Under the wave of always-on Agents, Chinese ICT infrastructure vendors are the first to secure orders

By Alter

In the early hours of October 22, Anthropic, in the middle of its IPO sprint, released a brand-new personal and work agent application, Claude Duet.

In the month or so before that, Meta's Muse, OpenAI's Dots, xAI's Grok Bot, and Manus Cue, built by a Chinese startup team, launched one after another, delivering their answers at almost the same time: give AI an independent virtual computer in the cloud.

In a demo Anthropic made public, an interesting scenario was shown:

Every morning at 5:30, before the operator of a coffee chain gets out of bed, Duet has already automatically pulled the latest data from the POS system, the Shopify online store, the customer service inbox, the Gusto payroll system, and bank accounts, generated a morning business briefing, and pointed out that in the East Bay area, due to a shipping fee adjustment, the cancellation rate rose from 2.6% to 3.4%.

The debut of Claude Duet quickly sparked a new round of discussion in the tech world: why hasn't China produced a cross-platform "Chinese version of Duet"?

If we shift our gaze away from the battle for entry points at the application layer and look at the more foundational industry chain, a different picture emerges—the first to feel the impact of the always-on Agent wave and the first to realize commercial value are not the AI companies in the spotlight, but the infrastructure layer beneath the surface.

An architect at a leading Chinese cloud vendor admitted that when the product form of AI changes from "a chat box of question and answer" to "a 24-hour online cloud virtual machine," the ratios of compute, storage, and networking in data centers all have to be recalculated from scratch.

01 From chat box to cloud computer

2023 was the Chatbot stage: ChatGPT defined question-and-answer interaction, the model was responsible for generating text or code, and the actual execution of actions was still done by humans.

2024 to 2025 entered the task-based Agent stage: the first-generation Manus and Claude Code proved that AI could break down multi-step tasks and deliver results, but interaction was still triggered one time only, and the session was cleared once the task ended.

The turning point came in early 2026.

The open-source framework OpenClaw (known in Chinese developer circles as "xiaolongxia" because its icon and name sound alike) gained 240,000 GitHub stars in three months, and major Chinese companies such as Tencent, ByteDance, and Alibaba quickly followed suit, building AI assistants that "live" on users' computers.

Agents no longer occupy the user's local computer; instead, they have an independent runtime environment in the cloud (Meta even allocates a dedicated virtual machine to each Muse user), and they generally introduce a "proactive inspection" mechanism—without waiting for the user to issue instructions, the system automatically wakes up in the background every few hours to check email, calendars, and business systems.

For users, it is an upgrade in experience from "passive tool" to "proactive butler"; for the underlying compute infrastructure, it means the load has fundamentally changed.

Traditional Chatbots are stateless: when a user asks a question, the GPU is invoked once, and the connection is disconnected when the answer ends, so nighttime traffic naturally falls back. Always-on Agents are stateful; taking Meta Muse as an example, multiple "worker sub-agents" operate in parallel underneath, along with an independent "Sentinel review agent" monitoring permissions and security in the background.

This change will directly push up the cost of serving a single user.

On September 30, the team of Brian Nowak, Morgan Stanley's chief internet analyst, published a unit economics model estimate for Muse: under a baseline scenario using only 25% of the weekly free Token quota, a single active user consumes about 108 million Tokens per month, corresponding to $37.12 per month in direct hard service spending—of which $24.30 is GPU inference cost and $12.82 is CPU sandbox environment and storage overhead.

Among heavy developer users, the numbers are even more astonishing.

OpenAI CEO Sam Altman publicly revealed in June that the engineer with the highest internal usage at OpenAI consumed 100 billion Tokens in a single month. When always-on Agents can work 7×24 hours in the cloud, 100 billion is by no means the upper limit of monthly Token consumption.

Stretching this timeline out, one easily overlooked detail becomes apparent: every leap in product form is accompanied by a transfer of the compute bill.

In the Chatbot era, bills were settled per use; in the "xiaolongxia" era, agents ran on users' own computers, and electricity costs and hardware depreciation were silently borne by users; by the cloud always-on stage, the bill has been moved back in its entirety onto the balance sheets of AI companies and cloud vendors.

The competitive variable for always-on Agents is shifting from "how smart the model is" to "how much it costs per month to support one active user." So-called "proactive inspection" is essentially a cost knob: the more frequent the inspection, the more butler-like the experience, and the higher the per-user cost. Whoever can find a sustainable balance between experience and cost may be able to keep this business going.

And before this question is answered, the first to be certain of getting paid are the people selling shovels.

When massive numbers of agents run continuously in the cloud, the first thing to run short in data centers is not the GPU that has been so sought after for the past three years.

02 CPU returns to the center of the table

For the past three years, the protagonist of the AI compute market has always been the GPU.

In traditional large-model training and single-inference servers, the CPU-to-GPU ratio is usually between 1:4 and 1:8. GPUs handle high-density parallel matrix operations, while CPUs only handle tasks such as data movement and basic scheduling.

Resident agents represented by Meta Muse are breaking the existing ratio.

After an agent receives a task, it needs to go through multiple steps such as intent understanding, task decomposition, spinning up a secure sandbox, executing code, operating a browser, calling external APIs, verifying results, and reflecting and retrying. Apart from a small portion of "model inference" completed by the GPU, the vast majority of process control, tool invocation, and environment isolation are typical serial logic operations, handled by the CPU.

The spread of cloud virtual machines will further drive up demand for CPU cores.

For example, Meta Muse provisions each user in the cloud with a standard Linux virtual machine with 2 vCPUs, 8GB of memory, and a 100GB SSD.

Overseas hardware media outlet Wccftech once made an estimate based on static peak values: if Muse expands to 100 million users and each person exclusively occupies physical cores without sharing, calculated on the basis of a single 126-core AMD EPYC processor, this one application alone would require 1.58 million server CPUs.

Of course, in actual engineering, cloud providers will not let virtual machines sit idle 24 hours a day.

But even after accounting for hibernation mechanisms and multiple oversubscription ratios, when tens of millions of active users concurrently spin up multiple temporary sub-agent sandboxes during peak periods, data center demand for independent "Agent CPU racks" will still grow significantly.

Bernstein judged in a research report that the CPU share of compute in agent workloads will rise from 14% in traditional large models to 50%, that the CPU-to-GPU ratio is converging from 1:8 toward 1:1, and raised its forecast for the global server CPU market size in 2030 to $223 billion.

The change in demand has already been transmitted to the supply chain.

KeyBanc data shows that, affected by concentrated procurement by large cloud service providers, Intel's and AMD's server CPU capacity for all of 2026 is basically sold out; both companies plan to raise server CPU prices by 10% to 15%, delivery lead times have lengthened from the usual 1–2 weeks to 8–12 weeks, and some models have risen by more than 40% since the beginning of the year.

Faced with high CPU costs and high-concurrency sandbox demand, domestic cloud and server vendors have accelerated underlying architecture upgrades in 2026:

Alibaba Cloud launched the ACS Agent Sandbox based on micro-VM isolation, focusing on optimizing large-scale low-latency elastic spin-up and low-cost state retention.

Tencent Cloud open-sourced Cube Sandbox v0.5.0 in July, and together with the Arm engineering team completed native mainline support for the Arm architecture, enabling the Arm platform to match x86 in sandbox creation speed, state recovery time, and per-machine concurrency density.

H3C proposed a sandbox image pool solution based on CXL memory pooling, pre-residing commonly used operating systems and dependency libraries in a shared memory layer, so that when a new sandbox is spun up it does not need to read from disk and can achieve zero-copy startup directly through memory mapping, compressing sandbox cold-start time from seconds to milliseconds.

Against the backdrop of rising CPU prices and lengthening lead times, how many sandboxes a single machine can pack in and how quickly they can be woken after hibernation directly determine per-user cost. Industrial value is migrating from "selling more cores" to "making every core run more agents."

03 After VRAM is blown out

Along with the CPU, another area undergoing value reassessment is the storage system that has long sat at the edge of the data center.

In traditional cloud computing architectures, storage is mainly responsible for statically preserving data, and millisecond-level read/write latency is enough to meet business needs.

But in long-horizon agent scenarios, storage directly enters the main compute path, and the core bottleneck becomes KV Cache.

Just as in the Claude Duet demo scenario, to help a user prepare for a Series A financing roadshow, the agent needs to continuously read 25 data room documents and background materials on 6 investment partners, and perform dozens of rounds of tool calls.

As model context windows expand to the million level and multi-round slow thinking becomes the norm, KV Cache capacity demand under a single task chain has surged dozens of times.

GPU high-bandwidth memory (HBM), which has limited capacity and a high price, is quickly filled up, and the overflowing warm KV Cache and long-term memory data have to sink down to the next tier: DRAM memory, CXL memory pools, and enterprise-grade solid-state drives (NVMe SSDs).

In addition, hundreds of millions of cloud sandboxes, when performing "snapshot hibernation and wake-up recovery," also need high-frequency read/write access to their own independent file spaces.

Wu Junjie, vice president of Huawei's data storage product line, said in a media interview that when agents access context 24/7 without interruption, the latency requirements for underlying storage have risen from the previous millisecond level to the microsecond level—any bit of delay in data return will cause the GPU to wait idle; if the context is forcibly truncated due to insufficient cache space, a large amount of repeated computation must be performed.

The tightness of supply and demand has already been reflected in prices and financial reports.

Domestic DRAM maker ChangXin Technology said in its performance forecast for the first three quarters of 2026 that, driven by AI compute demand, DRAM products are in short supply and prices have risen sharply, and the company's capacity utilization remains high; storage module leader Longsys expects to achieve revenue of 38 billion to 40 billion yuan in the first three quarters of 2026 (up 127% to 139% year on year).

To alleviate the "VRAM wall" and the storage isolation challenge of hundreds of millions of sandboxes, major domestic technology companies are also advancing storage architecture reconstruction.

Huawei launched CMS (Context Memory Storage) supporting heterogeneous computing power and The UCM (Unified Cache Management) solution builds a PB-scale shared KV cache pool through dedicated DPU semantic offloading, reducing inference first-token latency by up to 90% and cutting the comprehensive per-token cost by about 30%.

In July, Tencent Cloud released the industry's first native storage Agent Bucket for agent scenarios. By introducing an independent "Space" unit under the bucket, it moves a single Agent's credential isolation, capacity quotas, and read/write rate limiting down from the business code layer to the system layer, working with GooseFS distributed caching to handle high-frequency small-file access.

If CPU demand depends on whether consumer-grade Agents can make their business models work, the logic for storage is more solid. Long context and multi-turn reasoning are not unique to resident Agents; enterprise Agents, coding Agents, and even ordinary deep reasoning tasks are all driving up the scale of KV Cache. Even if growth slows for a certain star application, storage demand will not recede.

A deeper change lies in storage "converting capacity into compute": if the cache can hold data, the GPU does less repeated computation; if it cannot, more expensive GPU time must make up for it. Storage thus shifts from a cost item to an efficiency item, and its say in AI infrastructure keeps rising.

04 The inverted network funnel

Beyond compute and storage, communications and data center networks are also undergoing a reversal in traffic structure.

For the past two decades, the internet has been designed mainly around "people consuming content." When users browse web pages, watch videos, and download apps, data flows mainly from data centers down to terminals, and in traditional scenarios uplink traffic typically accounts for only about 10% of network traffic.

In Agent scenarios, the traffic structure is inverted.

Whether a cloud Agent continuously reads local workflows, or a user transmits high-definition screenshots, audio and video streams, and documents in real time from a phone or smart glasses to an Agent, the prerequisite for AI to execute tasks is that terminals upload context to the cloud at high frequency.

For example, at MWC 2026 earlier this year, Huawei disclosed measured data from an art museum in Shanghai: after introducing a multimodal AI agent as a guide, more than 3,000 visitors per day uploaded photos or videos of exhibits for interaction, and uplink traffic exceeded 60% for the first time.

Inside data centers and in cross-cloud collaboration scenarios, machine-to-machine (M2M) communication density is also climbing in tandem.

A typical example: when Claude Duet generates a business briefing in the background, it needs to call the APIs of more than ten external SaaS systems in succession through the MCP protocol within a few minutes; and inside the data center, separately deployed "Agent CPU racks," "GPU inference racks," and "CXL storage racks" need constant high-bandwidth, low-latency state synchronization.

Facing expectations of hundreds of billions of Agent connections, the domestic communications and network industry chain has long begun adjusting its layout:

On the operator side, China Mobile is exploring adjusting the uplink-downlink traffic ratio to 3:2 in the 4.9GHz band to match Agent uplink demand, and has proposed the lossless intelligent computing optical network (HIC-OTN) architecture; China Unicom in September upgraded its traditional CDN to AICDN, enabling coordinated scheduling of AI Token compute and edge content.

On the hardware and edge services side, data center demand for 800G and 1.6T high-speed optical modules continues to expand, and TrendForce estimates that the global AI optical transceiver module market will reach $26 billion in 2026.

It should be noted that although the direction of change on the network side is clear, its realization pace is likely to lag behind compute and storage. Adjusting spectrum ratios and building lossless optical networks take years, and investment must precede demand; moreover, current empirical evidence of the uplink traffic reversal still comes mainly from tests in individual scenarios.

The real focus is the role of operators.

For a long time, networks were designed around downlink bandwidth and charged by content distribution; if Agents make uplink the main traffic, "uplink" itself may become a new billing dimension. Operators may not be the first to benefit, but they may hold the longest-term pricing leverage.

05 Closing remarks

Back to the question at the beginning.

In the short term, the Chinese market will probably find it hard to produce a cross-platform "Chinese version of Duet"; the crux is not technology but ecosystem boundaries: as of late October, more than a hundred apps, including WeChat, Taobao, Alipay, and Meituan, had refused screen automation operations by the Doubao phone assistant, and cross-App agency in China is first and foremost a commercial question.

This creates an intriguing mismatch:

The battle for the application-layer entrance is temporarily deadlocked in China, while the CPU expansion, storage upgrades, and compute network reconstruction driven by the global Agent wave have already allowed infrastructure vendors in China's ICT industry chain to secure orders first.

But the quality of the dividend ultimately still depends on the application layer: only when resident Agents find a business model that can "afford users" will today's expansion not become tomorrow's excess.