Agent Is Ending the Cloud Computing “Assembly Line,” Infra Must Learn to “Think” | Interview with WuWen CoreSpace Xia Lixue

Introduction

Edited by: Luo Yanshan, Tina

Organized by: Yu Qi

A new era built around intelligent agents is accelerating. Infrastructure is evolving from AI Infra to Agent Infra, and now towards Agentic Infra — becoming the key driving force for large-scale agent deployment.

As China’s leading AI infrastructure service provider, Wuwen Xinqiong is tackling challenges in agent collaboration, security, and continuous learning — transforming intelligent agents into real productivity.

At the QCon Global Software Development Conference (Shanghai, Oct 23–25), InfoQ interviewed Xia Lixue, co‑founder and CEO of Wuwen Xinqiong. He shared in detail how they build agent-oriented infrastructure and explained how Agentic Infra supports industrial-scale agent deployment.

---

Key Highlights

• China’s computing power resources are diverse and geographically dispersed. To move toward the AGI era, infrastructure must allow agents to use computing power as easily as water or electricity.
• The first step is building solid Agent Infra — evolving infrastructure from a “factory” into a solution provider, ensuring task quality via environment, tools, context, and security.
• The next step is Agentic Infra, enabling agents and infrastructure to gain greater autonomy — achieving more efficient resource integration and higher-value innovation.
• Smarter infrastructure creates new demands, driving R&D upgrades and forming a virtuous cycle between technology and applications.

---

From “Processing” to “Thinking”

Continuous vs. Discrete Tasks

InfoQ: Traditional cloud computing follows a deterministic request–response model, but agents use a non-linear, stateful loop of perception–reasoning–action–memory. Where does this most disrupt infrastructure?

Xia Lixue:

Agent tasks are continuous and interconnected — not discrete. Like an outsourcing factory, older infrastructure just receives instructions and executes. But agents require intelligent infrastructure to coordinate task systems carrying “state.”

Infrastructure must evolve from:

• Factory production lines
• → Intelligent solution companies

Infrastructure Upgrade Needs

Practical upgrades include:

• Runtime Environment Adaptation — flexible environments matching agents’ execution modes.
• Comprehensive Tool Sets — enabling full resource invocation.
• Context Provision — precise, sufficient context to ensure task consistency.
• Security & Monitoring — controllable and observable task flows.

Key: Multi-task collaboration, sandbox environments, flexible scheduling — essential for complex task chains.

---

Resource Flexibility — “Thinking Longer”

InfoQ: Is the shift about computing faster vs. thinking longer?

Xia Lixue: Yes — and thinking requires different resources than computing. This demands flexible allocation strategies and real-time scheduling to fit dynamic workloads. Infrastructure design must adapt.

---

Scaling Obstacles for Agents

Demand Fluctuation & User Expectations

Large drops in users (e.g., Lovable platform) are natural adoption cycles. But scaling agents faces obstacles:

• Infrastructure maturity
• Interoperability
• Sustained cost efficiency

Many users expect "no-code" to mean instant full apps via natural language — reality requires iterations and skills. Frustration comes from high barriers and uncertainty.

Core Bottleneck

The biggest bottleneck is not model capability but immature infrastructure and tools.

Example: An agent might try calling diverse libraries but lacks permissions or interfaces — forcing guesswork.

---

Controllability and Observability

To fix this:

• Transparent process management — show decision paths at each step.
• Stronger tool support — proper access, permissions, and APIs.

---

Agent Infra vs. Agentic Infra

Stages:

• Agent Infra
• Transforms lab demos into production tools.
• Supports algorithmic capabilities and basic scalability.
• Agentic Infra
• Designed for large-scale, next-gen AI evolution.
• Enhances environment, context, tools, security, observability.
• Agents participate in core infra workflows.
• Enables anomaly detection, bottleneck fixing, elastic scaling.

Goal: Shift from Agents as tools → Agents as collaborators.

---

Timing for Agentic Infra

Xia Lixue: Models are already "smart enough" — but lack environments to unleash full potential. Now is the right time to invest in Agentic Infra.

---

Supporting A2A (Agent-to-Agent)

Traditional workflows are human-designed. Agentic Infra aims to let agents design their own workflows:

• Backend Ops Agents + Frontend Service Agents
• Efficient cooperation surpassing human handovers

We are in early A2A. Infrastructure must support agent swarms with real collaboration capabilities.

---

Autonomous Infrastructure

Once Infra embeds Agent capabilities, it gains cross-domain autonomy — enabling:

• Cross-boundary resource integration
• More valuable function innovation

---

Agent-Native Architecture Outlook (3–5 years)

Future: Agents form organizations to complete complex tasks, able to share or isolate KV caches and contexts as needed, freeing human productivity for creative work.

---

Efficiency Challenges in Compute Paradigm Shift

Traditional Model Inefficiencies

• Coarse-grained containers/VMs
• Slow cold starts
• Locked resources during scheduling

Solution Architecture

Micro-virtualization sandboxes with:

• Dynamic capacity assembly
• Millisecond environment switching
• Near 100% utilization

---

Unified Scheduling of Heterogeneous Computing Power

Core innovation: Standardization system that bridges multiple architectures while ensuring efficient allocation. In China’s diverse ecosystem, computing should be as accessible as water/electricity.

---

From Technical Adaptation to Industrial Integration

Initial challenge: bridging architectures (libraries, ops). Once done:

• Easier industrial integration
• Users only perceive performance differences, not hardware differences

---

AI-Native Infra — Balancing Advancement & Service

Wuwen Xinqiong focuses on building AI-Native Infra to let customers iterate products while infra manages computing and resource complexity.

---

Elastic Scaling Sandbox System

Features:

• Millisecond-level start/kill of sandboxes
• Dynamic resource allocation by task type
• Intelligent CPU/GPU/domestic chip mounting
• Elastic expansion during peaks, auto recovery during low-load

Result:

High cluster resource utilization and a positive feedback loop between infra innovation and application needs.

---

Vision and Mission

Personal Core Objective:

Witness the AGI era, contribute to making AGI efficient and sustainable via system-level innovation.

Company Vision:

"Unleash infinite computing power, make AGI within reach."

---

Connecting Agentic Infra to Global AI Creativity

Many creators use AiToEarn官网 — an open-source global AI content monetization platform — to:

• Generate, publish, and monetize content via AI
• Distribute across Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X
• Track performance via unified analytics and model ranking

Integration Potential:

Agentic Infra could link autonomous agent swarms directly with such ecosystems, enabling automated, scalable content creation and deployment at unmatched efficiency.

---

Summary:

Wuwen Xinqiong is positioning Agentic Infra as the next critical evolution in AI infrastructure — creating autonomous, collaborative, self-optimizing systems that bridge intelligent agent capabilities with real-world deployment at scale. This solid foundation supports both industrial AGI progression and global AI creativity ecosystems.