AI’s Shift From Thinking to Taking Action

欢迎收听《市场洞察》。

Welcome to thoughts on the market.

我是肖恩·金，摩根士丹利欧洲与亚洲科技团队负责人。

I'm Shawn Kim, head of Morgan Stanley's Europe and Asia technology team.

今天，我们来聊聊人工智能发展中的根本性转变，以及它对市场的广泛影响。

Today, a foundational shift in the development of AI and its broad market implications.

现在是五月五日，星期二，伦敦时间下午三点。

It's Tuesday, May 5 at 3PM in London.

想想你上次让聊天机器人帮你写个总结、草稿，或者回答问题的情景。

Think about the last time you asked a chatbot to write a summary or draft or maybe answer a query.

当时可能挺有用，但整个过程还是你主导的——你提问、修改、复制、核对，推动工作前进。

It was probably useful, but you were also still driving the interaction, asking, refining, copying, checking, and moving the work forward.

现在，想象一个不仅能回应，还能主动行动的系统。

Now imagine a system that does not just respond, but acts.

它记得你上周提过的需求，了解你的偏好，跨多个数字工具工作，规划工作流程，并根据情况变化自动调整。

It remembers what you asked last week, understands your preferences, works across digital tools, plans a workflow, and adapts as circumstances change.

这标志着从生成式AI向代理式AI的转变。

That is a shift from GenAI to agentic AI.

从帮助你思考的AI，变成帮助你行动的AI。

From AI that helps with thinking to AI that helps with doing.

生成式AI大多是被动的。

GenAI is mostly passive.

它收到一个指令，就给出一个回应。

It takes a prompt and produces an answer.

代理式AI是主动的，不只是帮您完成单个任务的副驾驶，而是能自动执行多步骤流程的自动驾驶系统。

Agentic AI is active, less a copilot for one task, but an autopilot for multi step workflows.

这个区别很关键，因为计算需求正在发生变化。

The distinction is key because computing requirements are changing.

在生成式AI中，大型语言模型和GPU承担了大部分的思考任务。

In GenAI, large language models and GPUs handle much of the thinking.

GPU，也就是图形处理器，能并行处理大量计算，因此成为现代AI模型的核心。

GPUs or graphics processing units process many calculations in parallel, making them central to modern AI models.

在智能代理AI中，CPU变得更为关键。

In agentic AI, CPU becomes more important.

CPU，也就是中央处理器，负责协调任务，并将系统连接到更广泛的数字基础设施。

CPUs or central processing units coordinate tasks and connect systems to the broader digital infrastructure.

生成式AI也依赖三个层级：大脑——也就是大语言模型，协调层——由CPU负责执行任务，还有知识层——相当于记忆。

A GenAI also depends on three stacks, the brain or the large language model, orchestration where the CPU manages the doing, and knowledge, which is memory.

记忆可能是最重要的那一层。

Memory may be the most important layer.

一个了解你的偏好、文档、语气和任务历史的代理，会随着时间越来越有用。

An agent that knows your preferences, documents, tone, and task history becomes more useful over time.

这就形成了一个情境飞轮。

That creates a context flywheel.

它收集的情境越多，就越个性化，也越让人离不开。

The more context it collects, the more personalized it becomes and the harder it is to leave.

通常在计算领域，我们把内存主要看作是存储空间。

Typically in computing, we think of memory as storage mainly.

我们需要重新思考这个问题。

We need to rethink this.

记忆还意味着连续性。

Memory is also continuity.

当AI系统能利用过去的经验时，记忆就变成了长期状态、共享知识和行为基础。

When an AI system can use past experiences, memory becomes a long term state, shared knowledge, and behavioral grounding.

这很关键，因为大语言模型的上下文窗口是固定的。

And that matters because LLMs have fixed context windows.

一旦对话超出这个窗口，之前的所有内容就全丢了。

Once a conversation exceeds that window, all the contents falls off.

简单问题可能还好，但对一个几天甚至几周内处理大型代码库的编程助手来说，这就成大问题了。

For simple questions, that may be fine, but for a coding agent working across a large code base over days or weeks, it is a major limitation.

真正的工作需要持久的记忆、短期的聚焦，以及主动检索能力。

Serious work requires persistent memory, short term orientation and active retrieval.

记住之前的决策，理解哪些文件被修改过，还能在不用用户逐个指明依赖的情况下找到相关代码。

Remembering prior decisions, understanding changed files, and finding relevant codes without the user pointing to every dependency.

对投资者来说，这意味着什么很清楚。

For investors, the implication is clear.

代理式AI改变了瓶颈所在。

Agentic AI changes the bottlenecks.

我们现在看到CPU成了新的瓶颈，而内存的需求增长最为显著。

We see CPUs as new bottleneck with memory seeing the highest content increase.

我们预计到2030年，因这一趋势新增的CPU市场规模可能高达600亿美元，占整个CPU市场超千亿美元的很大一部分；同时，高达70%的DRAM新增出货量也将与此相关。

We estimate as much as 60% or $60,000,000,000 of incremental CPU total addressable market by 2030, within a total CPU market of more than $100,000,000,000 We also estimate up to 70% of incremental DRAM bit shipment tied to this theme.

因此，我们对整个供应链更加乐观，包括内存、晶圆代工、基板、CPU和内存接口、电容以及CPU插槽等环节。

That makes us more positive on the supply chain, including memory, foundry, substrate, CPU and memory interface, and capacitors and CPU sockets.

这些领域将在2027年前持续受益于需求增长、定价能力提升和产能限制。

These areas benefit from content growth, pricing power, and capacity constraints into 2027.

随着AI从回答问题转向主动行动，投资者应该关注支撑这一转变的底层基础设施。

As AI moves from answering questions to taking actions, investors should watch the infrastructure behind the shift.

因为在代理时代，AI的下一个重大突破，可能不再取决于提示词本身，而更在于处理器的性能。

Because in the agentic era, the next big AI leap may be less about the prompt, but more about the processor.

感谢收听。

Thanks for listening.

如果你喜欢这个节目，欢迎在你听的平台给我们留个评价，也今天跟朋友或同事聊聊你对市场的看法。

If you enjoy the show, please leave us a review wherever you listen and share thoughts on the market with a friend or a colleague today.

以上内容仅供参考，基于制作时可获得的信息。

The preceding content is informational only and based on information available when created.

这不构成任何投资建议或要约，也不是税务或法律建议。

It is not an offer or solicitation, nor is it tax or legal advice.

内容未考虑您的财务状况和目标，可能并不适合您。

It does not consider your financial circumstances and objectives and may not be suitable for you.