"Cities are never just a combination of steel, concrete, and roads; they are dynamic entities that continuously perceive, flow, collaborate, and grow."

This was our direct impression at the Huawei Cloud INSPIRE Innovators Conference.

Several years ago, discussions about smart cities mainly focused on topics such as government services migrating to the cloud, one-stop online government services, city brains, and data platforms. At the 'Advanced Public Cloud Industry Summit,' the most frequently mentioned topics were large models, AI-native approaches, silicon-based black land, and urban super-intelligent agents.

As mentioned by Zhou Yuefeng, Huawei's Board Director and CEO of Huawei Cloud, in his speech: 'AI technology is rapidly evolving, with model parameter sizes surging. Public clouds, with their intensive construction advantages, are highly aligned with the current trend of large-scale and efficient development. Public clouds are not just a single form of public cloud but encompass diverse models such as enterprise alliance sharing and government exclusivity. They represent the best infrastructure for small and medium-sized enterprises and research institutions to access advanced computing power.'

In the context of smart cities, AI CITY does not emerge out of thin air. It requires a continuous supply of intelligence through a 'silicon-based black land,' enabling computing power, data, models, applications, and security to become public capabilities of the city, just like 'water and electricity.'

Huawei Cloud has raised the question and provided the answer.

Reviewing the evolution of smart cities, it can be summarized in three words: connection, convergence, and visualization.

During the informatization phase, the focus was on bringing business processes online, with numerous government processes, public services, and departmental systems moving from offline to online.

In the digitalization phase, the focus shifted to data convergence and collaboration, with data dispersed across different departments and systems gradually being interconnected.

In the early stages of intelligence, situation awareness and decision support were achieved through algorithms and models. For example, the 'city brain' enables managers to understand real-time events in the city, such as traffic congestion, environmental changes, emergency risks, and public services.

The evolution of smart cities is far from over: they can 'identify problems' but may not automatically understand them; they can display status but may not proactively break down tasks; they can support decision-making but may not collaborate across systems to execute actions.

The year 2026 is set to be a historic turning point for smart cities.

Large models have endowed AI with understanding and generation capabilities, while agents have further equipped AI with task execution abilities. Agents can not only answer questions but also understand intentions, plan paths, call tools, access data, collaborate across systems, and autonomously complete a series of complex tasks.

In the current stage of smart cities, humans are needed to identify problems, determine causes, assign tasks, and follow up on results. In contrast, AI CITY enabled by large models and agents allows humans to set goals while intelligent agents autonomously analyze, decide, and execute.

A direct example is Karamay's urban super-intelligent agent.

Rewinding to 2012, Huawei's first global government cloud landed in Karamay, marking the beginning of the industrial city's digital and intelligent transformation journey. Over 13 years, Karamay has completed three stages of transformation: digitalization, digital intelligence, and digital governance. Through the full-stack capabilities of the 'urban super-intelligent agent,' including perception and interaction, cognition and decision-making, and autonomy and evolution, Karamay has upgraded from 'problem identification' to 'problem-solving.'

For example, in developing government applications, the traditional model took 2-3 months from project initiation to delivery. Relying on Huawei Cloud's CodeArts code intelligence agent and full-stack government cloud services, the development cycle has been reduced to just half a month. Currently, Karamay has implemented multiple mature government intelligent agents, significantly improving the city's comprehensive operational efficiency and delivering impressive results in people's livelihood services, economic development, and ecological construction.

Another example is the deployment of intelligent agents. Ecological partners can seamlessly integrate through the A2A protocol, fostering a new ecosystem of open and growing intelligent agents in Karamay.

Comparing it to the 'city brain': after an emergency event occurs, the city brain can only present data, while the urban super-intelligent agent can further understand the nature of the event, correlate historical data, determine involved departments, break down tasks, dispatch relevant agents, and precipitate (chen4 dian4: accumulate) disposal (chu3 zhi4: handle) results for reuse in similar future scenarios.

While many cities are still troubled by issues such as inconsistent technological frameworks, severe data silos, difficult coordination between computing power and algorithms, and poor adaptation of scenario implementations, why has Karamay, known as the 'Gobi Oil City,' delivered a visible digital governance achievement?

The answer is not difficult to explain: Karamay is the first city to implement 'AI CITY 2.0.'

In the definition of Token Economics, the most crucial aspect is 'how to efficiently produce tokens.' For a city, the prerequisite for intelligent transformation across all industries is how to connect, use, and operate AI at lower costs and higher efficiencies.

At the Huawei Cloud INSPIRE Innovators Conference, the new Agentic Infra paradigm was officially introduced: 'Efficient Token Factory + Continuous Learning + Integrated General and Intelligent Scheduling + Secure Autonomy.' Simultaneously, a series of new Agentic Infra products were released, including the AICS Lingqu Intelligent Computing Cluster, AMS Agentic Memory Storage Solution, CCE VolcanoNext Integrated General and Intelligent Scheduling Engine, and the secure and autonomous Agent operating environment, AgentSphere, to help industries build a 'silicon-based black land' with coordinated software, hardware, and chips.

Among them, in the field of smart cities, the new Agentic Infra paradigm has profoundly influenced the construction of urban AI cloud services.

Hu Yuhai, President of Huawei Public Cloud, responded by stating that Huawei Cloud, with 'AI CITY' as its core concept, has jointly built a full-stack solution capability from AI infrastructure to intelligent agent applications with ecological partners. It has released the 1+2+3+N AI CITY reference architecture to help cities achieve comprehensive upgrades in 'good governance, people's livelihood benefits, and industry prosperity.'

'1' represents a set of AI infrastructure.

AI's empowerment of cities has rapidly extended from traditional government services to public services. Hu Yuhai specifically mentioned the deployment model of 'one cloud, two domains,' which coordinates resource pools in the government domain and public domain to ensure unified scheduling and efficient collaboration of city-level resources.

'2' represents two AI innovation engines.

Model as a Service lowers the barriers to model invocation, adaptation, and continuous iteration. Data as a Service transforms data dispersed across different departments, systems, and levels into callable, trainable, and serviceable city assets after governance.

'3' represents three AI enablement platforms.

AgentArts, for developing and running enterprise-level and government-level intelligent agents; CodeArts, enabling application development to transition from manual coding to AI-assisted and automated closed-loop development; and DataArts, transforming urban data from precipitate (chen4 dian4: accumulated) resources into governable, callable, and decision-making assets, helping various departments quickly implement intelligent agent applications for segmentation (xi4 fen1: segmented) scenarios.

'N' represents N AI intelligent scenarios.

Huawei Cloud, in collaboration with ecological partners, has accumulated 'N out-of-the-box' best practices, including expert capabilities in government, people's livelihood, and industry sectors, as well as replicable processes, tools, and templates. This enables cities to avoid starting from scratch and quickly replicate, deploy, and achieve results in scenarios such as government services, urban governance, people's livelihood services, and industrial development.

Additionally, a full-stack proactive defense system is covered, including role-based authorization, identifying malicious tasks, blocking destructive behaviors, and full-link monitoring of agent malicious behaviors, achieving a transition from passive 'unplugging to stop losses' to proactive 'prevention and controllability,' ensuring the secure and reliable operation of intelligent agent applications.

To draw an analogy, the '1+2+3+N' AI CITY 2.0 reference architecture is like a city-level 'AI operating system.' It addresses a series of issues hindering AI implementation at the bottom layer (di3 ceng2: underlying) level, such as computing power, data, and scenarios, enabling every city to build an AI CITY based on its unique strengths and continuously inject new momentum into industrial upgrading, people's livelihood improvement, and urban governance.

The rise and fall of a city hinge on its ability to enhance governance efficiency, service capabilities, and industrial vitality.

The biggest difference between AI CITY and traditional smart cities is that AI CITY does not focus on functional upgrades around individual systems or treat AI as an add-on tool. Instead, it organizes AI infrastructure, data services, model capabilities, intelligent agent platforms, and scenario practices, allowing AI to truly enter the mainstream of urban operations and transform into tangible new quality productive forces.

The case of Karamay mentioned earlier is not an isolated example. More and more cities are exploring diverse paths based on the AI CITY 2.0 reference architecture.

In Shenzhen, Guangdong, AI has become the best epitome of benefiting the people.

Based on the successful practice of the 'Citizen Cloud,' Longgang District in Shenzhen has comprehensively upgraded digital people's livelihood services, providing each household with 4T of exclusive digital living space and gradually implementing personalized AI applications such as 'personal doctors,' 'study companions,' and 'office assistants.' This supports Longgang's transition from 'digital empowerment of life' to 'digital is life,' ultimately ensuring that 4.8 million residents can step into the new digital world equally, calmly, and with a sense of gain.

In government applications, the Shen Xiaoi government AI assistant has comb, sort out, organize, arrange, streamline (shu1 li3: organized) knowledge graphs of over 2 million words in six high-frequency areas such as business establishment. By innovating a 'general-specialized combination' multi-agent collaborative elastic scheduling technology, it has achieved a 97.7% response rate and a 94.3% accuracy rate for government questions, becoming an 'efficient and reliable' government intelligent customer service.

In Wuhan, Hubei, AI is changing the lives of every citizen.

As early as 2021, Wuhan partnered with Huawei to build the nation's first 'urban cloud,' providing a 'black land' for intelligent transformation across various industries.

Taking smart healthcare as an example, by constructing a municipal imaging data sharing center, 'Wuhan Cloud' has achieved cloud-based convergence and cross-institutional sharing of imaging data, greatly optimizing patients' healthcare experiences. As of early April 2026, through city-wide imaging data access and mutual recognition, Wuhan's medical insurance imaging cloud is expected to help citizens save approximately 1.2 billion yuan in inspection costs annually.

In Wuxi, Jiangsu, AI exploration is delving into the core of industries.

When surge (peng2 pai4: surging) AI computing power is readily accessible, AI CITY is not just a carrier for AI applications but can also serve as an 'incubator' for industrial intelligence.

In this manufacturing hub, Wuxi focuses on the potential sector of 'industrial embodied intelligence,' building a 'city of data and scenarios.' It pioneered the 'data spectrum theory,' systematically conducting large-scale collection of three types of 'data crude oil': real machine, simulation, and human-like data, covering over 300 industrial scenarios.

Relying on Huawei's Cloudrobo platform, Wuxi has built its data pipeline, completing the entire process of model infrastructure from 'crude oil extraction' to 'refining.' Simultaneously, it innovated a city partner mechanism to promote data rights confirmation, assetization, and capital closure, establishing the Wuxi paradigm of 'data-driven production, scenario-nurtured production, and finance-catalyzed production,' solidifying the digital foundation for China's industrial embodied intelligence.",