China's AI Terminal Revolution: How Smart Devices and Humanoid Robots Overtook the Old World

*The invisible infrastructure powering China's AI terminal revolution — where every device is becoming a thinking machine*

The Moment the Register Tipped

At 9:47 AM on a Tuesday in early July 2026, the transaction counter at a Suning electronics store in Shanghai's Xujiahui district hit a number that would have seemed impossible eighteen months earlier: the 10,000th AI-powered device sold that week. Not the 10,000th smartphone. Not the 10,000th computer. The 10,000th device embedded with an AI co-processor, a neural engine, or an on-device large language model — capable of reasoning, generating, and adapting without ever phoning home to a cloud server.

"Last year, customers asked 'What's an AI phone?'" said the store manager, pulling up the week's sales dashboard. "Now they ask 'Why would I buy anything else?'"

That casual remark captures the most significant shift in consumer technology since the smartphone era began. On July 7, 2026, China's National Development and Reform Commission (NDRC) confirmed what the retail data had been whispering for months: AI phones and AI PCs in China are on track to outsell their non-AI counterparts for the first time in history. Not in some distant future. Not in a niche segment. This year. Right now.

The numbers are staggering. Over 100 million AI-capable terminals — phones, laptops, tablets, and wearables — shipped in China in 2025. The 2026 figure is projected to grow by another 30% or more. And if that were the only story, it would still be remarkable. But the terminal revolution is just one thread in a much larger tapestry. While consumers were upgrading their pockets and backpacks, factories were upgrading their floors. China's humanoid robot production is expected to exceed 100,000 units this year. The country's AI-related industry already surpassed one trillion RMB in 2025. And all of this is accelerating toward a single, unmistakable inflection point: the 2026 World Artificial Intelligence Conference (WAIC), opening July 17 in Shanghai, where more than 300 AI products will make their global debut.

This is not hype. This is the crossover.

From Gimmick to Necessity: The AI Device Journey

Rewind to early 2023. The launch of ChatGPT had triggered a global AI frenzy, but in China, the consumer-facing reality was underwhelming. Early "AI phones" offered little more than smarter photo editing and slightly better voice assistants. AI PCs were essentially marketing labels slapped onto machines with modest performance gains. Consumers were skeptical. Analysts were cautious. The phrase "AI washing" — attaching the AI label to ordinary hardware — became a common critique.

What changed between then and now? Three converging forces rewrote the script.

First, on-device AI models reached critical capability. By late 2024, Chinese developers had compressed large language models to run efficiently on mobile NPUs (Neural Processing Units). Qualcomm's Snapdragon 8 Elite and MediaTek's Dimensity 9400 both shipped with dedicated AI engines capable of 50+ trillion operations per second. On the PC side, Intel's Lunar Lake and AMD's Ryzen AI 300 series brought comparable NPU performance to laptops. The hardware finally existed to run meaningful AI locally.

Second, killer use cases emerged. Real-time translation that works offline. AI-generated meeting summaries that actually capture context. Photo editing that understands semantic relationships — "make the background more dramatic without changing the subject's face." Code completion that runs entirely on-device, preserving privacy. These weren't gimmicks. They were daily productivity tools that consumers could no longer imagine living without.

Third, China's regulatory framework created a distinctive ecosystem. While Western AI assistants struggled with inconsistent cloud connectivity and privacy concerns, China's 169 government-approved large language models provided a regulated, reliable foundation for on-device AI. Companies like Huawei, Xiaomi, OPPO, and vivo integrated these models deeply into their operating systems, creating AI-native experiences rather than bolt-on features.

*Sources: NDRC press briefing, July 7, 2026; IDC global AI PC forecast, April 2026; company filings.*

The table above tells a story of acceleration that would make any semiconductor executive smile. But the consumer side is only half the narrative. The other half is happening on factory floors, in warehouses, and on assembly lines — where a different kind of terminal is taking shape.

The Factory Floor Gets a New Worker

While consumers were deciding whether their next phone should have an AI chip, China's manufacturing sector was quietly deploying an entirely new category of worker: the humanoid robot. And in 2026, that deployment reached industrial scale.

According to Gan Xiaobin, Deputy Director of the Department of Science and Technology at China's Ministry of Industry and Information Technology (MIIT), humanoid robot production in China is expected to exceed 100,000 units this year. That figure, announced at the WAIC press conference on July 7, represents a quantum leap from the 13,318 units shipped globally in 2025 — of which Chinese companies already accounted for roughly 87%.

The numbers are even more striking when you consider the competitive landscape. While Tesla's Optimus and Figure AI's humanoids have captured Western media attention, their combined 2025 deliveries totaled approximately 300 units. Chinese firms, led by Unitree (targeting 75,000 annual capacity), AgiBot (5,168 units shipped in 2025, ranking #1 globally), and UBTECH (commercial deployments with BYD), have built an industrial pipeline that is orders of magnitude larger.

*Sources: Omdia global humanoid shipment report, 2025; SVRC Research China Robotics 2026; MIIT press briefing, July 7, 2026; company announcements.*

What's driving this production explosion? Three factors, each reinforcing the others.

Supply chain density is the first. Within a 100-kilometer radius of Shenzhen, a robotics startup can source brushless DC motors, harmonic drives, strain wave gears, CNC-machined aluminum frames, custom PCBs, depth cameras, and force-torque sensors — often with first samples delivered within a week. A prototype that costs $15,000–$25,000 and takes 10–14 weeks in the United States can be produced in Shenzhen for $4,000–$8,000 in 10–14 days. This is not a marginal advantage. It is a structural one that shapes the entire competitive landscape.

The EV-to-robot manufacturing pipeline is the second. China's dominance in electric vehicles created a vast ecosystem of precision actuators, battery management systems, and sensor fusion technologies — all directly transferable to humanoid robotics. BYD alone, the world's largest EV manufacturer, employs over 700,000 workers and represents a demand base for tens of thousands of robotic systems. The same factories that learned to assemble batteries and motors at scale are now learning to assemble robots.

Policy coordination is the third. China's Humanoid Robot Action Plan, issued by MIIT and five other ministries in 2025, set a national target of 100,000 deployed humanoids by 2027. The plan is not merely aspirational. It is backed by dedicated robotics industrial parks in Beijing, Shanghai, Shenzhen, Guangzhou, Hangzhou, and Chengdu — each offering subsidized facilities, streamlined permitting, and local government investment. National, provincial, and municipal programs are designed to complement rather than duplicate each other, creating a layered support structure that addresses everything from fundamental research to factory-floor deployment.

*Shenzhen's hardware ecosystem enables humanoid robot iteration cycles measured in days, not months — a structural advantage no other geography can match*

Shanghai's 300 Billion Bet

If China's AI terminal revolution has a capital, it is Shanghai. And the city is betting big.

At the July 7 WAIC press conference, Shanghai's Vice Mayor Chen Jie revealed that the city's 394 large-scale AI enterprises generated a combined 637 billion RMB ($87.8 billion) in revenue in 2025 — a year-on-year growth of 39.5%. That makes AI one of the fastest-growing sectors in Shanghai's economy, and the city is not content to rest on those laurels.

By 2027, Shanghai aims to grow its smart terminal industry to 300 billion RMB ($41.4 billion). The plan focuses on AI phones, AI PCs, AI wearables, and the software ecosystems that bind them together. It is not simply a manufacturing target. It is an ecosystem play: Shanghai wants to own the full stack, from chip design (via companies like Cambricon and Horizon Robotics) to operating system integration (via HarmonyOS and Xiaomi HyperOS) to application development (via the 169 approved large models).

*Sources: Shanghai Municipal Government WAIC press briefing, July 7, 2026; NDRC innovation division.*

The smart computing power figure is worth pausing on. Shanghai's 160,000+ petaflops of intelligent computing capacity — roughly equivalent to 160 exaflops — represents one of the largest AI-dedicated compute clusters in the world. For context, the U.S. Department of Energy's Frontier supercomputer, the world's fastest traditional supercomputer, operates at approximately 1.2 exaflops. China's AI infrastructure is being measured in entirely different units.

This computing foundation is what enables the on-device AI models that are now standard in Chinese terminals. It is what trains the humanoid robot control policies that allow machines to walk, grasp, and navigate without falling over. And it is what will power the next wave of AI-native applications — from real-time medical diagnostics to autonomous logistics coordination — that are expected to debut at WAIC 2026.

What 169 Large Models Built

Behind every AI terminal and every humanoid robot lies a large language model — or, more accurately, 169 of them. That is the number of AI models that have passed China's regulatory filing and review process as of mid-2026, a figure that has grown from just 45 in 2023.

This regulatory framework, often criticized by Western observers as overly restrictive, has created something unexpected: a curated, reliable ecosystem of foundational AI models that device manufacturers can trust. Unlike the Wild West of unregulated model deployment in some markets, China's approved models have been vetted for safety, accuracy, and alignment with national standards. The result is a platform layer that hardware companies can build on with confidence.

*Sources: Model provider technical reports; China AI regulatory filing registry; industry benchmarks.*

The diversity of this model ecosystem is its strength. Where the U.S. market has consolidated around a handful of models (GPT-4, Claude, Gemini, Llama), China's landscape remains genuinely competitive. A developer building an AI-native application for Chinese terminals can choose from models optimized for coding, medical diagnosis, scientific computing, gaming, or social interaction — each with its own strengths and target hardware.

This model diversity also drives hardware diversity. Qualcomm and MediaTek dominate the AI phone chip market globally, but in China, Huawei's Ascend and Kirin chips, Cambricon's MLU series, and Horizon Robotics' Journey chips all compete for design wins. The result is a hardware ecosystem that is more resilient to single-supplier disruption than any other major market.

The Global Implications

What does China's AI terminal crossover mean for the rest of the world? The answer depends on where you sit.

For semiconductor companies, China represents both the largest opportunity and the most complex challenge. The AI terminal market is driving unprecedented demand for NPU-equipped chips, memory, and sensors. But U.S. export restrictions on advanced AI chips have pushed Chinese manufacturers toward domestic alternatives. Huawei's Ascend 910B and 910C chips, while not yet matching NVIDIA's H100 in raw performance, are closing the gap rapidly — and they are available in quantities that NVIDIA's restricted products cannot match in the Chinese market.

*Sources: IDC AI PC market forecast, April 2026; Canalys smartphone AI market analysis; AI in China estimates.*

For consumer electronics brands, the message is clear: AI is no longer a premium feature. It is the baseline. Apple's delayed entry into on-device AI (Apple Intelligence launched in late 2025 with limited Chinese market support) has ceded precious ground to domestic competitors. Samsung's Galaxy AI has gained traction in China but faces the same challenge: Chinese consumers increasingly expect AI capabilities that are deeply integrated with local services, languages, and regulatory frameworks.

For robotics companies, China's humanoid scale represents a competitive threat — but also a potential partnership opportunity. The sub-$10,000 humanoid arms already being produced by Chinese manufacturers are finding customers in Southeast Asia, South America, and Africa — markets where affordable automation is unlocking industrial capability for the first time. Chinese robotic arm exports have grown 4.45x from 2020 to 2025. This is not a zero-sum game. It is a market expansion.

For policymakers, the China model offers a study in coordinated industrial policy. The combination of national targets (100,000 humanoids by 2027), local investment (Shanghai's 300 billion RMB terminal target), and regulatory framework (169 approved models) creates a system that is more predictable than laissez-faire approaches but more flexible than central planning. Whether this model is replicable elsewhere depends on local institutional capacity, but the results are hard to ignore.

*China's 160,000+ petaflops of intelligent computing power — one of the largest AI-dedicated infrastructures on Earth — is the invisible engine behind every AI terminal and humanoid robot*

The Road Ahead: WAIC and Beyond

All of these threads converge in one place: the 2026 World Artificial Intelligence Conference, opening July 17 in Shanghai. With an exhibition area exceeding 100,000 square meters for the first time, 1,100+ exhibitors, 3,000+ exhibits, and 300+ products making their global debut, WAIC 2026 is positioned as the definitive showcase of where AI is heading — not just in China, but globally.

The conference has expanded beyond its traditional Shanghai Expo venue to a "three locations, four venues" layout spanning Expo, Zhangjiang, and West Bund. This physical expansion mirrors the thematic expansion: WAIC is no longer just a trade show. It is an ecosystem event that spans forums, exhibitions, competitions, application experiences, innovation incubation, and talent recruitment.

*Sources: Shanghai Municipal Government WAIC press briefing, July 7, 2026; WAIC official announcements.*

Among the products debuting at WAIC: Huawei's Atlas 950 supercomputer (the largest supernode display in the industry), MiniMax's M3 multimodal model, Stepfun's Agent OS, 3D near-memory computing chips, the world's first AI agent phone, and multiple humanoid robots and AI dexterous hands. The intelligent computing and embodied AI tracks alone each feature more than 200 companies.

The academic dimension is equally significant. WAIC Academic, the conference's first-ever international academic track, has received 284 submissions from 11 countries. Papers accepted will be published by Springer. The presence of nine Turing Award and Nobel laureates — including Andrew Yao as conference chair and Richard Sutton as international co-chair — signals China's ambition to be not just a consumer of AI research but a producer of foundational knowledge.

Looking beyond WAIC, the trajectory is clear. China's AI industry is projected to grow at 30%+ annually through at least 2027. The 100,000 humanoid robot target for 2026 is likely to be exceeded, not missed. And the AI terminal crossover — the moment when AI devices became the majority — is not an endpoint. It is a starting point for the next phase: AI-native devices that are not merely smarter versions of old products, but entirely new categories of product that we do not yet have names for.

What the World Is Saying

Zhihu — *Comment on AI terminal market analysis*

"从2023年的AI洗白到现在的真AI设备，中国只用了一年半。关键是供应链成本下来了，NPU芯片的价格比2024年降了60%以上。这才是真正的护城河。"

"From AI washing in 2023 to genuine AI devices now, China took only a year and a half. The key is supply chain costs — NPU chip prices dropped over 60% since 2024. That's the real moat."

Weibo — *User reacting to 100,000 humanoid robot target*

"100000台人形机器人！这比科幻电影还科幻。但是想想深圳工厂14天就能出原型，突然觉得这数字很真实。"

"100,000 humanoid robots! That's more sci-fi than sci-fi movies. But then you remember Shenzhen factories can produce a prototype in 14 days, and suddenly that number feels very real."

Xiaohongshu — *Lifestyle blogger reviewing AI phone features*

"刚换了AI手机，最大的感受不是某个功能多厉害，而是整个手机'懂'你了。拍照不用调参数，文档不用自己总结，连日程都会自动安排。这已经不是手机，是生活助理。"

"Just switched to an AI phone. The biggest feeling isn't any specific feature — it's that the whole phone 'gets' you. No need to adjust camera settings, no need to summarize documents yourself, even your schedule arranges automatically. This isn't a phone anymore. It's a life assistant."

Twitter/X — *Tech analyst on global AI hardware competition*

"China's AI terminal numbers are impressive but let's not forget — 90% of these devices run on Arm or Qualcomm chips. The real test is whether Huawei Ascend and Cambricon can replace them before the next export ban."

Douban — *Discussion on AI companionship and humanoid robots*

"AI陪伴要被禁了，人形机器人却要量产10万台。这很矛盾吗？仔细想想并不——一个情感依赖，一个工具替代。政策在试图划一条清晰的线：AI可以帮你干活，但不应该让你爱上它。"

"AI companionship is being banned, but humanoid robots are going into mass production of 100,000 units. Is that contradictory? Think carefully and it isn't — one is emotional dependency, the other is tool replacement. Policy is trying to draw a clear line: AI can help you work, but it shouldn't make you fall in love with it."

GitHub — *Developer comment on open-source robotics frameworks*

"The most underrated part of China's robot boom is the open-source ecosystem. AgiBot open-sourced their entire control stack. Unitree's ROS 2 packages are better documented than most Western alternatives. If you're building robots, you should be reading Chinese GitHub repos right now."

The Crossover Is Not the End

In 2026, China's AI terminals crossed a threshold. For the first time, devices that think outsold devices that don't. Humanoid robots moved from prototypes to production lines at scale. And a trillion-RMB industry proved that it could grow faster than skeptics predicted.

But the crossover is not the end. It is the beginning of a new phase — one where the distinction between "AI device" and "non-AI device" becomes meaningless because every device is AI-capable. Where humanoid robots are not novelties but standard equipment. Where the world's largest manufacturing base becomes the world's largest robot manufacturer, and the world's largest consumer market becomes the world's largest AI application laboratory.

The 2026 World Artificial Intelligence Conference will open its doors in seven days. It will showcase 300 products that do not exist in the global market today. Some will fail. Some will transform industries. All of them will be built in an ecosystem that has crossed the threshold from potential to kinetic energy.

For the rest of the world, the question is no longer whether China's AI revolution is real. The question is how to engage with it — as competitor, partner, customer, or observer. Because one thing is certain: the devices that just overtook the old world are not looking back.

*Related Articles:*

- China's AI Companion Ban: When 345 Million Users Lose Their Digital Friends

- China's Embodied Intelligence Infrastructure: How Robots Are Invading Factories, Malls, and Homes

- How China's AI Video Models Captured the Global Creative Frontier

- China's Multimodal AI Revolution: The End of Text-Only Models