Huawei’s latest moves signal a decisive escalation in its bid to dominate the next phase of intelligent computing. At its developer conference, the company unveiled HarmonyOS 7 with an architecture explicitly built for autonomous AI agents, while simultaneously advancing a new chip-design paradigm that aims to match leading-edge transistor densities without relying on restricted EUV equipment. These parallel efforts—software agents that can orchestrate complex, cross-device tasks and hardware innovations that compress signal travel time through three-dimensional folding—position Huawei to challenge both Apple’s ecosystem and the U.S. technology blockade on multiple fronts.
The timing is strategic. Apple’s iOS 27 AI overhaul remains blocked from mainland China by regulatory hurdles, leaving domestic users open to locally compliant alternatives. At the same time, Huawei’s semiconductor team, led by He Tingbo, is resurfacing after years of quiet work to promote the Tau Scaling Law as a viable alternative to Moore’s Law. The combination of an agent-centric operating system and novel chip architectures suggests Huawei is no longer merely surviving sanctions but actively reshaping competitive dynamics in consumer devices, cloud infrastructure, and automotive intelligence.
HarmonyOS 7 and the Rise of Specialized AI Agents
HarmonyOS 7 introduces an “agent-friendly” framework that connects users to more than 2,000 specialized AI agents through an upgraded voice assistant. The assistant maintains context across sessions, recalls prior interactions, and executes multi-step workflows without constant human intervention. Concrete examples include generating personalized marathon training regimens from health metrics and calendar data, or retrieving files stored on a linked laptop via seamless cross-device orchestration.
Richard Yu Chengdong, chairman of Huawei’s Consumer Business Group, reported a 90 percent success rate on complex tasks and claimed the framework reduces the effort required for developers to build new AI skills. A developer beta released on June 12 includes AI coding agents, with consumer rollout expected on the company’s next flagship smartphone later this year. Performance gains of 30 to 40 percent alongside lower memory consumption are positioned as direct advantages over rival mobile platforms.
These capabilities matter because they shift the value proposition from reactive assistants to proactive agents that can operate across phones, laptops, and IoT devices. If the 90 percent reliability figure holds under real-world conditions, developers may accelerate migration to HarmonyOS, particularly in markets where Apple’s AI features remain unavailable.
Tau Scaling and the Push for 3D Chip Density
On the hardware side, Huawei is promoting the Tau Scaling Law, which prioritizes reducing signal propagation delay through vertical stacking rather than单纯 shrinking transistor geometries. Researchers at Peking University demonstrated a “true-3D” EDA prototype compatible with Huawei’s LogicFolding architecture just one day after the framework’s unveiling. Empyrean Technology has already released Argus, a physical verification platform for 3D IC design that integrates with this approach.
The goal is to reach transistor densities equivalent to a 1.4-nanometer process by 2031 using existing DUV lithography tools. He Tingbo, widely known inside the company as the “chip queen,” presented the framework at the IEEE International Symposium on Circuits and Systems, marking her first major public appearance since 2019. By stacking circuits vertically, Huawei argues it can bypass both the physical limits of planar scaling and the export controls that restrict access to EUV equipment.
Success would validate a sanctions-resistant pathway for advanced silicon. However, analysts note that commercializing 3D IC designs at scale requires mature thermal management, yield improvements, and an entire ecosystem of compatible EDA tools—areas where U.S. vendors still hold structural advantages.
Ascend AI Chips Target Latin American Cloud Markets
Huawei Cloud Latin America president Mark Chen confirmed the company is evaluating deployment of its newest Ascend 950 family in regional data centers. The 950PR variant launched commercially in China in March and is already used by Alibaba, ByteDance, and DeepSeek. While Chen offered no firm timeline, the move would extend Chinese-designed AI accelerators into a region traditionally dominated by U.S. suppliers.
Latin America represents both an opportunity and a test case. Many governments there face fewer explicit restrictions on Chinese technology than Europe or North America, yet they still value performance-per-dollar in AI training and inference workloads. If Ascend clusters demonstrate reliable results on large language models, local enterprises could gain leverage in negotiations with Western vendors. The deployment would also generate real-world data on the chips’ power efficiency and software stack maturity outside China.
Automotive Ambitions and Efforts to Diversify Dependencies
Huawei’s automotive strategy is evolving on two tracks. Through the HIMA alliance with Chery, the Luxeed RX SUV will feature four LiDAR units and electromagnetic thermal-control die-casting, positioning it as a premium “FUV.” Yet the vehicle’s launch has been clouded by disputes over design claims, with Ferrari Greater China publicly questioning assertions that a former Ferrari chief designer contributed to the project.
Meanwhile, Seres, maker of the Huawei-co-branded Aito vehicles, launched the Aiva brand to reduce reliance on its dominant partner. Aiva’s first model, the ME7, will integrate ByteDance’s Doubao large language model and DeepRoute.ai autonomous-driving technology instead of Huawei’s Qiankun ADS. The strategic pivot underscores a growing industry pattern: Chinese automakers that once embraced deep Huawei integration are now hedging to retain greater control over software roadmaps and supplier relationships.
Ultra-Low-Power Optimization Expands the IoT Horizon
Beyond flagship smartphones and vehicles, Huawei executives highlighted HarmonyOS’s extreme memory efficiency. The operating system currently runs on 128 KB of RAM and is targeted for further reduction to 64 KB, enabling deployment on devices powered by a single dry-cell battery for a full year. Such specifications open entirely new IoT categories—industrial sensors, medical wearables, and agricultural monitors—where traditional mobile operating systems are impractical.
This efficiency focus complements the agent architecture. If lightweight agents can run reliably on minimal hardware, Huawei could extend its ecosystem into millions of edge devices that feed data back into centralized AI models, strengthening its data advantage.
Taken together, these initiatives reveal a coherent strategy: use architectural innovation to neutralize hardware sanctions, leverage an agent-centric OS to lock in developers and users, and diversify automotive exposure while still advancing in core vehicle intelligence. The coming year will test whether the Tau Scaling Law delivers manufacturable chips, whether HarmonyOS agents achieve sustained reliability, and whether global markets accept Chinese AI accelerators at scale. Success on any two of these fronts would materially alter competitive balances in consumer technology and cloud infrastructure.
