Lenovo Routes AI Queries to Devices to Curb Soaring Token Costs

联想将AI查询分流至终端以遏制Token成本飙升 联想正通过在笔记本电脑和手机本地部署智能体AI来运行压缩模型，在削减Token开支的同时，仅将复杂工作负载分流至云端。 联想CFO Winston Cheng探讨了这家全球科技巨头在当前AI扩张期如何管理资本配置、运营成本和Token预算。作为技术价值链中的“中间力量”，联想在保持PC制造主导地位的同时，正迅速拓展模块化数据中心建设和液冷基础设施。Cheng还就AI落地的经济现实分享了核心见解，包括实现成本优化的模型路由、高带宽内存及电力方面的供应链瓶颈，以及中国激烈的市场竞争如何推动实现超高效、低成本的Token生成。 联想将当前阶段定义为“AI十年”的第一年，正加速从传统PC巨头转型为涵盖端侧设备与云端AI基础设施（pocket-to-cloud）的全球服务商。 (5:00) AI的普及正将企业软件支出由固定的SaaS订阅模式转向按Token计费的弹性模型，Token预算与运营支出（Opex）管理因此成为CFO的核心把控点。 (8:30) 针对企业AI治理，联想主张推行“智能体AI”（Agentic AI）架构，在端侧运行轻量压缩模型以保障隐私与成本，并采取主动收紧传统业务预算的方式倒逼各部门提高AI落地效率。 (15:20) 联想将自身定位为连接欧美科技巨头与中国AI生态的“中间力量”（Middle Power），依托横跨全球的30家制造工厂以及液冷技术，提供端到端的模块化AI数据中心部署方案。 (23:40) 面对国内“内卷”竞争与出口受限，中国企业在资源约束下被迫极度优化成本，模型Token生成成本显著低于西方同行；联想则保持平台中立，通过设备层编排器灵活调度不同大模型以优化性价比。 (45:00)

Lenovo Chief Financial Officer Winston Cheng argues that navigating the artificial intelligence boom requires disciplined capital allocation, managing token consumption as a variable operational expense while establishing the hardware giant as an essential, platform-neutral "middle power." Balancing historic dividend distributions with infrastructure investments, Lenovo aims to address physical data center constraints and deploy on-device AI across 180 global markets. Disciplined Capital Allocation and AI Operational Governance (3:00) Cheng explains that shifting from fixed software subscriptions to variable, usage-based token models turns AI into a primary operational expenditure challenge for corporate finance. Rather than micromanaging individual token quotas or permitting uncapped spending, Cheng advocates spurring AI adoption by intentionally restricting traditional operating budgets, compelling employees to use automated tools to complete tasks. Spending must be justified by quantifiable returns, such as optimizing real-time pricing and inventory across Lenovo's 180 markets, eliminating redundant external data subscriptions, and automating back-office finance workflows in tax, treasury, and accounting. At the enterprise level, capital allocation balances ongoing AI investments with immediate shareholder returns. Lenovo distributed the highest dividend in corporate history for the fiscal year ending March 31. Cheng acknowledges that margins may fluctuate quarterly as growth spending scales, but Lenovo’s overarching strategy targets sustained revenue growth alongside structural, long-term margin expansion. 'Middle Power' Positioning and Model Neutrality (23:40) Drawing on Mark Carney’s framework separating superpowers from middle powers, Cheng positions Lenovo—generating over $80 billion in annual revenue—as an indispensable "middle power." While trillion-dollar platform architects and chip designers dominate market valuations, Lenovo extracts value as an end-to-end integrator across its network of 30 factories and 2,000 suppliers. Its footprint spans personal devices, edge hardware, and enterprise servers. To mitigate technological risk, Lenovo remains platform-neutral rather than tethering itself to a single model provider like OpenAI, Anthropic, or Google. Lenovo instead builds an orchestration layer directly into device operating systems. This architecture executes compressed, privacy-compliant agentic models locally on PCs and phones to curb token expenses, routing only complex queries to external cloud infrastructure or custom hyperscaler silicon. Physical Data Center Scaling and Supply Bottlenecks (28:40) Cheng characterizes AI as an intensely physical industry constrained by multi-year supply bottlenecks. With semiconductor fabrication plants taking two to three years to build, the sector faces shortages across high-bandwidth memory, GPUs, CPUs, and grid power. To overcome these barriers, Lenovo's Solutions and Services Group deploys modular data centers in six to nine months, leverages 11,000-rack liquid cooling for dense compute, and expands capacity in regions with abundant energy, such as solar power in Saudi Arabia. Addressing fears of artificial demand inflation via customer "double ordering," Cheng distinguishes speculative early-stage pipelines from binding, multi-year backlogs required to process exponential data growth. Geopolitically, because trade restrictions limit foreign data center expansion within mainland China, international infrastructure is shifting to regional hubs in Malaysia and Indonesia, alongside high-density vertical facilities in Hong Kong and Singapore. U.S. Compute Scale Versus Chinese Cost Involution (23:40) The AI landscape reflects two contrasting operating paradigms. Cheng notes that while U.S. developers maintain superior access to advanced silicon, Chinese firms navigate export controls through relentless cost optimization. Driven by intense domestic competition—termed involution (neijuan)—Chinese labs have engineered ultra-low inference costs. Cheng cites market estimates indicating Chinese models like DeepSeek produce tokens at roughly 1/50th the cost of U.S. counterparts. This divergence shows that raw capital expenditure alone does not guarantee dominance. While Western hyperscalers develop custom silicon to hedge against vendor lock-in and high component costs, Chinese firms demonstrate resilience under tight resource constraints. Lenovo bridges both environments through local manufacturing agility, regulatory compliance across divergent markets, and device-level software that can route across diverse Western and Chinese model ecosystems. Lenovo's strategy demonstrates how legacy hardware makers seek to capitalize on the AI expansion by managing physical supply constraints and software integration without making exclusive platform bets. However, whether enterprise efficiency gains will translate into durable hardware demand remains dependent on how successfully businesses convert early AI experimentation into measurable profitability.

联想集团（Lenovo）首席财务官郑弘孟（Winston Cheng）指出，在当前人工智能基建投资热潮中，企业财务管理的核心已从单纯的算力追逐转向严格的投资回报率（ROI）考核与运营支出（Opex）控制。联想将自身定位为科技价值链中的“中间力量”，依托全球化制造与平台中立的设备编排技术，在算力瓶颈与地缘碎片化环境中实现资本配置与股东回报的平衡。 AI支出的财务约束与ROI量化评估 (8:30) 随着企业AI应用从固定软件订阅模式转向基于Token消耗的浮动计量模式，管理内部算力支出成为CFO的核心议题。郑弘孟提到业内曾有工程师单月消耗1亿美元Token的极端案例，强调非预算内的高额支出必须建立在明确的投资回报率之上。在拥有逾800亿美元年营收规模的组织中，微观管理单个员工的Token额度并不现实；更为有效的机制是设定明确的领域级能力指标，并主动压缩传统业务预算，以此倒逼团队利用AI工具提升效率。 在投资回报的量化方面，企业正在脱离宽泛的生产力口号，转向具体的关键绩效指标（KPI）。郑弘孟列举了数个可衡量的落地场景：通过AI模型在联想覆盖的180个市场中实时优化供应链配置与动态渠道定价，能直接节省数额可观的硬性成本；在财务和研究部门减少对高昂外部数据订阅的依赖；在投资者关系（IR）与并购团队中借助AI提高资本成本分析和标的评估效率；以及用内部AI内容生成替代高昂的外部市场营销和视频制作外包支出。 “中间力量”定位与平台中立战略 (23:40) 借鉴马克·卡尼（Mark Carney）在达沃斯论坛提出的地缘框架，联想管理层将自身定位为科技价值链中的“中间力量”，而非拥有数万亿美元市值的超级大国企业（如微软、谷歌等操作系统巨头，或英伟达、AMD等芯片设计商）。作为全球出货量第一的个人电脑制造商，联想协同超过2000家供应商，业务覆盖PC、平板、智能手机（包含ThinkPad与Legion等关键产品线）及边缘服务器，构成了连接底层算力芯片与终端企业客户的关键使能平台。 面对快速迭代的底层模型生态，联想采取了平台中立策略，避免将自身与单一模型厂商（如OpenAI、Anthropic或Google）绑定。联想在端侧设备中构建系统级“编排器”（Orchestrator），将压缩后的小型大语言模型部署于本地运行以保障数据隐私并降低Token成本，同时将高复杂度任务透明化路由至云端。为了规避全球供应链碎片化与关税风险，联想在全球布局了30家制造工厂，并在匈牙利、墨西哥、沙特利雅得等地设立了12个战略制造枢纽，以满足本地化监管与欧洲等市场对可循环材料的高标准合规要求。 数据中心供应链瓶颈与订单积压真实性 (28:40) 在企业级服务器市场，产品竞争已脱离单纯的硬件参数拼比，转向端到端的定制化系统交付。联想方案服务业务集团（SSG）能够利用模块化方案在6至9个月内为客户完成数据中心整体搭建，并提供具备11000个机架交付能力的液冷技术，以支撑高密度GPU运算所带来的严苛散热需求。 针对当前AI基建的供应瓶颈，郑弘孟指出高带宽内存（HBM）、DRAM、GPU、CPU以及电力基础设施和变压器普遍短缺，且新建半导体晶圆厂通常需要2至3年周期；为此，联想正协助客户将算力部署转移至电力充沛且廉价的区域（如沙特阿拉伯的低成本太阳能电网）。针对市场关于客户“重复下单”导致算力泡沫的担忧，郑弘孟澄清称，尽管销售漏斗前端的初期线索可能存在对冲性重复需求，但正式进入积压订单（Backlog）的均为此类具备法律约束力的严肃承诺，生成式AI所引发的海量数据膨胀决定了算力扩张具备坚实的底层支撑。 中美AI生态分化与资本回报平衡 (50:00) 中美两国在AI发展路径上展现出鲜明的分化特征：美国的核心优势在于获取最先进的前沿芯片与高效算力，促使超大规模云服务商（Hyperscalers）自主研发定制芯片以摆脱单一芯片供应商依赖并压低长期成本；而中国企业受制于出口管制与先进制程瓶颈，在激烈的本土竞争（“内卷”）下被迫向极致的成本效益创新演进，行业传闻显示以DeepSeek为代表的中国模型在单位Token生成成本上仅为美国同类模型的约五十分之一。 在资本配置战略上，联想在截至3月31日的财年派发了公司历史上最高额度的股息。面对AI基础设施扩张所需的巨额资本开支，管理层坚持在即时现金分红与研发再投资之间寻求动态平衡。郑弘孟强调，尽管季度间的短期利润率可能伴随基建投资规模扩大而出现波动，但长线财务目标依然锁定在推动营收扩张的同时实现结构性利润率增长。 当前整个科技行业仍处于“AI十年”的初级阶段，非超大规模云厂商在摸索运营支出优化标准的同时，仍面临长达数年的芯片与电网供应瓶颈；企业如何在保障高额股东分红的前提下平衡重资本研发投入，仍将是检验硬件巨头财务韧性的长期考题。

## Background 本期节目由彭博社（Bloomberg）旗下的知名财经播客《Odd Lots》于 2026 年 6 月 27 日发布。节目聚焦全球科技巨头在 AI 军备竞赛中近乎无上限的支出狂潮，邀请了联想集团执行副总裁兼首席财务官郑弘孟（Winston Cheng），深入探讨联想如何在人类历史上最大规模的资本开支周期中进行资本分配、将智能体融入终端设备、联手英伟达共建“AI 云”基础设施，以及如何应对供应链关税与中国市场的“内卷”竞争。 ## TL;DR 面对生成式 AI 引发的算力狂热与使用量计费（Token）激增，联想 CFO 郑弘孟分享了从“口袋到云端”的全栈布局与极为严苛务实的资本配置哲学。联想坚守生态链“中间力量”的定位，一方面推行端云协同与中立的端侧智能体编排器以防范算力账单失控，另一方面凭借横跨全球 30 座工厂的制造韧性、模块化数据中心与液冷交付能力承接长周期算力基建。节目同时深入剖析了中美 AI 演进路径的差异，揭示了中国极端“内卷”倒逼出的极致单位成本优势与全球竞争力，并探讨了在追求长期业务增长与维持高额现金分红之间的平衡之道。 ## Atmosphere & style 整场对话在香港现场录制，氛围既专业严谨又充满机智幽默的即兴碰撞。主播 Joe Weisenthal 与 Tracy Alloway 凭借标志性的敏锐洞察，从经典的 ThinkPad“小红点”情怀与工程师单月刷爆一亿美元 Token 的圈内奇闻切入，将宏大的算力叙事无缝锚定在冷酷具象的财务资产负债表上。郑弘孟的应答条理清晰、直言不讳且不打官腔，既有大厂掌门人驾驭地缘关税突变与供应链危机的沉着老练，又透着看透咨询泡沫与资本狂欢后的清醒务实，整期访谈在硬核的企业财务推演与通俗生动的现实案例之间流畅自如地切换。 ## Bulletin - `30:00` 联想明确将自身定位于AI生态中的“中间力量”，在全球布局30座制造基地并拥有逾1.1万个液冷机架交付能力，能以模块化方案将整座数据中心的建设周期压缩至6至9个月。 - `05:22` 针对大模型时代可能出现的Token开销失控，联想推行“端侧AI+AI编排器”架构，通过本地压缩模型处理高频及高隐私任务，并在多模型生态中保持中立路由分发。 - `10:00` 联想CFO郑弘孟提出通过“断供传统预算”倒逼业务拥抱AI工具，将AI投资回报严格锚定于180个市场的供应链库存与定价优化，以及投资者关系、并购等高杠杆职能。 - `36:51` 全球AI基础设施面临至少2至3年的多重物理瓶颈，包括存储芯片晶圆厂建设周期、大型变压器紧缺与电网容量枯竭，正驱动算力部署向沙特等清洁能源枢纽转移。 ## Key takeaways - `07:30` 企业软件正从平滑可预测的固定SaaS席位费，急剧转向基于算力消耗与Token生成量的动态使用量计费模式，CFO的核心职责是在建立风控的同时进行主动的资本分配。 - `05:22` 联想通过在终端部署“AI编排器”实现端云协同，利用轻量化端侧大模型优先处理高频常规与隐私任务，唯有复杂超长计算才路由至云端，有效遏制非预期的Token飞涨。 - `10:00` 评估AI投资回报率（ROI）必须摆脱泛泛而谈的提效口号，落脚于全球180个市场的渠道库存资金占用优化，以及针对IR、并购等虽人员精简但牵动数十亿资本成本的高杠杆部门。 - `15:45` 面对沉迷于交互幻觉而缺乏实质产出的“AI精神病”，有效的管理手段不是微观监控个人配额，而是通过主动削减传统流程预算、切断旧模式退路，倒逼团队挖掘AI落地潜能。 - `23:05` 联想将自身定位于科技价值链中的“中间力量”，认为长期以来底层芯片与操作系统攫取了万亿美元估值，而集聚供应链、高良率制造及全球交付信任的硬件集成商价值严重被低估，AI浪潮正开启价值重估窗口。 - `26:45` 受制于先进芯片获取限制与国内白热化的“内卷”竞争，中国AI团队（如DeepSeek）被迫将单Token算力效能与成本控制推向极限，其展现出的达尔文式生存韧性在全球具备极强性价比优势。 - `30:00` AI服务器竞争已从单机组装升级为深度的交钥匙工程，涵盖端到端测试、高密度GPU机架组装及上万机架的液冷工程，联想通过模块化设计可实现数据中心6至9个月的极限交付。 - `36:51` 全球算力扩张正遭遇存储芯片扩产周期（2至3年）、变压器短缺与电网可用容量枯竭等多重物理阻碍，迫使算力资本超越传统枢纽，在全球协同匹配如沙特等地充沛且低廉的太阳能与绿色电网。 - `40:00` 针对供应链“牛鞭效应”与重复下单的担忧，实际具有法律约束力的订单积压（Backlog）与跨年度机房、电力资产投入表明，生成式AI对新增数据的几何级裂变构成了长周期的刚性算力需求支撑。 - `41:30` “SaaS末日论”下大模型代码生成虽大幅拉低了轻量应用构建门槛，但企业级系统的深层壁垒在于异构核心数据层，缺乏清洗打通内部数据的“中间层”是目前AI落地的最大现实瓶颈。 - `43:03` 面对AI数据中心对先进半导体产能的挤占，联想依赖全球顶级采购规模在基础设施与个人终端（ThinkPad及游戏第一品牌拯救者）之间动态调度芯片配额，筑牢消费硬件基本盘。 - `47:11` 超大规模云服务商自研芯片不仅是为了专有模型的微架构调优，更是面对英伟达高溢价设立的战略“解绑缓冲垫”，以压低长期总拥有成本（TCO）并掌握底层架构主动权。 - `50:00` 个人电脑在全球普遍关税政策中获得豁免，源于各国监管层将其定性为连接数字社会与保障教育生产力不可或缺的底层生存刚需，尤其在发展中国家具有极高的普惠门槛属性。 - `54:12` 在资本配置的天平上，联想在完成创纪录现金股息分红回馈股东的同时，允许短期季度投入波动，战略重心旨在通过AI基础设施与终端的持续资本注入驱动长期的利润率加速扩张。 - `55:18` 亚洲数据中心版图面临地缘政策导致的结构性割裂：中国大陆优质的电力与机房受限闲置，迫使国际资本流向东南亚、香港与日本等高成本地区，进而推高了全球AI算力部署的综合通胀水平。 ## Detailed walkthrough ### `00:00` 从香港之行到“Token”词源：AI 时代的资本与算力计量 在播客开篇的一段企业赞助内容中，展示了 AI 落地当下的具体场景：IBM 提及利用内置 AI 助手服务其全球 30 万名员工，能够直接解决 94% 的常见人力资源疑问，强调在喧嚣的概念之外交付切实的业务回报。随后，节目主播 Tracy Alloway 与 Joe Weisenthal 在香港现场开启了对话。他们分享了在此次行程中与各界交流的新鲜收获，其中最耐人寻味的一桩轶事来自前一天与百度 CFO 的非正式闲聊：在中文语境下，“Token”这一大语言模型的基本计量单位，已经被正式收录进现代汉语规范词汇，其对应的意涵巧妙融合了“字符”与“通货/代币”的双重概念。 这一词源的交汇让 Joe 大为兴奋。他回顾指出，“Token”这个词在英语语境中向来有着丰富的层次——从游乐场里用来玩投币机和街机游戏的物理代币（如 Chuck E. Cheese 的游戏代币），到区块链世界里的加密代币；而在 20 世纪中叶，语言学家就早已用它来指代语言中的词元或符号单位。如今在生成式 AI 与 LLM 大爆发的背景下，中文里将这两者统合的构词逻辑显得尤为精妙：“字币”这一称谓本身就赤裸裸地揭示出它与金钱支出（Spending）之间的直接羁绊。 随着全球科技巨头在 AI 基础设施与大模型调用上掀起巨额军备竞赛，CFO（首席财务官）这一角色的核心权重被推到了聚光灯下。Tracy 与 Joe 讨论道，眼下媒体头条往往被动辄百亿美元的数据中心 CapEx（资本支出）所占据，但在企业内部日常运营中，如何管控 OpEx（运营支出）同样充满挑战——各部门、各员工之间的 Token 使用额度与调用权限该如何科学分配？倘若两名工程师从大模型中挖掘出的业务产出迥然不同，机械地给他们赋予相同的 Token 预算显然是极其低效的。 ### `03:15` PC 情结与联想的 AI 转型：从“小红点”到全栈基础设施 在切入正题前，两位主播展开了一番饶有趣味的个人科技偏好闲聊。Joe 问及 Tracy 私下里究竟是苹果 Mac 阵营还是 Windows PC 阵营，Tracy 坦言自己目前全靠工作设备度日，但在有得选的时候毫无疑问是铁杆 PC 党，甚至直言“非常不喜欢 Mac”。Joe 深有共鸣，并自曝多年来一直对经典的 ThinkPad 笔记本情有独钟——特别是机身键盘正中央那枚标志性的红色指点杆（TrackPoint，“小红点”）。 这台带有传奇色彩的电脑正是联想（Lenovo）的标志性资产。IBM 当年将 PC 业务出售给联想，如今在席卷全球的 AI 浪潮中，联想正在走出一条与众不同的竞争路线。不同于仅在纯软件或纯公有云端发力的竞争对手，联想的打法横跨了物理硬件终端与底层基础设施：一边是将智能体（AI Agents）深度植入个人电脑与智能终端，另一边则是联手英伟达等芯片巨头，大举投资构建面向未来的“AI 云”数据中心基础设施。为了看清这轮人类历史上罕见的资本开支狂潮在硬件与企业财务视角的投射，节目邀请到了联想集团执行副总裁兼首席财务官 Winston Cheng。 ### `05:22` 从口袋到云端：联想的“端云协同”与防爆仓的 Token 经济学 Winston Cheng 将联想的核心定位概括为一家“提供从口袋到云端（Pocket-to-Cloud）全栈 AI 基础设施的全球科技公司”，服务对象覆盖广大普通消费者与各行业企业级客户。从云端算力来看，联想一方面为那些因模型预训练（Training）爆发巨大算力需求的超大规模云服务商（Hyperscalers）供应算力底座；另一方面，得益于此前收购 IBM x86 服务器业务所继承的雄厚技术底蕴，联想在 CPU 通用计算架构上根基极深，这使其在应对企业级推理（Inference）以及混合云部署需求时占据了极具优势的身位。 谈到 Token 消费对财务管理的冲击时，Winston 分享了一个科技圈内流传的极端个案：某家企业的一位软件工程师因缺乏风控，个人在一个月内竟调用消耗了高达 1 亿美元的 Token。Winston 笑着强调“这绝对不是联想内部发生的”，但他直言，任何一位 CFO 面对这种突如其来的巨额非预算开支都会感到毛骨悚然。这种混乱暴露了企业亟需更加精细化、高性价比的算力解决方案。 为此，联想大力推崇“端侧 AI（On-device AI）”与端云协同架构。用户或许前期为高性能终端硬件支付了更高的溢价，但换来的是确定的本地计算能力、坚固的数据安全防线以及无可替代的个人隐私交互环境。联想的智能体系统（Agentic AI）在本地扮演着“总调度师（Orchestrator）”的角色： ```mermaid flowchart TD accTitle: 联想端云协同 AI 调度架构 accDescr: 展现联想智能体如何在本地压缩大模型与云端大规模算力集群之间分配推理任务 subgraph Client["终端设备 (本地计算)"] User["用户交互 / Prompt"] Orchestrator["联想 Agentic AI 调度器"] LocalLLM["端侧压缩版轻量 LLM"] end subgraph Cloud["云端 / 基础设施集群"] CloudLLM["公有云 / 超级大模型"] DataCenter["AI 云基础设施 (CPU/GPU)"] end User --> Orchestrator Orchestrator -->|"常规查询 / 隐私高敏任务"| LocalLLM Orchestrator -->|"复杂计算 / 超大上下文"| CloudLLM CloudLLM --- DataCenter ``` 在这一架构中，联想 Agent 能够统合设备上并存的各种操作系统与多重子智能体。对于日常高频或隐私敏感的交互，设备直接调用压缩蒸馏后的端侧轻量化 LLM 完成本地推理，免除额外的网络往返与云端计费；只有当遇到超出本地算力极限的复杂长尾任务时，智能体才会定向将请求路由（Route）至云端模型，从而帮助个人与企业精准遏制 Token 消耗失控的风险。 ### `07:30` 工程师单月烧掉一亿美元？CFO 的资本分配逻辑与 ROI 权衡 针对 Winston 提到的“工程师单月刷爆一亿美元 Token”的骇人案例，Joe 提出了一个针锋相对的逆向思考：这笔开销对 CFO 来说难道绝对是灾难吗？假设一家企业原本规划了一个耗时数年、预算高达 5 亿美元的复杂数据库跨架构迁移大工程，而这名工程师借助前沿大模型的强悍生成能力，仅花了一个月时间、烧掉了 1 亿美元的 Token 便奇迹般提前收工——省下了 4 亿美元预算并抢回了宝贵的时间窗口。如果站在最终商业成效的角度，这难道不是一笔性价比极高、值得 CFO 开香槟庆祝的超值买卖吗？ Winston 毫不犹豫地赞同了这一商业本质。他指出，现代企业财务的核心逻辑永远建立在投资回报率（ROI）与清晰的规划之上：“作为一名 CFO，你的职责是去分配资本（Allocate Capital），而不是去卡死资本、限制资本（Constrain Capital）。”敢于在产生高价值回报的地方重注投资是财务掌门人的必修课，但大前提是企业必须对资金流向与预期回报拥有明确的度量和预期。上述案例之所以令人警惕，症结在于那是无规划、无预警的盲目消耗。 Winston 强调，整个企业级商业社会眼下正站在一个极其关键的转折点上：过去十几年间，企业早已习惯了按照固定的席位月费支付 SaaS 订阅费用，成本曲线是平滑且可预测的；然而在 AI 时代，商业模式正在急剧向基于算力消耗与 Token 生成量的动态使用量计费（Usage-based model）裂变。各大企业目前仍处于这一剧烈范式转换的探索初期，如何在激进拥抱 AI 红利的同时建立起追踪、衡量与预算化 Token 支出的全新财务秩序，正是所有高管面临的全新大考。 ### `10:00` 衡量 AI 投资回报率：从库存优化到高杠杆职能 当被问及企业如何为内部 AI 项目或 Token 支出设定具体的评估基准时，联想集团首席财务官郑弘孟（Winston Cheng）直言，许多管理层口中泛泛而谈的“提升生产力”或“削减成本”往往显得过于空泛。在联想这样庞大的全球企业中，AI 的投资回报率（ROI）必须落到极具象的财务指标上。他特别强调了联想横跨 180 个市场的庞大供应链与渠道网络，如果能借助 AI 精准预测市场需求、管理渠道库存与动态定价，而非单纯依赖分布在 180 个市场的人工作业，其中撬动的经济效益将直接体现在具体的美元数字上，绝非虚谈。 除了供应链，郑弘孟还列举了数个能够迅速量化 AI 收益的具体场景。首先是传统数据开销的压缩，许多投资圈与对冲基金的朋友已经在探讨利用 AI 取代昂贵的第三方专业数据订阅。其次是像投资者关系（IR）和并购（M&A）这类团队人数少但财务杠杆极高的职能：AI 在这里的目标并非裁减本来就不多的员工，而是大幅提升决策质量。更精准的 IR 沟通有助于提振股价并压低股权融资的资本成本；而更高效的并购分析则能从根本上优化巨额资本部署的回收期。此外，营销领域也是立竿见影的试验田，以往耗资巨大的视频制作和内容生成，如今完全可以直接交付给现有的 AI 工具来完成。 ```mermaid flowchart TD accTitle: 联想 AI 投资回报评估场景 accDescr: 展示联想如何将 AI 投资落实在具体的高价值与高杠杆业务场景中。 subgraph ScaleOps["规模效应场景（直接降本增效）"] A["180 个市场的渠道库存与需求预测"] --> B["动态定价与供应链资金占用优化"] C["传统外部数据采购与订阅"] --> D["AI 替代与开支缩减"] E["市场营销与视频内容制作"] --> F["AI 原生内容直接生成"] end subgraph HighLeverage["高杠杆职能（决策赋能）"] G["投资者关系 (IR)"] --> H["优化市场预期与降低资本成本"] I["企业并购 (M&A)"] --> J["提高资本配置效率与投资回报"] end ``` - **渠道与库存管理**：涉及全球 180 个市场的渠道伙伴与库存供需协同，优化其背后的经济账目是释放营运资金的关键。 - **高杠杆部门**：IR 与 M&A 等团队虽然编制小，但决策直接牵动数十亿美元市值的资本成本和重大交易回报率。 - **即刻可落地的替代项**：营销部门的视频及宣发素材生成已达到可用水平，能够立竿见影地削减外部制作费用。 --- ### `15:45` “AI 精神病”与资本控制：通过紧缩传统预算倒逼技术变革 节目话题转向了 AI 支出管理中的棘手现实：面对每月可能烧掉上百万美元 Token 费用的工程师或业务团队，CFO 该如何甄别真实的回报与无意义的浪费。主持人调侃道，业内极容易出现一种所谓的“AI 精神病”（AI psychosis）——即使用者完全沉浸在与 AI 交互的魔法感中，自我催眠认为自己极其高效，但实际上根本没有产出任何实质性成果。在一家年营收超过 800 亿美元、业务遍及 180 个市场且拥有数万名员工的全球 200 强企业中，CFO 根本无法自上而下微观管理每个人的 Token 配额。 郑弘孟分享了他在管理哲学上的务实做法：与其事无巨细地监控每个员工，不如从宏观预算层面倒逼变革。他的策略是首先评估现有 AI Agent 在各个业务领域内究竟能做到什么水平，一旦确认 AI 能够胜任，CFO 办公室就会主动“卡死或断供传统预算”（starve certain budgets）。他指出，如果企业继续保留旧流程的预算，员工一定会出于惯性延续旧有行为模式；但人类具有极强的适应能力和生存本能，当退路被切断、传统预算被拿掉时，业务部门为了生存自然会全力拥抱并挖掘 AI 的功能。 主持人对此提出了进一步探讨，认为这折射出当下科技巨头 CFO 之间一种鲜明的哲学分歧：一派倾向于像联想这样建立严格的预算约束，通过资源倒逼来筛选真正的创新与提效能力；另一派则以部分美国激进科技公司为代表，主张放任自流甚至鼓励内部“Token 刷量竞赛”（token-maxing），给足极高的开销上限让员工肆意探索，试图通过彻底放开缰绳来淘洗出最具生产力的超级员工。这种约束机制与放任试错之间的张力，正是当前 AI 基础设施狂潮中企业资本配置的核心博弈点。 ### `20:00` 联想的“AI十年”战略与资本配置的ROI哲学 在谈及科技公司如何在早期摸索最优发展路径时，联想集团首席财务官郑弘孟（Winston Cheng）指出，企业内部往往并存着两种截然不同的投入路径。当前整个市场正处于创新的剧烈爆发期，联想必须在技术创新上不遗余力地注入资本。董事长兼首席执行官杨元庆也将此定调为战略重心，集团内部将其确立为“AI十年”（AI Decade）的开局之年。面对碎片化的全球格局、复杂的数据主权以及供应链风险，联想凭借遍布全球的本土化制造网络，能够灵活适配各个区域的合规与可持续发展要求——例如在欧盟市场，联想供应的笔记本电脑100%采用可降解包装纸盒，机身内部大量运用可回收材料，并在元器件供应紧张的环境下大力推进二手翻新业务。 然而，在业务职能部门（如财务体系）内部，AI预算绝不能等同于“无上限开支”，必须秉持务实理性的投入产出比（ROI）思维。郑弘孟以财务规划与分析（FP&A）和会计流程为例解释道，联想目前向第三方服务商外包了部分后台职能，每年的开支相当可观。这类工作本质上是标准化的流程流转，长远看应通过AI全面自动化来实现降本；联想愿意继续外包，前提是合作伙伴也能同步创新并与联想共享降本红利。在规划、税务、资金、投资者关系等各项细分职能中，效率提升带来的财务回报往往立竿见影。正如郑弘孟早间与税务总监沟通时提到的，只要团队协同推进下一年度的税务优化，省下来的真金白银“足以为这二十多位团队成员支付一次愉快的团建旅行”，这种投入显然极具价值。此外，企业在引入外部AI工具时，对于未知的企业数据安全隐患与合规风险仍存有深刻担忧，这直接影响着AI在企业内部落地的广度与深度。 - **AI十年战略**：联想将当前阶段定义为集团“AI十年”的第一年，全面加大创新研发投入。 - **可持续制造标准**：在欧洲等严苛市场实现100%纸质可降解包装，积极提升机身可循环材料比例与设备翻新再利用。 - **职能端资本节约**：对后台外包流程进行AI自动化改造，拒绝盲目跟风，以严格的ROI考量各业务条线的开支。 ```mermaid flowchart TD accTitle: 联想AI战略下的资本分配逻辑 accDescr: 展示联想在核心创新业务与内部运营支持两条路径上的资本配置策略与差异。 subgraph Innovation["核心业务：创新扩张驱动"] direction TB A["战略定调：AI十年首年"] --> B["全球本土化制造与合规交付"] B --> C["绿色供应链与可持续产品标准"] end subgraph InternalOps["职能运营：严格ROI驱动"] direction TB D["外包流程审查（FP&A/会计/税务）"] --> E["自动化替代与效率优化"] E --> F["成本节约与数据安全风控"] end Innovation --- InternalOps ``` --- ### `23:05` “中间力量”的自我认知与硬件价值重估 针对美国资本市场上诸如SpaceX等企业凭借“无限全景市场空间（TAM）”吸引海量资本涌入的狂热现象，主持人抛出疑问：身处资本狂欢之外的企业该如何竞争？中美两国在AI路径上又有何本质差异？郑弘孟坦言这需要拆解为两个维度。他提及今年初参加达沃斯世界经济论坛时，被马克·卡尼（Mark Carney）关于“超级大国”（Superpowers）与“中间力量”（Middle Powers）的演讲深深触动。回到公司后，他便将这套概念引入管理委员会讨论，以此明确联想在整个科技生态位中的定位：“我们必须承认，联想绝不是超级大国，我们是一家中间力量。” 作为全球出货量第一的个人电脑制造商，联想拥有横跨PC、平板与智能手机的完整终端产品线，上承微软、谷歌等操作系统巨头以及英伟达、英特尔、AMD、高通等芯片供应商，同时正积极拥抱基于ARM架构的新型芯片与蓬勃发展的中国本土技术栈。郑弘孟指出，在近期股价上涨之前，资本市场长期低估了联想的生态枢纽价值——联想连接着产业链上下游超过2000家供应商，撬动逾2万亿美元的科技支出，并将产品高效分销至全球180个市场。回顾PC时代的利润分配格局，操作系统与芯片巨头市值纷纷迈过万亿美元门槛，而以联想为代表的设备制造厂商市值过去长年徘徊在200亿美元以下，近期才回升至300亿到400亿美元区间。郑弘孟强调，大规模的供应链聚合能力、极致的制造良率、遍布全球的服务响应以及全生命周期的售后与安全信任，其内在商业价值绝不应被持续低估，当前的AI硬件浪潮正是终端设备商重塑自身价值分配格局的关键窗口期。 - **产业定位分级**：认清自身作为科技价值链中“中间力量”的赋能者定位，保持务实谦逊。 - **网络辐射能力**：协同全球逾2,000家供应商，参与调动超2万亿美元产业资金池，覆盖180个国家和地区。 - **价值分配失衡**：底层芯片与系统软件厂商攫取数万亿美元市值，而设备集成商长期被给予估值折价，AI时代有望迎来硬件端溢价修复。 --- ### `26:45` 中国市场的“内卷”进化与服务器竞争的深层逻辑 对比中美两国的AI产业演进，郑弘孟分析称，芯片获取层面的客观限制倒逼中国AI企业必须在算力受限的环境下将算力效能与成本控制榨取到极致。业内近期广泛热议DeepSeek等中国模型的高效表现，其单Token生成成本据传甚至低至美国同行的1/50。从经济学上的达尔文演化论来看，唯有最残酷的自由竞争才能淬炼出最具生命力的市场主体。中国科技圈常说的“内卷”（Involution）虽然竞争异常惨烈，但“如果一家企业能在芯片供给受限、竞争极度残酷的中国市场杀出重围，那么它在世界任何地方都具备极强的生存能力”。 | 对比维度 | 美国 AI 生态 | 中国 AI 生态 | |:--- |:--- |:--- | | **算力与硬件资源** | 掌握顶尖高性能计算芯片，算力资源供给相对充裕 | 面临先进芯片获取限制，极度依赖算法优化与算力挖潜 | | **成本与研发重心** | 追逐极限模型参数与通用前沿能力，资本投入容忍度高 | 追求极致单位Token成本效益，专注推理优化与降本增效 | | **市场竞争机制** | 巨头高额资本开支（Capex）主导，注重生态全景叙事 | 经历极度白热化的“内卷”竞争，倒逼企业磨炼出韧性 | 当话题转向服务器领域的竞争格局——究竟是戴尔（Dell）、超微（Supermicro）等企业侧重的单机硬件性能更重要，还是如联想这般依赖准时制生产（JIT）与全球顶级零部件采购的供应链掌控力更关键？郑弘孟指出，这从来不是一道非此即彼的单选题，而是高度集成的全栈技术工程。在当下的AI基础设施建设周期中，客户已经不再是传统的采购方，而是深度绑定的长期战略“合作伙伴”，因为数据中心系统架构的重构与资金开支周期往往跨越数年之久，唯有兼具供应链交付弹性与硬核工程性能的整体集成方案，才能在这场算力竞赛中立足。 ### `30:00` 全球化供应链韧性与模块化数据中心交付能力 联想集团（Lenovo）首席财务官郑弘孟（Winston Cheng）指出，如今的头部客户并非带着单笔采购订单而来，而是带着长达数年的综合建设计划，这也要求联想必须以同样的长期战略态度对待上游供应商。在过去十余年间，联想持续在全球制造网络上深耕，目前已在全球运营着 30 座制造基地，在各大核心区域均拥有深厚的本地化生产储备。无论是在匈牙利工厂供应欧洲和美国市场，在墨西哥和美国本土工厂直供北美，还是立足亚洲深耕亚太市场，抑或在中国保留强大的生产基地，联想都展现出极高的全球化灵活调度能力。为了进一步开拓中东市场，联想正在沙特阿拉伯利雅得新建一座先进工厂，专门生产面向中东客户的服务器硬件。 郑弘孟强调，在当下的算力军备竞赛中，瓶颈早就不局限于单一元器件的采购配额，而在于“拿到芯片后是否具备系统组装能力”、“是否拥有充足且可靠的工业制造产能”，以及“产线末端是否有能力对极高复杂度的算力系统进行严苛测试”。联想为客户提供从供应、制造到测试的端到端（end-to-end）全生命周期交付，构筑了高安全壁垒的一站式服务体系。更具差异化优势的是联想的方案服务业务集团（SSG），它颠覆了传统 OEM 仅能交付机架式服务器的局限，能够直接承接整座数据中心的交钥匙建设工程。无论客户是超大规模云厂商（Hyperscalers），还是仅持有特定土地与电力指标的跨界开发商，联想都能通过其成熟的模块化（modular）数据中心方案，在短短 9 个月内完成整座数据中心的搭建；若现场已有基础配套设施，交付周期甚至可以压缩至惊人的 6 个月，极大缩短了客户的投产周期并加速算力资产变现。 ```mermaid flowchart LR accTitle: 联想端到端AI算力交付与基础设施布局 accDescr: 展示联想从全球供应链网络到模块化数据中心与液冷算力的交付体系 subgraph GlobalFootprint ["全球制造网络布局"] HU["匈牙利工厂<br/>(欧洲/美国市场)"] MX["墨西哥/美国本土工厂<br/>(北美市场)"] ASIA["亚洲及中国工厂<br/>(亚太与全球)"] RY["沙特利雅得新厂<br/>(中东服务器)"] end subgraph EndToEnd ["SSG 方案服务与工程交付"] MOD["模块化数据中心方案<br/>(6-9个月快速交付)"] LC["11,000 机架级液冷温控能力"] GPU["高密度 GPU 算力集群集成与测试"] end GlobalFootprint --> EndToEnd ``` - **全球制造布局**：在世界主要市场布局 30 座工厂，其中匈牙利服务欧美、墨西哥服务北美、利雅得新厂深耕中东，中国工厂持续保障本土与全球供给。 - **超快交付周期**：模块化数据中心建设通常仅需 9 个月，基础条件完备时可缩短至 6 个月上线。 - **先进散热储备**：具备超过 11,000 个机架的液冷（Liquid Cooling）部署能力，为高能耗高密度 GPU 计算集群提供全方位温控支持。 --- ### `35:36` 走出 90 年代 PC 同质化困局：AI 基础设施的真正护城河 节目主持人乔·维森塔尔（Joe Weisenthal）和特雷西·阿洛威（Tracy Alloway）探讨了当下服务器与算力市场与 20 世纪 90 年代个人电脑（PC）黄金时期的鲜明对比。乔回忆道，当年的 PC 市场几乎演变成了一场残酷的同质化自我绞杀，戴尔（Dell）、康柏（Compaq）、惠普（Hewlett-Packard）等品牌的产品本质上都是千篇一律的灰色铁盒，消费者走进百思买（Best Buy）门店甚至完全分不清它们的区别，必须依赖导购推销。ThinkPad 笔记本键盘中央标志性的“小红点”（TrackPoint）之所以至今仍被人们津津乐道，恰恰是因为当时绝大多数硬件厂商根本无法在产品本身建立起任何实质性的辨识度。 但郑弘孟的阐述表明，当前的 AI 数据中心与企业级基础设施市场展现出了截然不同的商业逻辑。在这场浪潮中，供应商不再只是货架上待价而沽的组装商，而是与客户深度绑定的战略工程伙伴。面对极其庞杂的算力互联、严苛的散热标准以及千变万化的软件栈调优，单纯交付硬件已失去竞争力，能够提供从芯片封装、机架制造、液冷工程到全栈运维的一体化解决方案，构成了无法被轻易替代的核心壁垒。 --- ### `36:51` 全产业链供需挤压：从存储芯片到电网电力的多重瓶颈 特雷西将话题引向现实中最严峻的供应链制约，追问在 GPU、CPU、存储芯片以及近期美国尤为紧缺的变压器等核心环节中，采购到底有多艰难。郑弘孟直言，全球算力需求极其旺盛，在香港及全球资本市场，大规模的 IPO 与惊人规模的再融资都在密集涌向 AI 基础设施建设。这种由前所未有的海量需求所驱动的元器件短缺态势，至少还将持续两到三年。尽管半导体产业链已在积极追加资本开支，但芯片制造具有硬性物理周期，尤其是尖端存储芯片，建设一座全新晶圆厂通常需要两至三年才能真正实现产能爬坡上线。 郑弘孟指出，硬件供应链之外的物理资源瓶颈同样严峻，包括可用土地、大型变压器以及最根本的“并网电量供应”。正是因为很多传统核心枢纽地区面临电网超载与电力短缺，联想才深度布局中东，与沙特政府紧密合作。沙特不仅拥有低成本、充沛且可持续的太阳能资源，还具备充足的本地绿色能源供应能力，能够高效承载大负荷 AI 基础设施的持续运转。在全球发展 AI 算力，绝不能盲目执着于在电力早已枯竭的单一区域强行建仓，而必须在全球范围内进行能源与算力的协同资源配置。 --- ### `38:56` 暴增的订单簿背后：牛鞭效应与重复下单隐忧 乔随后抛出了一个充满警惕意味的宏观问题：在经历过新冠疫情期间全球供应链大中断的洗礼后，大家对“牛鞭效应”（Bullwhip Effect）的破坏力记忆犹新。不论是联想，还是此前做客节目的 NeoCloud 代表企业 CoreWeave 联合创始人，行业参与者都在频繁提及跨越数年的超长订单簿（multi-year order books）与积压订单。 乔提出质疑：鉴于当前市场上对算力与关键零部件短缺的普遍恐慌，各大下游厂商是否正在出于避险心理进行大面积的双重甚至三重重复下单（double and triple ordering）？如果这种抢跑与恐慌性囤货现象普遍存在，那么眼下看似无穷无尽的算力需求与漫长的订单队列，其真实的可持续性究竟有多高？这是否意味着当前的繁荣景象中潜藏着严重的产能过剩与虚假繁荣风险？ ### `40:00` 订单泡沫真伪与AI驱动的指数级算力扩张 面对市场上关于GPU和算力设备“超额订购”（over-ordering）或重复下单的疑虑，联想集团CFO郑 Winston（Winston Cheng）从财务与订单周期的本质进行了厘清。他指出，算力基础设施的最终采购方都是极其成熟的大型科技企业。在销售管道（pipeline）阶段确实可能存在重复排队以测试不同供应商交货能力的情况，但在正式的未交付订单（backlog）层面很难出现双重签约，因为标准的待交付订单是具有法律约束力的正式协议，企业不会重复签署；而到了收入确认（revenue recognition）阶段，设备更是已经实际出货并等待回收应收账款。 Cheng 强调，目前在市场上看到的对于数据中心机房空间、服务器及电力设备的投入，全部是多年期（multi-year）的重资产战略承诺。科技企业从互联网时代和云服务商的发展中深刻吸取了经验，清晰地预见到数据体量将产生几何级爆发。生成式AI不仅要消化现存的所有历史数据，还会持续衍生出海量的新数据，而对这些新数据的激活与赋能，必然需要成倍甚至指数级增长的底层算力支撑。 - 订单辨析：销售潜在需求池（Pipeline）可能存在多头下注，但形成正式背书的待交付积压（Backlog）与实际出货收入（Revenue）极具刚性。 - 基础设施周期：当前数据中心建设并非短期投机，而是锁定长周期电力、空间与算力资源的跨年度战略投资。 ```mermaid flowchart TD accTitle: 算力基础设施订单转化与数据增长循环 accDescr: 展示从潜在销售管道到收入确认的订单流程，以及生成数据对新增算力的反馈循环。 subgraph OrderLifecycle["订单转化生命周期"] P["销售需求管道 (Pipeline)<br/>可能存在多方排队与重叠"] --> B["签约待交付订单 (Backlog)<br/>具有法律约束力的确定协议"] B --> R["出货与收入确认 (Revenue)<br/>实际交付与应收款项确认"] end subgraph DataFlywheel["数据飞轮与算力需求"] D1["吸收现有存量数据"] --> G["AI 生成指数级增量数据"] G --> E["数据激活与应用赋能"] E --> C["追加多年度基础设施算力投入"] C -.-> P end ``` ### `41:30` “SaaS末日论”与企业内部IT的数据中间层瓶颈 谈及市场上热议的“SaaS末日论”（Saspocalypse）——即大模型将让软件开发门槛骤降，任何企业都能轻易复刻类似微軟 Word 或 Excel 的工具，进而可能侵蚀软件利润——Cheng 从制造与硬件巨头的独特视角给出了回应。联想本身并非纯软件或 SaaS 公司，因此大模型代码生成能力的飞跃对联想反而是利好。它显著降低了开发门槛，让原本非软件核心导向的联想内部 IT 团队或工程人员能够以极低的成本自行搭建业务所需的特定应用。 然而，这并不意味着所有企业应用都能被随意替代。Cheng 强调，企业级系统的真正壁垒在于“核心数据层”（data layers），这是大中型企业既无法绕开、也不愿放弃的资产。当前所有企业拥抱 AI 时面临的最大瓶颈，并非缺少生成代码的模型，而是缺乏能够打破数据孤岛、整合多源内部数据并高效对接 AI 的“中间层”（middle layer）。要解决这一卡点，单靠纯软件产品远远不够，还需要大量的架构咨询与落地服务配合，整个行业距离完全打通这一流程仍有漫长旅程。 - 研发门槛变化：AI 编程工具大幅赋能硬件与制造企业的内部 IT 部门，自主构建轻量化定制应用的成本被大幅压低。 - 核心瓶颈转移：AI 落地的障碍不在于应用层界面，而在于企业内部异构数据的清洗、汇聚以及与底层大模型的打通机制。 ### `43:03` 算力被AI吞噬下，联想拯救者（Legion）与ThinkPad的护城河 主持人提到游戏玩家群体的普遍抱怨：早先消费级 GPU 被以太坊等加密货币挖矿抢夺，如今最高边际利润的高性能 GPU 又被人工智能数据中心席卷一空，导致游戏硬件面临原料被分流和定价高昂的局面。面对这一现象，Cheng 表达了联想捍卫核心硬件业务的决心。他特别提到了联想旗下两大王牌品牌：源自 IBM 日本大和实验室（Yamato Labs）、凝聚数十年工业设计精髓的 ThinkPad 商用系列，以及在全球游戏 PC 市场出货量居首的“拯救者”（Legion）系列。 Cheng 坦言，拯救者系列凭借极高的质价比在核心玩家群体中建立了极其忠诚的粉丝基本盘，联想不仅要守住现有的全球第一市场份额，更要寻求进一步扩张。面对 AI 基础设施需求与个人设备端对芯片资源的争夺，联想的策略是凭借自身庞大的综合采购规模，在终端设备部门（PC/移动设备）与基础设施方案部门（AI 服务器与数据中心）之间动态平衡，通过获取更多上游供应链的战略配额来满足不同产品线的供给。 - 经典传承：ThinkPad 保留了源自日本大和实验室的核心研发脉络，维持高端商用笔记本的标杆地位。 - 游戏基本盘：拯救者（Legion）在全球游戏 PC 市场位居出货量第一，凭借极致性能释放与性价比维系用户黏性。 - 资源调度：面对芯片厂商将晶圆产能向数据中心 AI 芯片倾斜的现状，联想依赖全球顶级采购量争取更多跨部门的元件配额。 ### `45:15` 不押注单一巨头：以“AI编排器”打造中立端侧生态 针对联想是否会与某一家大模型巨头签署排他性独家协议的提问，Cheng 明确表示联想坚持走开放与中立路线。当前整个大模型领域处于高速迭代周期，每隔数月无论是 OpenAI、Anthropic、谷歌 Gemini，还是中国本土的一系列创新模型，都会推出颠覆性的技术升级。市场未来是否会演变为“赢家通吃”尚无定论，因此联想的核心定位非常明确：“我们不去赌最终谁能赢，我们的职责是提供底层的 AI 算力”。 联想在 AI PC 和终端设备上的核心架构创新是“AI编排器”（AI Orchestrator）。通过这层中立的端侧调度架构，用户无需在设备上自行下载、切换多种不同的大模型应用，而是由系统底层根据任务类型，自动将指令路由分发给最优的本地模型或云端模型。这种调度机制不仅极大提升了用户交互体验，还能在端侧内存占用和硬件算力消耗上实现深度优化。 - 战略中立：不锁定单一模型供应商，对全球顶尖模型（如 OpenAI、Gemini）与中国本土生态均保持全面兼容与接入。 - 核心功能：端侧 AI 编排器实现智能化跨模型路由调度，降低终端能耗并最大化本地硬件资源的执行效率。 ### `47:11` 云服务商自研芯片：性能协同还是对英伟达的战略“防身”？ 随着微软（如推出 Maia 芯片）等超大规模云服务商（hyperscalers）纷纷加速定制自研芯片，并宣称其私有模型在自研硅片上运行效率最高，主持人探究这究竟是算法与硬件结合带来的真实性能红利，还是巨头为了摆脱对英伟达绝对依赖而设立的战略“解绑缓冲垫”。Cheng 认为，英伟达目前是毫无争议的技术领跑者与 AI 浪潮核心使能者，其营收爆发正是市场对其领先地位的最佳认可。 然而，自研芯片趋势的出现是由双重动力驱动的： 1. **掌控技术栈与成本防线**：超大型云厂商具备雄厚的资本储备与底层架构能力，面对英伟达高昂的硬件溢价，自研芯片能从长期大幅拉平总拥有成本（TCO），并让巨头牢牢掌控自身全链路的技术架构。 2. **场景定制与长远战略弹性**：特定芯片针对专有模型的微架构优化确实能压榨出特定算力红利，更重要的是为巨头未来的业务演进提供了不被单一硬件路线图锁死的战略回旋空间。 ### `49:15` 财年交替之际的突发变局：全球供应链与地缘关税冲击 在对话深入到供应链多元化布局时，主持人提到若在一两年前录制本期节目，地缘政治、关税壁垒和贸易保护主义必然是压倒性的核心议题。联想在全球范围内构建了极其庞大而精密的供应链与制造网络，面对贸易保护政策的死灰复燃，企业如何抗风险成为考验管理层的关键课题。 Cheng 在此分享了一段戏剧性的个人就任经历：联想的财年始于每年的 4 月 1 日，而他正是在 4 月 1 日正式接棒出任集团首席财务官（CFO）。然而就在他上任的第二天（4 月 2 日），公司就迎来了其历史上最为重大的地缘与关税突发事件考验，管理层被迫在履新伊始就全面投入到供应链应变与全球资产调度的硬仗之中。 ### `50:00` 关税豁免与科技普惠：个人电脑作为数字时代的生存刚需 在全球关税政策剧烈波动的背景下，联想的全球化足迹面临着直接冲击。全球有200多个国家和地区推行了相关关税政策，而联想在其中的180多个市场均有业务布局。然而，政策制定者最终普遍选择对个人电脑（PC）产品予以关税豁免。这一决策背后的核心逻辑在于，无论是每天装在口袋里的移动设备，还是摆在桌面与家庭中的计算设备，PC与智能终端早已成为现代社会生产力、教育及娱乐不可或缺的底层基础设施。在发展中国家和新兴经济体中，这种倾向尤为明显——各国政府极力避免其民众在数字化浪潮中被抛下，而低成本的计算设备正是接入数字世界的起点与门槛。因此，消除关税壁垒、以尽可能低的价格提供科技硬件，本质上是一项关乎数字公平的生存刚需。 - 联想在全球超过180个国家和地区开展业务，庞大的供应链网络使其对全球贸易壁垒极为敏感。 - 个人电脑之所以能在广泛的关税政策中获得豁免，是因为监管层将其定性为连接数字信息的最底层生产力工具。 - 发展中经济体对硬件准入门槛尤为敏感，降低硬件进口成本直接决定了大众能否享受到数字经济的发展红利。 --- ### `51:24` 算力路由与咨询迷局：AI 落地时代连顾问也在“交学费” 关于模型成本的讨论正逐渐深入到企业的实际运营层面。尽管业内公认 DeepSeek 在单 Token 成本上远低于 OpenAI 或 Anthropic 的前沿旗舰模型，但美国顶级实验室的旗舰大模型在综合能力上依然无可争议地领先数月。面对这种差距，企业 CFO 面临的新命题是：如何避免“杀鸡用牛刀”，即如何构建精细化的查询路由机制（Query Routing），让常规任务流向低成本模型，仅将极端复杂的推理交给昂贵的前沿模型。 谈及企业的 AI 学习曲线，Winston Cheng 分享了当天早晨的一段生动经历。为了梳理这套复杂的 AI 部署方案，联想专门聘请了全球顶尖的咨询机构，当天上午刚与对方的全球团队进行了一个半小时的访谈。然而他敏锐地察觉到，在这场 AI 淘金热中，“所有人都是新手”（everyone is new to this game）——即便是收取高昂咨询服务费的专业顾问，也同样处于边干边学的摸索阶段。面对主持人调侃“那你付钱给他们买什么”，Winston 幽默地回避对同行做评价，但强调这一现状恰恰说明：企业内部不能迷信外部专家，更不能陷入层层审批的官僚主义红毯（red tape）。在瞬息万变的 AI 时代，取胜的关键在于清醒的判断力和极快的执行速度，合理规划 Token 交互预算，把钱花在刀刃上。 ```mermaid flowchart TD accTitle: 复杂请求与常规任务的模型路由分流机制 accDescr: 展示企业如何根据任务复杂度将请求路由至低成本模型或前沿旗舰模型，以平衡性能与Token成本。 UserReq["用户或系统输入查询 (Query)"] --> Router{"智能路由判断引擎<br/>(任务复杂度评估)"} subgraph CostEffective["低成本高效层"] Router -->|"常规查询 / 文本总结 / 代码补全"| CheapModel["高效开源/性价比模型<br/>(如 DeepSeek 等)"] CheapModel --> OutCheap["以极低 Token 成本快速交付"] end subgraph FrontierLayer["前沿旗舰层"] Router -->|"高阶推理 / 复杂架构 / 极限任务"| FlagshipModel["前沿旗舰大模型<br/>(OpenAI / Anthropic 等)"] FlagshipModel --> OutFlag["高质量输出 (成本较高)"] end OutCheap --> FinalResult["终端应用集成与交付"] OutFlag --> FinalResult ``` - 企业对外部顶级咨询公司的访谈揭示：整个行业在 AI 落地和 Token 成本优化上均处于摸索初期，顾问与客户在共同经历学习曲线。 - 决定企业 AI 竞争力的核心不是咨询报告，而是消除内部官僚流程、以最快速度试错与部署模型路由。 - 现场主持人还打趣称，未来在社交场合人们不仅会报出名字和国籍，可能还会宣称自己用哪款基础模型，甚至在酒吧闲聊时调侃“我要找一个每月 Token 消耗量极高的男人”。 --- ### `54:12` 资本回报与增长投入的平衡：创纪录分红与利润率扩张 在科技行业陷入前所未有的 AI 资本支出（Capex）狂潮之际，投资者对资金消耗的合理性提出审视，管理层也面临巨大的资本分配压力。当主持人问及如何在无底洞式的 AI 军备竞赛中平衡股东回报时，Winston 指出，在刚刚于 3 月 31 日结束的上一个财年中，联想派发了公司历史上最高额度的现金股息。尽管联想股价表现强劲，管理层依然坚持将经营成果即时回馈给股东。 然而，持续的增长同样需要海量资本作为燃料。联想当下的资本配置哲学是在“提供即时现金分红”与“支撑底层业务增长以推动长期股价升值”之间寻找精密平衡。Winston 承认，为了驱动这种增长并伴随长期的利润率扩张，短期内不可避免会在个别季度出现增长投入与利润率表现不完全匹配的波动；但从长期来看，联想的核心战略目标非常明确，即保持业务规模扩张的同时实现利润率的加速提升。 - 在截至 3 月 31 日的最近一个财年中，联想支付了公司有史以来最高金额的股息分红。 - CFO 的资本分配天平两端，分别是股东对即时流动性的诉求，以及通过 AI 基础设施与端侧设备投入拉动长期市值的资本升值需求。 - 战略规划允许季度维度的利润与增长短暂错配，但长期基准是推动收入增长伴随利润率加速扩张。 --- ### `55:18` 亚洲数据中心版图：“北弗吉尼亚”何在与地缘政策导致的算力通胀 在美国，北弗吉尼亚州长期被公认为数据中心之都，但随着得克萨斯州场外离网发电项目的兴起以及多地反数据中心政治情绪（anti-data center politics）的发酵，美国的算力版图正在重构。当被问及亚洲是否存在类似“北弗吉尼亚”的算力枢纽时，Winston 梳理了亚洲复杂的地缘与能源格局。亚洲在产业建设层面整体表现出更强的灵活性，但目前面临结构性割裂：从规模和基础设施完备度来看，中国大陆无疑拥有最强大的电力供应、电网规模和供应链制造能力；然而受限于监管与地缘出口限制，全球大量企业无法在当地部署前沿算力，导致中国大陆许多优质的电力与数据中心资源处于闲置状态。这种扭曲迫使全球资本流向成本更高的法域，进而加剧了全球 AI 算力供应链的通胀效应。 在中国大陆以外的亚太区域，多处算力热点正在迅速崛起： - **东南亚（马来西亚与印度尼西亚）**：拥有极佳的土地与电力资源承载力，正成为国际科技巨头争相建设超大规模数据中心的首选自然承接地。 - **香港与新加坡**：受限于极小的土地面积，两地通过高度创新的建筑结构来实现算力密集化，例如香港正在发展建在多层高楼结构中的高密度数据中心，并采用多元化的混合供电方案。 - **日本**：虽然在法治与监管信誉上极具优势，但在成本控制以及各类许可审批的落地周期上相对缓慢。 数据中心建设不仅关乎技术，更是一项涉及土地出让、高压电力接入以及专用设备进口许可的综合系统工程，极其依赖与当地政府的深度信任与本地执行效率。节目最后，主持人 Tracy 调侃称香港“多的是便宜地皮”，随即便自嘲这是一句反讽——香港固然有未开发的土地，但由于严苛的规划管制，“有地也根本无法开工建设”。 ### `01:00:00` 主持人复盘：AI 的物理基础设施与非超大规模企业的资本考量 在与联想首席财务官郑伟（Winston Cheng）的对话结束后，主持人 Joe Weisenthal 与 Tracy Alloway 对整场采访进行了复盘。针对节目中频繁聚焦人工智能话题的现象，Joe 解释了背后的两大原因：首先，AI 几乎是所有人有生之年见证的最重大事件，每个人都在试图理解它，有太多认知需要更新；其次，这个话题天然契合《Odd Lots》的关注领域——尽管 AI 表面上是代码和算法，但它本质上具有极强的“实体性”（physical），与全球硬件供应链紧密相连。联想不仅生产服务器，还具备构建完整数据中心的能力，横跨在芯片厂商、云厂商以及终端客户之间，这种兼具上下游双重视角的定位使其成为探讨 AI 浪潮的绝佳标杆。 Tracy 对此深表赞同，指出人们在面对 AI 时往往存在一种奇特的割裂感：一面是毫无形体、没有面孔的虚拟人机交互界面（"bodiless, faceless AI interface"），另一面却是背后极其庞大、沉重的物理基础设施支持。谈到联想作为硬件提供商与超大规模云厂商（Hyperscalers）的区别时，Joe 指出，联想并非像谷歌、微软或 OpenAI 那样动辄宣布明年要砸数百亿美元自建数据中心的企业。面对巨额资本开支（CapEx）狂潮，联想不仅要考虑设备投资，还必须在日常运营成本（OpEx）的管控与转化上精打细算。正如受访者所言，无论是在 CFO 同行圈子还是管理咨询领域，关于如何在 AI 时代平衡 OpEx 和商业回报的讨论还处于摸索的第一天（"day one of figuring it out"），所有规则与模式都尚未成型。 ```mermaid flowchart TD accTitle: AI投资的双重视角差异 accDescr: 对比超大规模云服务商与硬件集成商在资本与运营层面的关注焦点 subgraph Hyperscalers["超大规模服务商 (Hyperscalers: Google/OpenAI等)"] H1["重度资本支出 (CapEx 驱动)"] H2["动辄百亿美元自建超大型数据中心"] end subgraph LenovoView["硬件与解决方案提供商 (如 Lenovo)"] L1["兼顾资本支出与运营支出 (CapEx + OpEx)"] L2["构建端到端实体基础设施与服务器"] L3["在严苛成本控制下探索商业化落地"] end Hyperscalers -->|算力与基础设施需求| LenovoView ``` - **物理基础设施的张力**：AI 展现出虚拟界面与重型硬件设备（供电、冷却、机架、供应链网络）之间的剧烈对比。 - **CapEx 与 OpEx 的再平衡**：超大规模云厂商主导了聚光灯下的巨额 CapEx 竞赛，但广大科技硬件制造与集成商更需在 OpEx 的精准把控中求生存。 ### `01:02:30` 市场“内卷”效应、全球竞争力与节目结语 Tracy 随后提到了访谈中关于中美 AI 发展差异以及“内卷”（involution）的精辟讨论。在中国市场，激烈的内部竞争达到白热化程度，倒逼所有参与者不计代价地压缩成本。这种残酷的市场环境产生了一个非常耐人寻味的客观结果：由于在中国本土被极限压缩了成本曲线，某些企业即便在国内算不上最头部的顶尖玩家，一旦将其产品放到全球市场维度去审视，其成本控制与性价比优势依然具有极其凶悍的全球竞争力（"even if you're not that competitive in China, you would still be competitive on a global scale"）。 复盘最后，主持人以轻松的语调结束了本期播客录制，并向听众致谢。两位主持人提醒听众可以通过社交平台关注他们本人及播客制作团队（包括制作人 Carmen Rodriguez、Dashiell Bennett、Kael Brooks 和 Kevin Lozano），查阅 Bloomberg 的《Odd Lots》每日通讯专栏，或加入官方 Discord 社区展开全天候交流探讨。 - **中国内卷的技术溢出效应**：极度残酷的本土竞争打磨出了极致的降本能力，形成了外溢到全球层面的结构性成本优势。 - **播客互动与社群渠道**：节目组鼓励听众在 Discord 社区参与讨论，并可通过 Bloomberg 订阅渠道收听无广告版本。 ## Notable quotes > `07:30` 作为一名 CFO，你的职责是去分配资本（Allocate Capital），而不是去卡死资本、限制资本（Constrain Capital）。 — Winston Cheng > `23:05` 我们必须承认，联想绝不是超级大国，我们是一家中间力量。 — Winston Cheng > `26:45` 如果一家企业能在芯片供给受限、竞争极度残酷的中国市场杀出重围，那么它在世界任何地方都具备极强的生存能力。 — Winston Cheng > `45:15` 我们不去赌最终谁能赢，我们的职责是提供底层的 AI 算力。 — Winston Cheng > `51:24` 所有人都是新手（everyone is new to this game） — Winston Cheng > `01:00:00` bodiless, faceless AI interface — Tracy Alloway > `01:00:00` day one of figuring it out — Joe Weisenthal > `01:02:30` even if you're not that competitive in China, you would still be competitive on a global scale — Tracy Alloway ## Mentioned 联想 (Lenovo), Winston Cheng (郑弘孟), Joe Weisenthal, Tracy Alloway, IBM, 百度 (Baidu), Nvidia (英伟达), ThinkPad, TrackPoint, 杨元庆, Mark Carney (马克·卡尼), 微软 (Microsoft), 谷歌 (Google), 英特尔 (Intel), AMD, 高通 (Qualcomm), DeepSeek, 戴尔 (Dell), 超微 (Supermicro), 康柏 (Compaq), 惠普 (Hewlett-Packard), CoreWeave, 拯救者 (Legion), OpenAI, Anthropic, Gemini, Maia, 大和实验室 (Yamato Labs), 方案服务业务集团 (SSG) ## Video description <p data-renderer-start-pos="122">We know that companies around the world are investing heavily in AI. So intense is the race to win the AI battle, that it feels like there's almost no upward limit on how much you could spend on it. So how are CFOs thinking about capex in the AI age? In this episode we speak with Winston Cheng, CFO of Chinese-founded multinational tech firm Lenovo. Lenovo is known for its personal computers, especially its Thinkpad line of laptops, but they are making a push to move beyond its role as one of the leaders in personal computing, integrating AI agents into their devices and investing in building out an &ldquo;AI Cloud&rdquo; infrastructure alongside Nvidia. We talk to Cheng about how Lenovo's allocating capital during one of the biggest capex booms in history. We also discuss involution and market competition in China, and how Lenovo's been adapting its supply chain to tariffs.</p> <p data-renderer-start-pos="998">Read more:<br><a href="https://www.bloomberg.com/news/articles/2026-06-25/ai-demand-begins-to-justify-massive-cost-of-data-center-buildout?utm_medium=referral&utm_source=podcast&utm_campaign=odd_lots&utm_content=article" data-renderer-mark="true" data-is-router-link="false" data-testid="link-with-safety">AI Sales Start to Justify Data-Center Spending Boom, Report Says</a><br><a href="https://www.bloomberg.com/news/articles/2026-06-24/anthropic-accuses-alibaba-of-illicitly-accessing-its-ai-models?utm_medium=referral&utm_source=podcast&utm_campaign=odd_lots&utm_content=article" data-renderer-mark="true" data-is-router-link="false" data-testid="link-with-safety">Anthropic Accuses Alibaba of &lsquo;Illicitly&rsquo; Accessing AI Models</a></p> <p data-renderer-start-pos="1136">Only <a href="http://bloomberg.com/" data-testid="inline-card-resolved-view">Bloomberg - Business News, Stock Markets, Finance, Breaking & World News</a> subscribers can get the Odd Lots newsletter in their inbox each week, plus unlimited access to the site and app. Subscribe at&nbsp; <a href="https://www.bloomberg.com/subscriptions/oddlots?in_source=oddlotspodcast" data-renderer-mark="true" data-is-router-link="false" data-testid="link-with-safety">bloomberg.com/subscriptions/oddlots</a></p> <p data-renderer-start-pos="1308"><a href="http://bloomberg.com/subscriptions/oddlots" data-renderer-mark="true" data-is-router-link="false" data-testid="link-with-safety">Subscribe to the Odd Lots Newsletter</a><br><strong data-renderer-mark="true">Join the conversation:</strong> <a href="https://discord.gg/oddlots" data-renderer-mark="true" data-is-router-link="false" data-testid="link-with-safety">discord.gg/oddlots</a></p><p>See <a href="https://omnystudio.com/listener">omnystudio.com/listener</a> for privacy information.</p> [Source](https://podtrac.com/pts/redirect.mp3/tracking.swap.fm/track/UVBrz8bN8aM2Xe47PEPu/traffic.omny.fm/d/clips/e73c998e-6e60-432f-8610-ae210140c5b1/8a94442e-5a74-4fa2-8b8d-ae27003a8d6b/b1fa9369-e153-4071-8b62-b47400f4a30f/audio.mp3?in_playlist=982f5071-765c-403d-969d-ae27003a8d83)

[7:30] Winston Cheng: 作为一名 CFO，你的职责是去分配资本（Allocate Capital），而不是去卡死资本、限制资本（Constrain Capital）。 [23:05] Winston Cheng: 我们必须承认，联想绝不是超级大国，我们是一家中间力量。 [26:45] Winston Cheng: 如果一家企业能在芯片供给受限、竞争极度残酷的中国市场杀出重围，那么它在世界任何地方都具备极强的生存能力。 [45:15] Winston Cheng: 我们不去赌最终谁能赢，我们的职责是提供底层的 AI 算力。 [51:24] Winston Cheng: 所有人都是新手（everyone is new to this game） [60:00] Tracy Alloway: bodiless, faceless AI interface [60:00] Joe Weisenthal: day one of figuring it out [62:30] Tracy Alloway: even if you're not that competitive in China, you would still be competitive on a global scale

[1:00] Tracy Alloway: word currency [5:00] Winston Cheng: pocket-to-cloud [5:00] Winston Cheng: unbudgeted spending of that scale is a major concern [15:20] Winston Cheng: starve [20:00] Winston Cheng: AI decade [23:40] Winston Cheng: middle power [23:40] Winston Cheng: involution [23:40] Winston Cheng: 1/50th the cost of U.S. counterparts [28:40] Winston Cheng: 11,000-rack liquid cooling capability [34:00] Joe Weisenthal: bullwhip effect [40:00] Joe Weisenthal: SaaS-pocalypse [45:00] Winston Cheng: orchestrator [60:00] Joe Weisenthal: day one

[0:00] Whatever your goal, trade show giveaways, client gifts, or team gear, 4imprint has the promo products to match. With thousands of options, from apparel and drinkware to tech and totes, it's easy to find the right fit for your brand and budget, with standout choices at every price point. And with their 360-degree guarantee, you can be 4imprint certain your order will show up just right, right on time. Explore more at 4imprint.com. 4imprint. 4certain. So there's a lot of noise about AI. But time's too tight for more promises. So let's talk about results. At IBM, we work with our employees to integrate technology right into the systems they need. Now, a global workforce of 300,000 can use AI to fill their HR questions, resolving 94% of common questions. Not noise. Proof of how we can help companies get smarter by putting AI where it actually pays off. Deep in the work that moves the business. Let's create smarter business. IBM. Health care doesn always work great If you ever waited on a refill or couldn schedule an appointment you get it That the kind of stuff Optum is changing They using data and technology to integrate patient care pharmacy and everything else So health care is connected not complicated What that look like Cheaper prescriptions that are easier to get and care that looks at the whole person how you need it Optum is helping make healthcare work as one for everyone. Learn more at business.optum.com. Bloomberg Audio Studios. Podcasts. Radio. News. Hello and welcome to another episode of the All Thoughts Podcast. I'm Tracy Allaway. And I'm Joe Weisenthal. So Joe, we are here in Hong Kong still. And we're learning a lot of different things, a lot of interesting things. One of the most interesting things we heard actually came yesterday from the Baidu CFO, where we were just talking casually ahead of our interview. But he was saying that the word token has now officially been added to the Chinese Mandarin dictionary and that the characters that actually make up the Chinese word for token are something like word currency. I think that's so fascinating. You know I fascinated by the etymology of the word token specifically So I was thrilled to hear that But it is you know it like we know the word token in the monetary context We know Chuck E Cheese tokens, crypto tokens. But we also know that since the middle of the 20th century, linguists have been using token to describe more or less a word. And then obviously with LLMs, we talk about these linguistic tokens a lot. So to see that in Chinese, the formal term is a merger of these two concepts is I find a very intellectually satisfying thing to learn about. Bringing up Chuck E. Cheese tokens is really a way to make sure everyone knows that you're a millennial. Yeah, that's right. That's right. Arcade. I should just say arcade token. That's right. Yeah. OK. And I think what's really interesting about word currency itself is it implies that it's connected to spending. Right. Of course. And so when you think about the big companies that are spending all this money on tokens and I guess the AI infrastructure build out more broadly, CFOs become really important, right? Totally. So we, you know, the headlines by and large are about CapEx, right? And that's going to continue to be because of the data center builder. We're going to talk about that. But a lot of it is also going to be OPEX and figuring out how to within a company allocating token permissions and caps and so forth And I doubt anyone has figured out the final answer But if two different people can get different value out of using AI models, then there is no way that it makes sense for them to have the same token budgets. By the way, Tracy, can I ask you a personal question that I've never asked you before? Oh, OK. Outside of the work context, are you a Mac or a PC person? Oh, I only have work computers at the moment. Okay. All right. Before that, but don't touch the question. Before that, definitely PC. Okay, good. Me too. Yeah. And in fact. I mean, it's not a choice, is it? Like, sorry, I really don't like Macs. I don't either. And in particular, I have long been a fan of what used to be the IBM ThinkPad laptop with the famous red button, which is now owned by Lenovo. That's right. You've talked about that computer before. I can say that this is categorically true. Joe likes that computer. But what's interesting about Lenovo is like, OK, it's famous for the computer with a little red button in the middle, but it's now making an AI play. I mean, everyone's making an AI play, but it's doing it from [5:00] a different perspective. So AI integrating into the actual computer, the hardware. But it's also doing cloud, right? So this is a really good opportunity to, I guess, take the temperature on AI, the AI build out, the AI spend from a bunch of different perspectives. So we do, in fact, have the perfect guest. We're going to be speaking with Winston Chang. He is the CFO of Lenovo. So Winston, thanks so much for coming on OddLots. Hey, thank you, Tracy. And thank you, Joe. Why don't you go ahead and describe what Lenovo is and what the AI play is and how it actually fits into the existing business? Yeah, Lenovo is a global AI infrastructure company that provides pocket-to-cloud AI infrastructure for the consumer and the enterprise. I think that's really in a nutshell. And so today we're able to provide this to hyperscalers, which are doing a lot of the spending are driven by training demand today. And then given our IBM x86 heritage, which we also acquired the server from IBM actually So we actually are very strong in CPU compute as well given that IBM was a dominant player in the x86 architecture So from that perspective we well positioned for inferencing needs of the enterprise and also the hyperscalers in terms of in the cloud as well. So I think from the perspective of then, you talked about tokens. And I think from a token perspective, really today, people are in the early stage of how much I'm really paying for. I heard someone saying that there was an engineer at a particular company, which I will not mention, that apparently spent $100 million a month on tokens. And so as a CFO, I would have concerns because that was clearly not in the budget, right? Not saying that that was from Lenovo. So we cannot have that happen. And I think we need to be able to drive the productivity or efficiencies as it relates to that budgeting of the token generation. And I think a lot of that would happen on device where you may just pay a higher price for a device, but you know what you're going to be able to do on compute for the security of the data and for the privacy that you want to interact with your AI agent. So just to be clear, for the computers themselves, they can do some inference, right? But where it makes sense you route it to the cloud Is that right That is our goal So the Lenovo Agentic AI today what we are good at is really integrating and maximizing the compute capabilities on a device. And from that perspective, given the various needs in terms of various operating systems, agents that may now want to sit on top of a device, I think our agent aims to orchestrate the various LLMs. We'll take the compressed versions of these LLMs. We will, depending on the partnership, be able to do the on-device compressed LLMs versions and do that as a local compute, but in certain queries allow it to go on the cloud and therefore probably would allow the user probably to spend in terms of the token generation. It's interesting this word orchestrate So you have the AI agent orchestrating maybe a bunch of different sub-agents to complete some tasks, and that is something that perhaps is done best on a CPU. A company that builds servers is also an orchestrator of the supply chain and acquiring the different components that go into a server, et cetera. I want to go back to something you said in your first answer because I think this action will get to the core of this new era You said okay an engineer spends million in a month on tokens and it like that would not make you happy as a CFO. But it could make you happy, right? What if you had a multi-year database migration plan that you think, oh, this would be a $500 million job, and the engineer does it in a month for $100 million via tokens, don't you have to at least be open to the possibility that that was money well spent? Absolutely, Joe. I think everything is about the return, right, and the planning. So we're not afraid to invest. As a CFO, you are there to allocate capital. You're not there to constrain capital. You have to allocate, but you have to be clear in terms of that return. And I think in that case is probably one where they weren't sure in terms of what they were particularly doing in terms of spending. It wasn't in budgeting. And that goes to the point of what is happening today. I think most enterprises were at the early stages of how people were changing from a subscription-based model to a token usage model in terms of the compute capabilities. And so that is at the beginning and enterprises are starting to figure out how do I really track that? [10:00] spend? And therefore, what do I really want to get out of that return? And so we're on that early stages. So you're absolutely right in terms of that we would want to spend if we can get a return out of it. You're absolutely right. And I think preliminary data and anecdotal evidence from a lot of people in our chat recognize the power of AI and the probability of generating increased productivity and potentially even cost savings. But say more about measuring the actual return And again, I realize it's early stages, but I'm curious how you think it will actually be done, because we've been asking a bunch of executives, how do you internally benchmark your AI projects or your token spend? And everyone says, oh, productivity or cost savings or whatever. But those are they sound very theoretical. So is there more concrete, I guess, KPIs that you're looking at? Yeah, I think it really depends on your specific enterprise. In our case, I would really hone in on a few areas. For me in terms of pricing and that visibility of demand in terms of channel inventory and the demand of 180 markets and what I would take to actually provide and work with my channel partners in terms of managing that inventory supply. There's a lot of economics involved in that. So if we can optimize that to run by AI rather than spread out in 180 markets, I think there could be a lot of productivity gains from that that could be seen from a dollar perspective rather than just generally in terms of making very blanket general statements. Other areas I would say that you could also increase productivity is pretty clear in terms of data spend that you have today. A lot of hedge fund friends and investor friends really talk about in terms of what their savings could be in terms of how much they're paying for certain data subscriptions. And then, of course, as it relates to IR functions, to M&A functions in terms of how you increase productivity, those functions tend to be smaller in headcount. So it's really less about the headcount reduction, but probably about the increased productivity. And therefore, are you making the right decision? Because those are small headcount, high impact areas, right You could have a higher stock price and therefore your cost of capital becomes much more more efficient in your financing Even M if you have better analysis you optimizing in terms of how you returning in terms of that capital deployment So from those perspectives I think it works out And then of course, areas like marketing, where we're spending a lot out there in terms of video and other production generation, those actually can be done by AI today. So we really need to look at that. So that's just a few examples, but I think there must be hundreds and hundreds, and we're in the process of trying to sort through that. Whether you're planning a big tech event, launching a new campaign, or just stocking up on team gear, finding the right promotional products makes all the difference. 4imprint offers thousands of options, from on-trend apparel and premium drinkware to tech, totes, and giveaways, so you can find the right fit for any audience, purpose, or budget. You can customize it all, your logo, your message, your look, and many items come with no setup charge to help you save. And if you're really watching the bottom line you find standout choices at every price point so you can and make a real impact while staying on budget Plus you get expert help fast turnaround times and their 360 guarantee So you can be 4imprint certain your order will arrive on time and look exactly right. Whatever your goal, 4imprint makes it easy to find your perfect promo match. Explore the possibilities today at 4imprint.com. 4imprint. 4certain. Support for the show comes from public.com. If you're actively involved in your portfolio, you probably catch yourself repeating the same actions. Buying the dip, manually sweeping idle cash, putting on a hedge. On public, you can now create AI agents that handle all these tasks on your behalf. Just describe what you want to do in plain English. Like, if the VIX hits 25, buy a put option on the S&P 500. Or, if my cash balance goes above $20,000, move the excess into my direct index. you approve the workflow and your agent handles the rest monitoring the market watching for your conditions and executing your strategies exactly as defined an investing platform driven by your intent not just your clicks you can also get full read and write access to your account via the public api go to public.com slash market and fund your account in five minutes or less that's public.com [15:00] slash market. Paid for by Public Investing. Brokered services by Open to the Public Investing, Inc., member FINRA and SIPC. Advisory services by Public Advisors, LLC, SEC Registered Advisor. Complete disclosures available at public.com slash disclosures. Wise is the smart way to manage the currencies you need around the globe. If you've ever sent money internationally using a traditional bank, there's a good chance you've paid more than you realize. Hidden fees, exchange rate markups, and extra charges can quietly add up before your money even arrives. There's a better way. Try Wise. Wise uses the exchange rate you'd usually find on Google, helping you avoid the unwelcome surprises that often come with international transfers. Whether you're sending money to family overseas, spending while on your holiday abroad, or paying bills across borders, Wise makes moving money simple, transparent, and straightforward. Wise offers 24-7 customer service and runs over 7 million daily checks to spot and stop fraud. And most transfers happen in under 20 seconds, which means your money arrives in less time than you've been listening to me. Join millions saving billions. Be smart. Get wise. Visit wise or download the wise app today T and C apply Let go back to the theoretical engineer who spends million on tokens in a month because, you know, it seems to me like you're going to have someone who actually can, with their, you know, intuitions and skills with AI, etc., actually deliver ROI, even if the nominal amount they spend is nose bleeding. And then other people are just going to waste a lot of money. And I guarantee I know a lot about wasting money on AI. And I guarantee that it's very easy to develop that psychosis whereby you believe you're doing some incredible things that feel like magic and productivity. And it turns out that you really haven't built anything, et cetera. It's very easy to delude yourself into thinking you're being productive with one's AI use. What I'm curious about specifically is how do you even go about, say, identifying the employees or the teams who might really be people who you really should give a long leash to with their spend? How do you even like identify who those are that should Yes let not be too aggressive in their consumption in capping their consumption Don give all your tokens to a Joe with AI psychosis How do you avoid giving me the uncapped budget and finding that person who actually deserves a very liberal spending budget Yeah. And these are great questions. They're so relevant, particularly in a large enterprise like us in a global for a global 200 company. right in 180 markets, 80 plus billion in terms of revenues. We have massive scales and so many tens and thousands of employees. So it's really difficult from a CFO's perspective to micromanage down to the person's allocation. So you really have to be able to say, first of all, what are the agents in each domain capable of doing today? Right. And how what is that productivity and also capability. And then you can say, well, what specifically can it do today? And then let's get the dollars, at least from the CFO perspective, let's then eliminate the dollar spent in those areas and really force the AI. I have come to probably the conclusion that to really effectively drive things I need to then force discipline or starve certain budgets to then allocate because that really changes behavior Because if you continue to allocate that budget they will continue to use it and go on the old behavior But I think human beings are really adaptable. And I think we tend to be good at survival, right? And so if you force a certain situation, I think people, particularly with the functions of AI today, people will be able to then really try to use it because they don't have the budget for the alternative. I think that's really from a CFO's perspective. Yeah. Well, let me press on it. I find this to be very interesting and maybe it even gets to a certain philosophical split, perhaps, among CFOs, because I think, OK, there's one school of thought would say, OK, you keep these very tight constraints and then you see who can do the most with the constraints. And then you derive valuable information that comes back into your office that that is valuable data. And then there seems to be another school, which is let everyone have really high caps. And we read these stories at American tech companies about the internal quasi token maxing competition. And I think that school of thought would say, look, let the employees go to town. And that is the way that we discover who can actually use these tools productively, et cetera. But would you say that that is actually a sort of a philosophical split in CFO? [20:00] foes of tech companies in the early years figuring out this optimal path? And I think that that decision exists within the same company, but it could be two paths that you just described. Because, for example, where we really it's a phase of innovation today in terms of where the market's going. And so we really need to put more dollars in terms of innovation. And you can see that in terms of our spend. So where we really and that's the emphasis from our chairman and CEO as well, is really around making sure that we take advantage of this period. We're calling this the AI decade within Lenovo, and we're in the first year of that journey. And I think this is an opportune time for us because the way we're positioned today, where we have global manufacturing in a fairly fragmented world, where we have a lot of data and production and supply chain concerns, where we are able to be local and be able to supply that on a local basis and complies to each region's needs, sustainability needs for certain regions, particularly like EU. 100% of our laptops that you like is actually in recyclable box paper. And also within the laptops a lot of them are also recyclable materials as well And we also have a refurbished effort given the components constraints today So I think from that perspective we really want to spend dollar on innovation Now, as you say, then there's the other pocket. Should the likes of even within finance be able to have an unlimited budget on AI? I think you could probably be more sensible in that and say, well, specifically within certain analysis in FP&A or accounting, what was it that you actually used to spend? So for example, without naming a party, we outsource certain back office function to a service provider. And that's a pretty costly one for us annually. Now, in terms of that capability set, it's really a process flow. So should we ultimately be able to automate everything and be able to achieve savings in that. Now, we are okay to continue to outsource as long as that partner also innovates and be able to share that savings with us as well. So I think it really depends on where you want to go very specific in each area within your function and your company. So in my function, there are so many areas planning tax FP accounting treasury IR M right you name it corporate finance So within all of these areas where are we going to optimize be more efficient productive And I think that cost dollar sometimes to me is the return is significant. I was just with the head of my tax this morning. I said, if we just pull together and drive the next year's efforts, the tax savings or optimization, right, that we do, we'll more than pay for a little small trip of 20 something people together. That's very, very much well spent. But I think that's just a philosophy. It's less AI driven, but it's really the philosophy of where that return is. And I think you allocated earlier, you spend based on the ROI generation. And I think today you have to recognize first the AI capabilities that you can have. I think the other thing we haven't touched upon is really the security within the enterprise of using AI, right? And that's a broad topic that we really should touch on. Because today, I think there is really concerns about what we don't know about using external AI within our enterprise. So I think that's also a concern in terms of how effectively you can use AI. We should definitely talk more about that. But you just mentioned the word unlimited in the context of AI spend. And coming from the U this is something that we really seeing and I guess living through at the moment like people talk about the total addressable market being like basically infinite at this point If you look at the SpaceX IPO, like the total addressable market is now the entire universe, right? I guess my question is, how do you compete against those types of companies who seem to have investors throwing money at them? I know your stock price has gone up quite a lot. So that obviously helps. And then a more broader question, what are the key differences between how Chinese companies are approaching AI versus how U.S. companies are approaching AI? Sounds good. I think there's two very distinct questions there. The first one really with respect to Lenovo's position in the entire tech ecosystem. I was in Davos earlier this year, and when I came back, I was particularly affected by Mark Carney's speech. and really talking about the superpowers and the middle powers. And of course, within the middle power, there are superpowers. I actually use that within my own management committee in terms of referring our own position within the tech stack. And I think as big as we are- As a middle power, you mean? I think we're definitely not a superpower. We have to recognize that. But I think from our perspective, being the number one device PC manufacturer, [25:00] in the world, having a strong ecosystem and the best, broadest portfolio set across PC, tablets, and mobile phones, and great partners to Microsoft and Google's of the world in terms of that type of operating system. And of course, the chip suppliers like NVIDIA, Intel, AMD, and Qualcomm, and of course, now ARM-based chips as well, and also the Chinese tech stack that's also coming up. I think from that perspective, we are there to really enable. And I think prior to this stock run, we're really not recognized by the market in terms of our ability to be partners to 2,000 suppliers in the tech ecosystem, driving almost $2 trillion plus of spend. And we're right in the middle of it and distributing this across the board evenly in 180 markets. So I think from that perspective, we really needed to be recognized. But at the same time, we're very humble. And I think our chairman really used that word in terms of staying humble. And I think we are because we're also micro. And from that perspective, most of the profits have been towards the IC companies, the OS companies If you look at the PC revolution and the tech stack the OS company is now trillion dollars The chip companies are now crossing trillion dollars but have before been hundreds of billions of dollars And we only most recently became around I think 30 to 40 billion dollars But before that, sub 20 billion. So I think that's the value chain that you're in. But I think we probably did not optimize it as well. And I think there's an opportune time for the device makers today to optimize. And of course, distribution is very critical. Supply chain is critical. The ability to aggregate supply chain to manufacture at scale, at cost efficiently, and deliver to your customers, to be able to service it, to provide the security and trust and the aftermarket service support, that must have a value beyond what we're being recognized by the market. I mean, I absolutely believe in that. And I think we need to drive that in terms of financial returns from that perspective. and AI wave definitely creates an opportunity for the likes of Lenovo. And I think we're the best position in this because there's no company like us from the end-to-end portfolio and the global manufacturing. So, you know, I think that's a very important point. And in terms of how the US and Chinese tech stack today are differentiated or AI companies are differentiated Today obviously because of the restrictions I know that probably the most efficient compute today is probably from the US chip company. So obviously the chip availability is something that the Chinese AI companies probably have to work with. But in that way, they probably managed to produce at much lower cost and very efficiently as well. So you see today a lot of chatter about the cost per token generation from the like of a deep sick being and i don't have official numbers so this is just quoting what i've heard like 150th of those in the u.s so i think from that perspective right as i said human beings or companies and if you really think about the darwinism in economic theories the free markets create the toughest competitors and the most healthy competitors so i think in china for those that compete in such a there's a word called involution in china yeah of course We know well. It only happens in China market. So if they can survive within their cutthroat market under the constraints of the chip supply, they are pretty strong. You can survive anywhere. They are pretty strong. So I think that would be one difference I would say. Yeah. I want to talk about competition You mentioned the deep and broad relationship you have with all these suppliers across the supply chain And when I think about like when I think about competition within the server space specifically I feel like there are two dimensions through which you could win And one of them is, of course, like, OK, you have total supply chain mastery. mastery. You can keep inventories low because you're just in time delivery of everything. And you just have this beautifully efficient supply chain with the world's best parts makers all around the world. And then there's another way you could win, which is your server is just more performance than a competitor's server, et cetera. And we know other competitors like Adele or Supermicro, et cetera. Those both strike me as dimensions upon which a company could win the server market. Which to you is more important? Is it the sort of supply chain excellence or is it the quality of the server itself? I think it's not an either or question. It's fully integrated tech stack. Today, the value to a customer and a partner, and I think there are less customers today than they are really a partner because their architecture is not driven and the spend is a multi-year spend. [30:00] So they're coming to you with a multi-year plan, not just a single purchase. So we take that attitude, and we need to have that attitude with our suppliers as well. We have invested well over 10 years in these factories. We have 30 factories in the world, 12 in each region of the world. We can produce in Europe, in Hungary, for the European market or the U.S. market. We can produce in Mexico, in the U.S., for the U.S. market. We can produce in Asia for the Asian market. Now we're building that factory in the middle, in Riyadh. to produce servers also for the Middle East market. And of course, China market, we always have factories there as well for the China market and other markets. So I think from that perspective, we are really truly global from that perspective. Today, the bottlenecks are not only in getting supply. After you supply, can you then put it together? Do you have the production capability to be able to produce? Then do you have the capability to test what is being produced? So we at Lenovo provide that. end-to-end capability to our customer, and that ensures them to have the safety and one-stop shop. In fact, our SSG can actually also, so the Solutions and Services Group at Lenovo can also build data centers So different from our OEM competitors they can actually build data centers So today, the customers can, hyperscalers can actually come to Lenovo with a plan to say, I have a plan to have this site. So people who are not even hyperscalers, people who are land owners and they have power in specific areas, they come to Lenovo and they said, help me build a data center because we also have a modular solution that can build this data center within nine months. Depending on what you already have available in terms of infrastructure, we can do it as fast as six months. So that is very, very fast. And that gets the time to market and enable revenues immediately for our partners. And then in terms of that total infrastructure, plus if they can have GPU compute in the local market, then, of course, Lenovo can also provide that, particularly with our 11,000 rack liquid cooling capability to be able to service a GPU compute today, right? And we're expanding on that capability. So today, we're the most end-to-end and relevant partner really from that domain. So it really not from a perspective of really having the supply and having the technology Technology is always important but that end look if you think about how people are affecting and enabling AI compute today they need everything. And of course, as Tracy said earlier, I think capital is also an important part of that. And it does help that that market is starting to recognize Renovil in terms of our value add. Thank you. your message, your look, and many items come with no setup charge to help you save. And if you're really watching the bottom line, you'll find standout choices at every price point so you can make a real impact while staying on budget. Plus, you'll get expert help, fast turnaround times and their 360 guarantee so you can be 4imprint certain your order will arrive on time and look exactly right Whatever your goal 4imprint makes it easy to find your perfect promo match Explore the possibilities today at 4imprint.com. 4imprint. 4certain. Support for this show comes from public.com. If you're actively involved in your portfolio, you probably catch yourself repeating the same actions. Buying the dip, manually sweeping idle cash, putting on a hedge. On public, you can now create AI agents that handle all these tasks on your behalf. Just describe what you want to do in plain English. Like, if the VIX hits 25, buy a put option on the S&P 500. Or, if my cash balance goes above $20,000, move the excess into my direct index. You approve of the workflow and your agent handles the rest. Monitoring the market, watching for your conditions, and executing your strategies exactly as defined. an investing platform driven by your intent not just your clicks you can also get full read and write access to your account via the public api go to public.com slash market and fund your account in five minutes or less that's public.com slash market paid for by public investing brokered services by open to the public investing inc member finra and sipc advisory service [35:00] By Public Advisors, LLC, SEC Registered Advisor. Complete disclosures available at public.com slash disclosures. WISE is the smart way to manage the currencies you need around the globe. If you've ever sent money internationally using a traditional bank, there's a good chance you've paid more than you realize. Hidden fees, exchange rate markups, and extra charges can quietly add up before your money even arrives. There's a better way. Try WISE. WISE uses the exchange rate you'd usually find on Google, helping you avoid the unwelcome surprises that often come with international transfers. Whether you're sending money to family overseas, spending while on your holiday abroad, or paying bills across borders, Wise makes moving money simple, transparent, and straightforward. Wise offers 24-7 customer service and runs over 7 million daily checks to spot and stop fraud. And most transfers happen in under 20 seconds, which means your money arrives in less time than you've been listening to me. Join millions, saving billions. Be smart. Get wise. Visit wise.com or download the wise app today. T's and C's apply. Tracy I thinking about the difference between you know the server market today and say like the PC market of the 90s And I think part of the reason we talk about and remember the red button on the laptop is simply because the companies were unable to differentiate themselves at all product wise. And it was like truly the PC market kind of killed itself because it was just so unbelievably commodified. No one really knew the difference or cared between the difference between a dull box and a compact box and you heal a Packard box and whatever. But it does sound, listening to a Winston, that you really can, it's the partner, not the customer, you can really differentiate your offering. Yeah, I remember that environment. So I remember you had to go to Best Buy, right? And you had to basically get a salesman to tell you what the difference was between all these different things. Okay, since we're on the topic of supply chains and you mentioned data centers as well, Give us some color on what it's like to try to get, I guess when I think about constraints on the AI build out, I think about memory, GPUs, CPUs, chips, and transformers now for data centers, at least in the US. How difficult is that at the moment getting those key things Yeah the market is very robust because the demand is very robust And as we sitting in Hong Kong today but in the Bloomberg offices but really there are some really exciting IPOs coming right And also capital raises We seeing capital raises at such significant levels And that really all going back to AI infrastructure spend So really, from the perspective of that component shortage or demand driven challenge on the supply chain will continue probably for at least two to three years. But we know that the supply chain is starting to invest. But in terms of, you know, some some of the memory specific areas, it does take almost two to three years for a new fab to come online. So I think from that perspective, and I think you're seeing bottlenecks across the board. And as I said earlier, land, power, not just the transformers, but actual power from the grid. And that's why we're in the Middle East to work with the Saudi government on sustainable, but also cheap solar and much available energy in the local market. So it's very efficient there. And they have very low cost energy there that could actually support AI infrastructure from that perspective So we really need to optimize where specifically the power is around the world rather than just saying hey everything has to be generated within an area and that area doesn't even have power. So I think we really need to be able to do that, pull the global resources together. One of the things that we all learned during the pandemic is that, you know, and we all learned about how supply chain disruptions work and the bullwhip effect, or at least, you know, us lay people learned about these concepts for the first time. Everyone we talked to, we recently did an episode with one of the co-founders of CoreWeave, one of the NeoClouds, you know, everyone we talked to will talk about multi-year-long order books and backlogs, et cetera. And when I hear that and I started thinking back, I was like, yes, I believe you when you say that, but it could also be a bunch of players double ordering and triple ordering and trying to get ahead of themselves just because they're all aware of the shortages, et cetera, in which case it does seem like what looks right now like a major backlog and constraints and endless demand could not be as sustainable as people thought. Should we perhaps be worried that some of these endless order books are just, in fact, companies overworked? [40:00] ordering because they see everyone else over ordering? I think these are very sophisticated companies in terms of the final off takers. And so, but I think I would point out to the fact that the duplication could happen in the pipeline, but not necessarily in a backlog, because I think if people have standard definitions of the backlog, by that time, it's actually already a signed deal. So I think they wouldn't double sign with people. And then of course, revenue recognition is a totally different thing. I think you have already shipped the actual product and you expect to collect on your accounts receivables. So I think from that perspective, really looking at the pipeline that is massive in the market, there is likely duplication there and people are waiting to see who has the supply and who can deliver to you. But overall, I think the commitment that we see in the market, particularly for data center space, for equipment, for other things is really a multi-year commitment. And I think people see this as a critical infrastructure. They learn from the internet era in terms of the cloud service providers, how much data can be generated. And this is just going to be so much more. I think it going to be exponentially more than what it is because it will take the existing data plus generate more data And then you need to enable that data Right And that enablement I think needs additional compute Lenovo has software as well Right. And you have the gaming business. Is that right? We have a we're the leading online games. Oh, sorry. The game PC company in the world. Yeah. So what do you think about the Saspocalypse idea? So the idea that with AI, we're going to be able to replicate whatever application we want, whether it's Microsoft Word or Excel or something like that, would that eat into your margins? Or because you own the IP, can you just do it more cheaply? No, I think we are not a SaaS company by any means. But I think from the perspective of Lenovo or enterprises, I think really, as it relates to what we touched upon earlier, your IT department, and I think it actually works for a company like Lenovo, because we are not a software centric company. These tools now lower the bar for us. So I think our engineers or hiring engineers should be able to create certain apps that should be available or applications. So I think from that perspective, there are certain elements of applications that would be able to produce in and that would lower the bar because these tools are very powerful But in terms of the SaaS copopculus I think it hard for me to say but I think from that perspective, I believe there is certain data layers that are absolutely essential because enterprises almost cannot get away from this and not, nor do they really want to. I think you just need the middle layer to be able to bring the data together and to be able to connect to AI. And that's one of the biggest bottlenecks for enterprises today. And whoever can do that, right, would be able to be our helper. But I think it's rather than just purely a software solution. There's probably a consultant element to that as well to help enterprise enable this function. And that but I think we're a long journey from that. You know, I feel sorry for gamers because even prior to AI gobbling up some of the GPUs right before that, it was crypto. And I remember they used to complain that the Ethereum miners were buying up all of their NVIDIA GPUs. GPUs, et cetera. But I'm curious, like, you know, this is another one of the big meta themes of the time, like in your own gaming business or in the gaming industry in general, obviously, GPUs seems like the highest marginal value you can get from them is with artificial intelligence Do you see this continuing phenomenon where raw inputs of any sort that might have at one time go to gamers either because the gamers get priced out or no one builds them for gamers because you can make so much more money selling them into the AI ecosystem? Well, we got a fantastic franchise in terms of, as you alluded to, the ThinkPad business. It's an iconic franchise. And we love it. Our customers love it. I've actually visited the R&D facility that originated actually from Japan. It's from the Yamamoto Labs of IBM. So we kept that intact also with a lot of the people from it. But I think the original inventor actually retired some years ago. But there's a long history that we're very proud of, and we've actually maintained that. In terms of our consumer and also as it relates specifically to the online game business, which is our Legion business, it's actually also very famous. It's the largest by volume and people really like Lenovo for the price performance that we can deliver to them. And gamers really appreciate that. So these are amazing franchises that we want to be able to keep and they are a loyal fan base and we have the largest market share in the world. So we want to be able to protect that. [45:00] We actually, rather than protect that, we want to grow that market share. So I think from that perspective, we will have to balance between the device side of our business as well as the infrastructure side of the business, which at least they have a lot of demand today. I think it just means that we have to get more share of our supply. Talk to us about your decision to, I guess, not tie yourself to any one AI platform, because I'm sure there are some very, very big tech companies out there who would love to have exclusivity with Lenovo. And there are also broader arguments out there that eventually we're going to have one AI platform that emerges triumphant in the same way that, you know, we saw Google take over search in the early 2000s. Why did you decide to, I guess, work with a bunch of different people here? I think today what we see is that we're still in that journey of improvement and you see improvements every few months from the likes of OpenAI or Anthropic or Gemini or whoever it might be. And so I think from that perspective, you see a lot of innovation also from the China market. So from that perspective you really want to be able to be flexible And we think today we not there to bet on who is going to win We are there to provide the AI compute So from now the orchestration there makes a lot of sense for us to allow our customers to be able to go through the Lenovo AI and reach what works best for them, rather than them have to figure out and download multiple apps to be able to do that. Right. So I think from that perspective, the orchestrator would do that for them. And I think that makes it much more efficient, both from a memory perspective and compute capability perspective that is on your device. So with that architecture should actually potentially optimize the performance on the device. Should we believe these companies, let's say in Microsoft, so Microsoft recently announced their flagship model. I forget if it's pronounced like Maya or Maya or whatever, or MAI, I haven't played with it. And they're like, and it also runs best in our own custom silicon. how should we read that is there in your view a lot of a juice to be squeezed from model silicon alignment or should we read this as at least companies would like to begin having a little bit of a wedge so that they not so dependent on NVIDIA for their hardware needs I think there a lot of great things that NVIDIA is doing today And I think a lot of people are recognizing So their revenues are clearly growing significantly. I think they're probably at the core of driving this current and enabling this current AI wave. So I think that continues. But of course, as costs become to a level where people have to find alternatives, you're also seeing people from a cost perspective making selections, but also probably from, as you say, the flexibility of having their own architecture. But I think that's really happening with respect to the hyperscalers who have the scale and also the technological capability and the capital to be able to develop their own. But is it about their models really will run better on these chips that they're designing, Or is it about long term strategic at least, you know, I wouldn't want to say divorce from NVIDIA, but maybe, you know, some time apart. They had a prenup or something like that. Just like having a little bit of like not so dependent on one company's specific chip capacity and roadmap. I can speak for them but just really from an external point of view But yeah but I believe it a little bit of both right Because if you were a corporate really deciding on your strategy I think that's probably one where, especially given that a lot of them are also tech stack focused companies as well. So they probably want to be, have more control over their own tech stack. But the other one really is around the cost and also maybe future planning around what they're enabling. So probably give them more flexibility to be able to do more if they have their own chips in other areas as well. So speaking of diversification, if we were recording this episode a year or so ago, we would probably still be talking about AI, but I think we'd be talking about trade and tariffs as well. Right. Remember that show? I forgot about that story. That's right. So Lenovo obviously has a very large and complex supply chain all across the world. But how does, I guess, the general return of more trade restrictions actually impact your business? Yeah, it actually happened very interestingly on April 2nd. So I took over on the reins. Our fiscal year is April 1st. I took over on the reins on April 1st. So the second day, it was the most major event in the history of 100%. [50:00] I think it was 200 plus countries that were enacted on this, and we're probably in 180 of those. So I think it's a major event given the significance of our business. But it's really about enabling and providing our products at a very reasonable price to our end customers. So I think the end customer suffers if there is inefficiencies added. So I think that's the most important aspect, which is and I think eventually, right, I think PC products were actually exempt. So I think there was a conclusion that this is something that was needed for productivity, for entertainment, for a lot of things that people use devices for today. And we're one of the largest providers across the board, whether it's in your pocket every day or on your desktop or at home. So I think from that perspective, this is probably a very essential thing for consumers today. And I think that's the most important in terms of giving them that product at the lowest cost. You see that behavior really in a lot of probably third world countries or second world countries in terms of really emphasizing that because it's really about not leaving their citizens behind. It's really about making sure that they have access to digital information and the device is the start of that journey Right So that a very critical aspect of it So lowering the tariff barrier is actually essential to ensure that their citizens and any citizens should have the lowest cost access to technology. You mentioned something earlier and you're like, OK, on a per token basis, DeepSeek is significantly cheaper than, say, the most advanced models from, say, OpenAI and anthropic. But strictly speaking, there is a wide agreement that still the flagship models from the American labs are the best models in the entire world. And I just don't think there's much dispute about that in either the U.S. or Chinese communities, at least a few months behind, perhaps even inside Lenovo. But do you see in general, and this sort of gets back to the early part of the question about CFO decisions, people not being sophisticated about recognizing that queries don't always have to go to the most expensive, most advanced models. And how much learning is there still yet to do about how to optimize routing of the query such that it goes to the most cost model rather than just slamming the frontier model Yeah I going to answer your question in my how I started my morning today because I in the journey of figuring this out Yeah And I decided to spend some money with external consultants. And so we're in that interview process. And as I said, you know, we spent an hour and a half with the global team. And this is one of the major firms. I'm starting to get a sense from talking to various parties that everyone is new to this game. Yeah. Everyone, even the advisors who are trying to earn a service fee from you is really also learning as part of this journey. And that's what consultants do as well. Do you ever ask what you're paying them for? You're like, okay, we're all learning this together. But I would not make comments about another industry. I'm sure you love them. Absolutely, absolutely. But, you know, I think from that perspective, that goes to say that if they are the specialists or those that see a lot in the market, then what about specific departments or individuals in the company? So I think from that perspective, everyone is in that journey and it's the people who can be most clear-headed to learn the fastest. I think it's probably the most essential, right? So I think from that perspective, we need to be able to take out the red tape, enable the processes to work accelerate I think timing and speed is essential in the new AI world And then we need to deploy that where it sensible And then as you say where do you actually allow that spending and the interaction to generate tokens to really happen? And so we're in that journey. But I think we're not alone. I think every company is the same. I'm actually going to a conference next week hosted by a major consulting firm that is gathering a lot of CFOs. But in my regular dialogue with CFOs. I understand. I worry a lot, but I think we're also not alone. Joe, do you think in the future when people meet each other, they're going to say their names and their nationality and then declare... Well, I was going to say declare the foundational model that they use. Oh, I was going to say I wonder if at bars or women when they gossip with each other, they're like, I want a man who is this much token spending. You know, whatever. So it'll be part of their identities for sure. That's right. Well, OK, so speaking of rich men, I have to ask at least one CFO-ish question, which is when it comes to capital spend, I'm sure you could justify pretty much anything right now. And there's so much money that's actually going out on the AI build out. How are you thinking about returning capital? [55:00] capital to shareholders because the stock's gone up a lot. But, you know, people can always be wealthier. I'm sure they're into that. Yeah, we paid the in the last fiscal year, which just ended March 31st, we paid the highest dividend ever in the Lenovo history. So our shareholders are very happy. We share our success with our shareholders. But we also see a lot of opportunities for growth. And as you have mentioned many times on this discussion, is that there's also a lot of capital needs to fuel that growth. So we really need to strike that balance between giving our shareholders the immediate cash that they would like, but also at the same time, they want us to be able to enable the capital appreciation on the stock because the underlying fundamental of the business is growing. So we need to drive that growth. And there's definitely going to be a period where you drive that growth. At the same time, we're driving that margin expansion, but there will be quarters here and there that could potentially mismatch in terms of that growth versus the margins. But overall, the long term trend for Lenovo is a plan to drive growth with accelerated margin expansion. Northern Virginia in the U is sort of understood to be the data center capital of America But you know it expanding There a lot that sort of like off grid in Texas that happening And then there also of course anti center politics in the U And we don't know how that will affect the map. What is the northern Virginia of East Asia right now? Where do you see, like, which country is it because they have the most, whether it's regulations or access to energy? Where is everyone trying to put up data centers? And I'm just curious, like anywhere in Asia, is there any sort of equivalent anti-data center politics that's emerging? Well, I think Asia tends to be a little bit more flexible from that perspective. But clearly, the best infrastructure of scale is in China. But I think from that perspective, a lot of people, due to certain regulations and policies, cannot have data centers. So you have a lot of idle resources today whether it's power and data center, therefore accessibility and also supply chain, right? That's not that the world is spending additional money on in higher cost jurisdictions because of regulations that you cannot put it in China. So the world ends up having some of the inflationary effects is because you not actually using the most efficient place in the world that you can actually do things But having said that I think other countries are actually catching up very fast as well So we seeing even in the supply chain area slightly lower cost But in terms of the data center aspect and where the power, Southeast Asia has been very good. And I think what we're seeing in this part of the world, clearly in places like Malaysia and Indonesia, I think those are natural places where people have really been looking into. I mean, even Hong Kong these days, I think there's plans from the government, you would think, and also Singapore. So some of these smaller land mass areas, but I think also have plans. And I think right here in town, there's also a dedicated few players as well. And they have to be creative in terms of putting data centers in these higher elevated structures. But there's all sort of creativity. And I think Hong Kong has a mixed use in terms of that power generation. But again, that's a small area. But I think it's just giving a data set of examples. People are also doing it in Japan. Japan is also another one from a regulatory perspective. But cost-wise and also policy-wise in terms of getting necessary permits may not be the fastest. Tracy spent a long time there, so may have some knowledge about this. But I think it really depends on where you can have trust and partnership with the local government because it involves land it involves power it involves imports of specific goods And I think that part really depends on a lot of the local efficiencies. Joe, we got to go visit a data center in Hong Kong. I'd love to. I was going to say Hong Kong actually to me makes a lot of sense as a data center location because if there's one thing I know about Hong Kong, it's there's plenty of cheap real estate here. And there's more than enough space. And we know that people are very happy with the price of how affordable the residence is here, etc. So it makes sense that, you know, I'm being facetious. There is actually a lot of land. You just can't build on it. You just can't build on it, right? Yeah, so. All right. Winston Chang, thank you so much for coming on All Thoughts. Really appreciate it. Thank you, Tracy. Thank you, Joe. joe that was a lot of fun yeah i feel like we haven't actually spoken about ai from the perspective of a company like lenovo no like you've never done that before yeah no we haven't and that was very nice and look part of the [60:00] I would say two things about AI, which is we're doing a lot of AI episodes. And I would say there's two reasons for that. One is because, I mean, it's just the biggest thing of our lives, probably, in terms of the significance and trying to everyone's trying to wrap their heads around it. So there's a lot to learn. But it also is in our wheelhouse because it's so physical and because it interacts with supply chains. And so you think like, OK, a company like Lenovo, a server company that also happens to offer the ability to build complete data centers, not just the servers, who therefore is then interacting with all types of players is almost like the perfect sweet spot for like an odd lots guest, even among all of our perfect guests. Someone who has that visibility on both sides is really someone interesting to talk to. Yeah, I think I've said before, there's such a weird tension between these sort of bodiless, faceless AI interface. And then when you think about all the actual physical infrastructure that supports it. It's why I think it's so interesting. The other thing too that I think is interesting about hearing from a Lenovo and again hearing from a CFO of Lenovo is like look they not a hyperscaler I sure they investing a ton of money and building out their capabilities et cetera but they not one of these companies that announces big things like we going to be spending billion next year on building out data centers. So they really do have that dimension where they have to figure out the Googles and et cetera, and open AIs. They're talking about CapEx, and the Lenovo's of the world also have to think a lot more about OPEX. And the OPEX component of the conversation, as he said, whether it's with fellow CFOs or the consultant community, it doesn't sound like it feels like it's in day one of figuring it out. Everything's up for grabs. Yeah. I also thought the involution point was interesting when we were talking about the key differences between China and USAI and the idea that because competition is so cutthroat in China, that it just drives your costs down and down and down so that like, even if you're not that competitive in China, you would still be competitive compared to... On a global scale. Yeah, exactly. Yeah, it's kind of funny to think about. Yeah, absolutely. No, I love that chat. All right. Shall we leave it there? Let's leave it there. This has been another episode of the All Thoughts Podcast I Tracey Alloway You can follow me at Tracy Allaway And I Joe Weisenthal You can follow me at The Stalwart Follow our producers Carmen Rodriguez at Carmen Armin Dashiell Bennett at Dashbot, Kale Brooks at Kale Brooks, and Kevin Lozano at Kevin Lloyd Lozano. And for more Odd Lots content, go to Bloomberg.com slash Odd Lots. We have a daily newsletter and all of our episodes. And you can chat about all these topics 24-7 in our Discord, discord.gg slash Odd Lots. And if you enjoy OddLots, if you like it when you hear from the perspective of Chinese tech companies, then please leave us a positive review on your favorite podcast platform. And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening. Thank you Explore more at 4imprint.com. 4imprint. 4certain. Healthcare doesn't always work great. If you've ever waited on a refill or couldn't schedule an appointment, you get it. That's the kind of stuff Optum is changing. They're using data and technology to integrate patient care, pharmacy, and everything else. So healthcare is connected, not complicated. What's that look like? Cheaper prescriptions that are easier to get and care that looks at the whole person how you need it. Optum is helping make healthcare work as one for everyone. Learn more at business.optum.com. [65:00] A single layer of control, a single standard of trust. So whether an AI agent supports a single user or your entire enterprise, with Okta, you'll turn risk into opportunity. Secure every agent. Secure any agent. Okta secures AI.