▲ All images featured in this article are sourced from the internet. Should any copyright infringement be identified, please contact us for prompt removal.

Has the era of AI smartphones truly dawned?

This article originally appeared in Shadow Memo, penned by Mo Yingsheng.

July 2026, Shanghai, at the World Artificial Intelligence Conference (WAIC).

Venturing into the WAIC 2026 exhibition hall, you'll observe a fascinating trend: large AI models have taken a backseat, while AI-powered hardware steals the limelight.

Among the hardware displays, the most crowded section showcases three smartphones, each claiming to be the 'world's first' intelligent agent phone.

Honor unveiled the Robot Phone, the world's inaugural robotic smartphone, featuring a titanium alloy mechanical gimbal at its pinnacle. This gimbal can track users like a petite 'hand' and even sway to the rhythm of music. Nubia, in collaboration with ByteDance, launched the NaviX Ultra, powered by the Doubao assistant, asserting itself as the world's premier AI intelligent agent phone. Meanwhile, StepFun introduced the STEPX Neo, the first large model-native intelligent agent phone, rebuilt from scratch at the operating system level.

Three 'world's firsts,' each following a distinct path.

This scenario evokes memories of the smartphone landscape pre-iPhone era in 2007. Everyone sensed change was imminent, yet the ultimate victor remained a mystery.

Just two days prior to WAIC's opening, the Cyberspace Administration of China (CAC) made a significant announcement: the on-device generative AI services of seven manufacturers—Apple, Huawei, Xiaomi, OPPO, vivo, Samsung, and Nubia—were officially registered as an independent category. This marked China's first centralized registration of 'on-device AI for smartphones.'

With policy backing, product launches, and tech giants joining the fray,

AI smartphones appear to have genuinely arrived.

However, casting our minds back to 2023, when Qualcomm's CEO first introduced the concept of AI smartphones, we encounter a stark contrast: despite three years of buzz, user enthusiasm remains lukewarm.

From Conceptual Buzz to Systemic Overhaul

When AI smartphones first emerged in 2023, mainstream smartphone manufacturers swiftly and uniformly embraced the concept, igniting a wave of marketing fervor.

Huawei, vivo, OPPO, and Honor all integrated self-developed AI intelligent agents. During product launches, their PowerPoint slides grew increasingly elaborate, showcasing features like AI object removal, AI summarization, AI search, and voice-to-text transcription.

Yet, when these phones reached consumers, the experience fell short of expectations.

The so-called AI experience amounted to cleaner photo edits, faster voice-to-text conversion, and slightly more conversational voice assistants.

Ultimately, it was merely a more proactive voice assistant and a photo album capable of automatic retouching.

Back then, AI smartphones were essentially traditional operating systems with an AI veneer. The core operating system remained unchanged, and apps continued to operate in isolation.

The real turning point arrived in late 2025.

ByteDance and Nubia's first-generation Doubao phone, the M153, sold 30,000 engineering prototypes in a limited release for 3,499 yuan.

Without a launch event or preheat campaign, it quickly sold out, with secondhand prices skyrocketing to 7,999 yuan.

The market went wild for a simple reason: this phone offered an unprecedented experience. With system-level access, AI could recognize the screen, open apps, compare prices, order food, and book flights autonomously.

But the honeymoon period was short-lived.

The following day, some users found their WeChat accounts banned due to 'abnormal login environments.' Soon after, Taobao triggered human-machine verification, and apps like Agricultural Bank of China and China Construction Bank suspended logins and payments, citing 'risky environments.'

Within days, national-level apps like WeChat, Taobao, Alipay, Meituan, and Pinduoduo blacklisted the Doubao assistant.

Doubao relied on a 'simulated click' approach, using accessibility services to read an app's internal tag structure and then simulate touch controls to operate the app autonomously. The issue was that this mechanism acted as a massive backdoor for risk control systems.

More fundamentally, Doubao's approach disrupted the long-established entry logic of the global internet ecosystem.

Users typically access an app through search, clicks, and redirects, each step tied to app distribution, traffic referral, and monetization strategies. If AI agents could simulate all these steps, the entire traffic ecosystem built around apps would crumble.

Being a guest in someone else's house means you can be evicted at any time.

Three Players, Three Strategies

Doubao's setback made one thing clear to the industry: granting intelligent agents access to old systems only makes them permanent guests.

At WAIC 2026, three players unveiled three distinct solutions.

ByteDance + Nubia: A Deep Integration Between Hardware and Model Makers

Nubia's NaviX Ultra pursued deep integration between a hardware manufacturer and a large model company. Learning from the first-generation Doubao phone's lessons, both sides adjusted their strategy: the GUI Agent would only handle unoptimized general apps, while mainstream apps like payments and social media would share partial data through standardized interfaces like A2A or MCP.

This amounted to shifting from 'barging in' to 'negotiating entry.'

Notably, Nubia has suspended its traditional smartphone business. Its parent company, ZTE, registered Shanghai Shenqi Era Intelligent Terminal Co., Ltd., to fully focus on AI. The NaviX Ultra had an initial stock of about 200,000 units, no longer a limited 'engineering prototype.'

For a major hardware player to willingly grant unrestricted OS-level access to an external large model was unthinkable just two years ago.

StepFun: The Model Company Takes the Plunge

StepFun adopted an even more aggressive approach. Instead of embedding an intelligent assistant into the operating system, it added a dedicated execution layer on top of Android, rebuilding the entire underlying framework from scratch to create Step AOS, a native OS for intelligent agents.

StepFun's chairman, Yin Qi, succinctly stated: 'Granting intelligent agents access to old systems makes them permanent guests. Only by building a dedicated execution environment can they become true natives.'

Yin admitted that terminal experts he consulted unanimously advised against touching hardware. But he chose to 'ignore the advice,' arguing that instead of being a 'guest' in old systems, they should build a 'house' for intelligent agents to become 'natives.'

The team backing StepFun is noteworthy: terminal president Ni Jiayue hails from Honor, while manufacturing is handled by ODM giant Huaqin. StepFun's largest recent funding round of $2.5 billion included investors like Tencent and consumer electronics supply chain companies.

This mirrors similar moves overseas: OpenAI acquired the company founded by Apple's former chief design officer, Jony Ive, and was reported to be collaborating with MediaTek and Qualcomm to develop custom smartphone processors.

The rationale behind model companies diving into hardware is clear: only by controlling the terminal can they retain user relationships and data flywheels.

Honor: Giving Smartphones a 'Brain' and 'Hands'

Honor took the most unique approach. It didn't just give its phone a 'brain'—it added 'hands.'

The Robot Phone features a four-degree-of-freedom gimbal with multimodal perception, extending the intelligent agent's interaction range into the physical world. Meanwhile, Honor upgraded MagicOS to Agentic OS, the industry's first partner-type multimodal intelligent agent operating system.

Honor's chief AI scientist, Huang Fei, said Agentic OS isn't about 'adding an AI assistant to the system' but reconstructing a new OS centered around 'intent' and 'tasks.'

Honor CEO Li Jian elevated this philosophy: AI's evolution must transcend its cold, tool-like nature, moving from operating systems to embodied interactions to become a partner-like, human-like entity.

The debut of three 'world's firsts' underscores a reality: no universally recognized standard exists in this field. Anyone can redefine 'first,' precisely because no one has truly broken through yet.

Giants Enter the Arena, But What Lies Beyond the Hype?

If WAIC's three products represent the 'innovators' exploring new territory, the centralized registration of on-device AI services by seven manufacturers marks the industry's entry into a 'compliance-driven' new phase.

Apple deserves special attention.

Over the past two years, Chinese flagship phones have all integrated localized on-device large models, leaving domestic iPhones lagging in intelligent interaction capabilities.

In July, Apple's 'Apple Intelligence' completed CAC registration, partnering with Alibaba and Baidu for long-text generation and localized search services, respectively. This AI suite will debut on the iPhone 18 Pro.

In the high-end market, localized AI experience has become a core competitive edge. The iPhone 18 aims to help Apple catch up in the intelligent race and narrow the experience gap with Chinese models.

Samsung isn't far behind. On July 22, it released the Galaxy Z Fold8 series, with Galaxy AI as a key upgrade, introducing the 'Gemini Intelligence' system developed with Google for the first time.

Notably, MiniMax's on-device large model entered Samsung's global flagship lineup, marking the first time a Chinese on-device large model was featured in a global flagship product from an international leading smartphone maker.

Further afield, OpenAI is secretly developing its own AI smartphone. According to analyst Ming-Chi Kuo's report, this 'artificial intelligence agent phone' could enter mass production as early as the first half of 2027, with a target of 30 million units shipped over two years.

It will debut with a chipset deeply customized based on MediaTek's Dimensity 9600, using TSMC's N2P process.

From Apple to Samsung, from ByteDance to OpenAI, from smartphone makers to model companies, everyone is racing down the same path.

But after the hype, cooler heads are beginning to ask: What exactly is an AI smartphone?

Market data paints an exciting picture for AI smartphones.

Counterpoint Research predicts that by 2026, 45% of global smartphones with generative AI capabilities will be shipped, surpassing 50% in 2027.

IDC expects 147 million next-generation AI phones to be shipped in China in 2026, up 31.6% year-on-year, accounting for 53% of the market. Canalys reports that Chinese AI phone shipments surged 320% year-on-year in Q1 2026.

According to the National Development and Reform Commission's Department of Innovation and High-Tech Development, sales of AI smartphones and AI PCs are expected to surpass non-AI products for the first time.

But behind these shiny numbers lie several unavoidable challenges.

First is hardware cost.

Due to AI servers aggressively competing for wafer capacity, LPDDR4/5 memory prices roughly doubled in Q2 2026 compared to late 2025. High costs are forcing low-end models out of the market. Global smartphone shipments are expected to decline 13.9% year-on-year to 1.08 billion units in 2026, the lowest since 2013.

AI smartphones won't come cheap. Few can afford to spend thousands of yuan on a phone with a one- to two-year replacement cycle.

Second is the physical limits of on-device capabilities.

AI smartphones can't rely on remote code execution to fix their shortcomings. Algorithms are tightly coupled with a phone's NPU performance, memory bandwidth, system-level scheduling, and even thermal design.

This is nothing like the 'launch first, polish later' approach of apps.

Finally, there's the tough nut of profit distribution.

When AI agents can understand instructions and handle everything with a single click, they become the most dangerous 'gravediggers' of internet commerce.

Users no longer need to navigate, browse pages, or watch ads, meaning the trillion-dollar traffic ecosystem built on daily active users and engagement time is about to collapse.

For AI smartphones to break through at scale, technical risk control is just the surface-level challenge. The harder problem lies in redistributing profits.

How Will AI Smartphones Transform Everything?

Despite the challenges, the transformation brought by AI smartphones will be fundamental.

For users, human-computer interaction will shift from 'app-centric' to 'agent-centric.' Users won't need to open apps one by one, fill in information repeatedly, or navigate through redirects. A single verbal instruction will allow AI to automatically coordinate multiple apps to complete complex tasks.

Qualcomm president and CEO Cristiano Amon summarized this shift: users' tasks and experiences will flow across smartphones, PCs, cars, and other devices, guided by intelligent agents.

For the industry, the rise of AI agents brings operating systems back to the center stage after years of dormancy. Traditional app architectures cannot support the core capabilities of autonomous execution and cross-scenario coordination by intelligent agents. If users adopt AI agents as their primary gateway, weakening reliance on standalone apps, traditional apps' traffic dominance will erode.

Operating systems will once again become the core power hub of the smartphone industry.

For supply chains, AI terminal innovations will drive value reconfiguration across chips, storage, cooling, batteries, RF components, and more. On-device hardware is poised to enter a new upgrade cycle.

But transformation brings risks.

For AI smartphones to achieve widespread commercialization, they must surmount obstacles related to technical experience, privacy and security, and profit allocation. Intelligent agents necessitate the collection of massive volumes of sensitive data. Local processing demands substantial computational capabilities, whereas cloud processing poses the risk of data breaches.

More crucially, if all manufacturers eventually converge on a model of "system-level reconstruction plus intelligent agent assistance," could AI smartphones follow the trajectory of the smartphone industry—transitioning from a focus on parameter competition to functional uniformity?

Conclusion

At this juncture, the author cannot help but ponder: Have AI smartphones truly arrived?

The response is: They have made their entrance, yet they are not fully here.

Policy filings have been initiated, seven tech behemoths have joined the fray, three models claiming to be the "world's first" have been launched, and market adoption is rapidly increasing.

However, they have not truly arrived in the sense that a sustainable business model has yet to be established, there are no unified industry standards, the challenge of profit distribution remains unaddressed, and users are still adopting a wait-and-see approach.

Li Dahai, CEO of FaceMind AI, offered his perspective at WAIC: Currently, there are three prevailing trends in the AI smartphone sector—smartphone manufacturers developing in-house AI capabilities, the external procurement of AI technologies for devices, and model companies venturing into smartphone manufacturing. He believes this is not indicative of "fragmentation" but rather a common phase in the early stages of industry evolution.

In these nascent stages, anything is possible.

StepFun has opted to rebuild the operating system from the ground up; ByteDance and Nubia have forged a deep partnership; Honor has concentrated on hardware innovation with "embodied interaction"; Apple is "playing catch-up"; and OpenAI is formulating its own hardware strategy.

Who is on the right track and who is not? The answer remains elusive.

Earlier, the Beijing Business Today remarked: "On the surface, it appears that the starting gun for AI smartphones has been fired, but in reality, the countdown to the final outcome has already commenced. Once the answer is unveiled, there is often no opportunity for a second chance."

The moment consumers open their wallets, they have already cast their votes. If the initial wave of reputation falters, the prospects for a resurgence in this highly competitive market are virtually nonexistent.

No one doubts that AI smartphones may represent the future trend, but they are still in the exploratory phase.

As for how innovation will progress and whether homogenization will ensue, we will have to await the verdict.

After all, market feedback serves as the ultimate litmus test.