谁才是赢家。

最后，值得把目光投向DeepSeek。一篇关于它能否为中国节省1万亿美元算力支出的讨论，抓住了当前AI发展的核心痛点之一：成本。在所有人追逐更大模型、更高参数时，如何用更高效、更经济的算法实现同等甚至更优的效果，这才是决定AI能否真正普惠的关键。比起那些华丽的战略合作和未经证实的海外盲订，这种对效率和成本的极致务实追求，或许才是一场技术革命能走远的地基。巨头们忙着签约、直播和发布，但真正的胜负手，可能就在这些枯燥却致命的数字里。