In November 2025, the official website of the ICPC Beijing headquarters released minutes from a symposium between Ren Zhengfei and ICPC leadership, coaches, and award-winning contestants. At the meeting, Ren made a seemingly simple yet incisive remark: “Computing power without a network is an information island.” He emphasized that AI’s real value lies overwhelmingly in application rather than invention—2% versus 98%—a ratio that reflects both the technological essence of AI and the realities of global industrial competition. His observation cuts through the prevailing narrative of a “computing-power arms race” often amplified in the context of intensifying U.S.–China tech rivalry. In an era obsessed with GPU counts, model size, and the race toward AGI, Ren reminds us that computing power alone is insufficient: without connectivity, coordination, and integration into real-world scenarios, even the most formidable hardware remains an isolated and costly “digital bonsai.”
The “Computing Power Island” Dilemma: Beyond Chips—Huawei’s Asymmetric Leap in Interconnects vs. NVIDIA’s Ecosystem Supremacy
China’s rapid advances in domestically developed computing chips over the past three years—exemplified by Huawei’s Ascend 910B, Cambricon’s Siyuan series, and Moore Threads’ GPUs—have been widely viewed as milestones on the path toward semiconductor self-reliance. Yet these achievements conceal a deeper structural dilemma: without an equally capable high-speed interconnect and deployment ecosystem, computing hardware becomes an isolated “island.” Self-developed chips alone cannot deliver system-level autonomy.
This dilemma is starkly visible in the training of large-scale AI models. Teams that rely on clusters built with domestic accelerators often face severe interconnect bottlenecks. In the absence of NVIDIA-class technologies such as NVLink and NVSwitch—or equivalent ultra-low-latency optical switching—the practical efficiency of these clusters may be only a fraction of that achieved by DGX SuperPOD systems. Longer training cycles, lower utilization, and elevated operational costs are common outcomes. The result is not merely slower innovation but a structural disadvantage at the foundation of model development.
An equally challenging problem emerges on the deployment side. Even when a model has been successfully trained, its transition from cloud to factory floor is impeded by entrenched industrial ecosystems. Production lines still rely on real-time Ethernet protocols such as Siemens Profinet, Rockwell EtherNet/IP, and Beckhoff EtherCAT—protocols with stringent requirements for determinism, reliability, and long-term interoperability. Domestic AI platforms and edge devices often lack deep integration with these systems, leading to practical incompatibility with PLCs, CNCs, and other industrial controllers. The gap here is not only technical but institutional: factories cannot risk introducing AI components that lack proven stability or fail to “speak the language” of their existing control infrastructure.
Taken together, these training-side and deployment-side constraints reveal two critical Achilles’ heels in China’s AI development: insufficient high-end compute infrastructure and underdeveloped industrial integration pathways. The result is a recurring pattern in which advanced models remain confined to laboratories or cloud environments while real production lines continue to operate on decades-old architectures. This misalignment between technology creation and technology absorption slows the translation of research breakthroughs into industrial value.
Huawei’s Ascend platform offers an illustrative contrast. While still trailing NVIDIA’s leading GPUs and NPUs in raw compute performance, Ascend has pursued asymmetric innovation in high-bandwidth interconnects and optical communication. With technologies such as HCCS(Huawei Collective Communication Service) and 800G LPO(Linear Drive Pluggable Optics), the ecosystem has already achieved rapid short-term cluster efficiency improvements. In the medium term, Huawei’s strength in optical modules positions it to accelerate the adoption of CPO(Co-Packaged Optics) and optical switching. Over the long term, its exploration of optical-computing integration aims to define an alternative architecture for next-generation AI infrastructure. This strategic emphasis on “the network as the nervous system” acknowledges that computing strength alone is insufficient without equally powerful pathways for moving and synchronizing data.
Ultimately, the lesson is clear: chips may serve as the muscles of an AI system, but the interconnect and industrial communication layers form the nerves that coordinate and deliver their strength. Only when both advance in concert can China move from isolated computing power to true system-level self-reliance.
5G-R and 6G: China’s Shift—From Consumer Connectivity to Industrial Nervous System
China’s new wave of 5G-R and 6G development signals a structural shift in how the country conceives of connectivity—not as a consumer convenience, but as the operational backbone of industrial intelligence. The commercial launch of China State Railway Group’s 5G-R pilot at the end of 2024 illustrates this shift clearly. Designed to support real-time dispatching, axle and wheel monitoring, and track inspection for trains traveling at 450 km/h, 5G-R is emerging as the nervous system through which AI instructions reach the physical world with millisecond precision. As Ren Zhengfei noted, 5G-R’s radar-enhanced sensing layer continuously interrogates the safety of railway lines, while the broader railway IT system—exemplified by the high-concurrency 12306 ticketing and dispatching platform—reveals how China has built AI-ready infrastructure through engineering discipline, not speculative ambition.
This approach reflects a strategic orientation: China prioritizes scenario-driven automation over a race to abstract general intelligence. Instead of pursuing AGI for its own sake, it focuses on achieving fully automated scheduling of 30,000-ton heavy-haul trains or enabling nation-scale real-time logistics coordination. Connectivity becomes the organizing principle—AI models are valuable insofar as the network can deliver their control signals directly to field equipment, vehicles, and industrial assets.
China’s 6G roadmap extends this logic by formalizing three pillars: integrated sensing and communication, ubiquitous multi-layer coverage from orbit to sea level, and AI-native networks that optimize themselves. Integrated sensing and communication (ISAC) is particularly emblematic. By allowing a base station to function simultaneously as a radar and communication node, 6G collapses sensing and networking into one unified electromagnetic infrastructure. This expands the role of networks from carriers of information to active interpreters of the physical world.
A notable example is the ISAC-based structural health monitoring project on the Hutong (Shanghai–Suzhou–Tongzhou) Yangtze River Bridge, jointly conducted by Huawei, China Tower, and Tongji University. Using downlink channel state information and beamforming feedback from 5G base stations operating at both sub-6 GHz and millimeter-wave frequencies, the system infers micro-deformations of the bridge’s towers and cables without installing additional sensors. Millimeter-wave beams, with their narrow beamwidth, achieve sub-millimeter displacement detection, while structural dynamics models filter out noise caused by temperature shifts or wind. During Typhoon Doksuri in 2023, the system measured a lateral displacement of 12.7 mm with an error margin below 1.2 mm compared with GNSS reference points. The results now feed directly into Jiangsu’s provincial bridge-monitoring cloud platform, completing a communication–sensing–early-warning loop. This case, documented in IEEE Transactions on Wireless Communications, illustrates how communication infrastructure can be upgraded via software into high-precision sensing assets.
Beyond terrestrial deployments, China’s 6G vision extends connectivity through a tri-layer architecture of low-orbit satellites, high-altitude platforms, and dense ground networks. This “air–sky–earth–sea” fabric aims to provide uninterrupted coverage for autonomous systems, maritime operations, logistics corridors, and national emergency response, ensuring that sensing and AI-driven control are available everywhere, not just in populated regions. Complementing this is the concept of the AI-native network: a communication system that reasons, predicts congestion, allocates spectrum, and orchestrates edge computing resources autonomously.
Taken together, these developments underscore a distinctive national strategy. While the United States emphasizes scaling large AI models and raw compute, China is concentrating on activating industrial assets through pervasive, high-fidelity connectivity. The core bet is that the next epoch of digital transformation will be won not merely by smarter algorithms, but by networks capable of integrating sensing, computation, communication, and control into a single operational fabric—one that ties consumer-level connectivity directly to the deepest layers of production.
HarmonyOS NEXT: Building a Unified, Distributed Intelligence Ecosystem
Huawei’s 2024 release of HarmonyOS NEXT marked a decisive break from reliance on the Android Open Source Project. Yet the system’s significance extends far beyond the question of whether it can run popular global apps. Its true innovation lies in distributed soft-bus technology, which allows heterogeneous devices to share computing power and data with unprecedented efficiency and security. This architecture enables scenarios such as mobile phones drawing on car-mounted cameras for real-time AR navigation, hospital CT equipment synchronizing encrypted imaging data with diagnostic tablets and AI systems, and fleets of unmanned mining trucks coordinating obstacle avoidance through microsecond-level communication.
The core logic of HarmonyOS is therefore not to replace existing mobile operating systems like iOS or Android, but to establish a unified trust and communication protocol across devices and environments. By enabling continuous, secure data flows between disparate terminals, it addresses the long-standing problem of information silos and unlocks the potential for genuinely distributed intelligence.
In contrast, many of China’s purported “homegrown operating systems” concentrate primarily on imitating user interfaces or maintaining compatibility with existing apps. Because they do not address the deeper challenges of cross-device data fragmentation or service continuity, they ultimately function as isolated “islands within islands.” HarmonyOS demonstrates that genuine technological independence is not achieved through solitary reinvention, but through the creation of interoperable ecosystems in which capabilities move fluidly across devices.
Networked Mobility Divergence: Intelligent EVs as a US-China Roadmap Mirror
Intelligent electric vehicles have become a revealing microcosm of the diverging technological trajectories of the United States and China. Each country’s approach to autonomous mobility reflects deeper national philosophies regarding infrastructure, data governance, and the relationship between market-driven innovation and state-coordinated systems. Nowhere is this contrast more pronounced than in the differing pathways exemplified by Tesla in the United States and by Huawei ADS and XPeng XNGP in China.
The US model, typified by Tesla’s Full Self-Driving system, is rooted in the logic of single-vehicle intelligence and large-scale cloud-based model training. It relies on high-precision maps, pure vision sensing, and centralized computing power to rapidly iterate models. This architecture has shown remarkable scalability in relatively structured environments, yet its dependence on on-board computing and stable connectivity reveals limitations in complex or unfamiliar terrains. Tesla’s struggles in Chongqing’s mountainous regions in 2024—where the absence of real-time external perception severely constrained performance—highlight the structural fragility of a system built primarily around vehicle-centric autonomy.
China’s pathway, represented by Huawei’s ADS 3.0 and XPeng’s XNGP, advances a fundamentally different paradigm: vehicle–road–cloud integration. This architecture fuses vehicle sensors, roadside perception units, edge computing nodes, and 5G-V2X communication into a distributed intelligent network. The objective is not merely to enhance a single vehicle’s capabilities but to transcend the inherent limits of individual perception and decision-making. By orchestrating collaboration among vehicles, infrastructure, and cloud-based intelligence, China aims to deliver safer, more reliable, and highly scalable autonomous mobility.
This integrated model reflects the competitive advantages of China’s state-enabled technological system. Centralized deployment of roadside sensing and communication infrastructure helps overcome the market’s chicken-and-egg problem; national coordination around the C-V2X standard reduces fragmentation; domestic cloud and edge networks safeguard sensitive traffic data; and collaboration across firms—Huawei in MEC and 5G, Baidu in cloud control, Hikvision in roadside vision—creates a full-stack industrial ecosystem. Pilot programs in Guangzhou, where roadside radar already compensates for weather-induced blind spots, demonstrate how networked intelligence can guarantee near-continuous takeover capability independent of any single vehicle’s computational limits.
The divergence between the U.S. and Chinese models ultimately reflects different underlying views of how technology advances. The American model emphasizes maximizing the autonomous capabilities of each individual machine, while the Chinese model centers on collective intelligence enabled through coordinated systems. As Ren Zhengfei noted in another context, true sophistication arises not from isolated strength but from the synchronized operation of many interconnected elements. In intelligent electric vehicles—as in the wider U.S.–China technological race—the most significant breakthroughs may come not from ever-greater on-board computational horsepower, but from the depth and coherence of the networks that integrate machines, infrastructure, and data into a unified whole.
Returning to the Essence: 98% of AI’s Value Is in Implementation
Artificial intelligence today delivers its greatest value not through the distant quest for “ultimate intelligence,” but through the disciplined, grounded work of implementation. In industry after industry—transportation, mining, steel, water management, glass manufacturing, and healthcare—the transformative impact of AI emerges not from abstract breakthroughs, but from practical deployments that optimize processes, reduce risk, and elevate productivity. As Ren Zhengfei noted, progress is measured not by symbolic sophistication but by real operational change: even coal miners now enter fully digitized, automated workplaces, a vivid testament to AI’s integration into the physical economy.
This contrast illuminates a deeper divergence between American and Chinese AI strategies. The United States orients its efforts toward Artificial General Intelligence, pursuing a conceptual understanding of intelligence itself. Its ambition lies in universal models, theoretical frontiers, and the exploration of cognition and consciousness. While such research holds profound scientific significance, it risks drifting away from immediate societal needs, slowing the conversion of cutting-edge ideas into tangible economic or industrial value.
China’s approach, by contrast, is fundamentally problem-driven and execution-oriented. It treats AI as an instrument for empowering real-world industries and improving governance, evaluated by concrete, quantifiable outcomes: a fraction of a percent gain in production efficiency, a slight reduction in accidents, or faster customer service responses. This focus accelerates scenario penetration across the economy. Yet it also risks fragmentation, as numerous highly customized vertical models may lack a shared foundation, limiting long-term integrative and ecosystem-level advances.
Neither paradigm is inherently superior; each reflects distinct institutional logics and developmental priorities. But history demonstrates that the greatest societal transformations emerge from sustained incremental improvement rather than sudden intellectual leaps. A 0.1% improvement in coal washing multiplied across billions of tons yields tens of millions of tons in refined output. A 1% performance gain in steelmaking translates into massive energy savings. Automation deployed across major global ports can reshape supply-chain resilience at scale. These achievements arise not from speculative AGI breakthroughs but from disciplined application, systems integration, and deep domain embedding.
Ultimately, the future advantage may belong to those who can fuse the strengths of both traditions: the deep insight that expands the frontier of intelligence and the relentless execution that converts technology into real economic and social value. But for now, the overwhelming share of AI’s impact—its essential 98%—still lies firmly in implementation.
Final Thoughts
China’s response to the U.S. “chip wall” reflects a strategic choice: instead of pursuing isolated breakthroughs, it is building an interconnected technological continent—advancing 5G/6G infrastructure, developing HarmonyOS and EulerOS, and empowering industry with Ascend and Kunpeng computing. Ren Zhengfei’s insight that globalization is not ending but restructuring captures the essence of this approach: real power lies in transforming computing into services, data into decisions, and isolated systems into cohesive networks. The young engineer who helped shape 12306 into a world-leading real-time ticketing platform embodies this principle. Her work illustrates that great systems are not forged through singular feats, but emerge from the resonance of countless connections. In the AI era, connectivity itself becomes the highest form of productivity—and the defining advantage of those who can weave it.
References
- Minutes of the meeting between Ren Zhengfei and the Chairman, coaches and award-winning athletes of the ICPC, November 2025