The integration of computing networks into everyday life is no longer a distant dream. This year, AI Agents have transitioned from niche tech circles into the dAIly lives of ordinary people. tasks such as generating videos, creating spreadsheets, and ordering food delivery are all powered by computing resources. However, many users still find computing power difficult to access—either because they cannot find the right channels or are deterred by high costs.
Meanwhile, data centers are sprouting up across China. Both state-owned and private enterprises are heavily investing in infrastructure. Yet, beneath this boom lies an awkward gap: some equipment remains idle with computing power "sleeping," while businesses and indiViduals seeking affordable computing struggle to find it.
Recently, the Chinese government began systematically advancing the construction of the "Six Major Networks," which prominently includes the computing network. Backed by the Ministry of Industry and Information Technology's (MIIT) ACTion Plan for computing power interconnection, the "National Computing Power Internet Service Platform" has been launched by the China Academy of Information and Communications technology (CAICT). This platform aims to make computing power as accessible as water and electricity, allowing users to connect at a single point and use it on demand.
Advancing Computing Power Interconnection
"Currently, there is no official unified DeFinition of a computing network. Simply put, following the logic of power grids and highway networks, it connects data centers and server rooms of all sizes across the country into a single massive network," explained Deng Zhihui, an expert at the East China Branch of CAICT. The physical foundation of this network consists of Data Centers and edge nodes nationwide, with the eight national hubs and ten clusters of the "East Data West Computing" project serving as its backbone.
Discussing the purpose of this network, Deng highlighted a critical pain point: "In recent years, the AI boom led to a rush of capital into data center construction. Blind expansion resulted in low utilization and occupancy rates, while some regions faced severe computing shortages with no way to access external resources." The core goal of promoting the computing network is to optimize resource allocation.
Adhering to the principle of "connecting as much as possible," the National Computing Power Internet Service Platform prioritizes government-built computing sites. Assets built by tech giants remain under corporate ownership, and companies can voluntarily connect their idle resources without mandatory requirements. The platform is divided into four types of service providers: scheduling, resources, APPlications, and dedicated lines, with major players like Tencent and Tianyi cloud already on board.
Natural CLImate is also a crucial factor in computing layout. Deng explained that regions like Northeast China, Yunnan, and Guizhou have lower temperatures, making server room cooling more energy-efficient and yielding better Power Usage Effectiveness (PUE). In contrast, eastern regions face high cooling costs during hot summers. This is why Heilongjiang has been able to deploy tens of thousands of computing nodes.
To make this network truly functional, a "base" is not enough; the "highways" connecting these bases are equally critical. Telecom operators play a key role as comprehensive scheduling service providers. For instance, China Telecom's "Xirang" scheduling platform optimizes the distribution and matching of computing power. LeverAGIng their network infrastructure, operators act as vital bridges connecting supply and demand, with tech giants like Tencent and Alibaba also participating in scheduling.
Industry insiders revealed that local data center interconnect (DCI) ring networks are being built to integrate operators' own data centers with third-party facilities. Once completed, any two data centers can establish connections and launch businesses in the shortest possible time, ensuring on-demand connectivity.
"Previously, bandwidth between data centers was capped at 1G, but now clients frequently require 10G or even 100G," the insider noted. Traditional Ethernet architectures can no longer handle the massive demands of AI computing transmission. Operators are now introducing Wavelength Division Multiplexing (WDM) technology, multiplexing multiple wavelengths within a single fiber to support dozens of 200G ultra-high-capacity channels.
According to the "Six Major Networks" plan, by the end of 2026, major domestic computing centers will be fully interconnected. For ordinary users, the platform supports individual real-name registration, allowing them to rent affordable inference computing power for hundreds of yuan to enjoy AI generation and smart Q&A at home. Computing power is no longer an exclusive resource for large enterprises.
Will token pricing Be Unified?
With the expansion of the computing network, will computing power be priced uniformly nationwide, much like electricity?
"In the future, Token Pricing will not be unified, as vendors cannot apply a one-size-fits-all price. However, token measurement rules will be Standardized," Deng concluded.
Currently, computing usage is billed through various models, including high-end monthly server rentals, cloud on-demand billing, and private enterprise deployments. Vast differences in hardware costs and AI model capabilities mean there is no bASIs for uniform Token pricing. Billing rules vary significantly: some vendors equate one Token to one Chinese character, while others equate it to two, making it confusing for ordinary users.
"The future focus will be on standardizing the measurement criteria for Token-to-character conveRSIon, closing loopholes that allow arbitrary manipulation of statistical metrics," Deng stated.
In May, the three major telecom operators simultaneously launched Token packages. CAICT has also explicitly proposed that tokens are expected to become the unified value benchmark and trading medium for the computing internet. As standards unify, the most direct benefit for users will be transparent Token Consumption. For C-end users, basic tasks like writing and Image Generation can be handled by mid-to-low-end computing power for just tens of yuan a month. For B-end users, high-end computing power costing tens of thousands of yuan monthly will be available for enterprise model training.
Challenges in Collaborative Implementation
As the computing network rolls out, Data Security and computing-power-electricity coordination are two unavoidable challenges. Deng Zhihui and Jing Yiliang, founder of LUXstarlight, shared insights on the current industry landscape and breakthrough directions.
Addressing SECurity concerns regarding cross-network calls for idle computing power, Deng frankly admitted that highly sensitive industries like finance and healthcare fear data breaches and typically build private server rooms rather than joining public networks. SMEs with limited budgets must rely on public idle computing power, fortifying their defenses through three measures: tenant network isolation by cloud vendors, privacy computing and blockchain encryption, and enhanced industry-wide confidentiality self-discipline. Small, scattered edge server rooms with rudimentary security remain a risk, but continuous technological advancements are gradually resolving these issues.
Computing-power-electricity coordination is the key lever to making computing "affordable." Under multiple pressures, the construction of the computing network cannot happen overnight. However, as security technologies iterate and power market reforms deepen, the supply of affordable idle computing power continues to expand, steadily achieving the goals of being "user-friendly and affordable."
From isolated server rooms to a nationwide computing network, China's computing infrastructure is advancing steadily, driven by both policy guidance and market forces. This transformation is turning scarce computing resources into ubiquitous digital utilities, bringing them into the homes of ordinary citizens.
Comments & Questions (0)
No comments yet
Be the first to comment!