A billion-level bet on AI: Huawei, Xiaomi, OPPO, vivo and other phone makers are getting a reality check.

AI Smartphones: From Hype to Stress Test

Looking back from early 2026, 2025 was clearly the year when AI smartphones crossed the line from concept marketing to market reality.

It was a year defined by contradiction. On one side, smartphone makers poured tens of billions into AI Agents and on-device large models, pushing “AI phones” to the center of their narratives. On the other, the market responded with lukewarm sales, exposing the gap between ambition and consumer behavior.

Industry data shows that domestic smartphone shipments in the first three quarters of 2025 reached 70.9 million, 67.8 million, and 67.2 million units respectively, with year-on-year growth of 5%, -4%, and -3%. From another angle, this confirmed an uncomfortable truth: generative AI has yet to become a decisive reason for mass consumers to upgrade their phones.

As one distribution channel executive put it, “AI may reshape the industry someday, but today it’s not the key driver of replacement demand. Even government subsidies boost sales more directly than AI features.”

AI as the “Third Revolution”

Despite the “high praise, low sales” dilemma, leading manufacturers remain firm in treating AI as the industry’s “third revolution.” The cautionary tale of Nokia hangs like a sword overhead, pushing companies such as Huawei, OPPO, vivo, Xiaomi, and Honor to commit heavily to AI infrastructure, algorithms, and ecosystem control.

Since the rise of generative AI four years ago, these vendors—armed with massive user bases—moved early to build proprietary models, computing clusters, and data centers. Industry insiders estimate that leading brands have invested well over tens of billions of yuan in computing power and data infrastructure alone.

Take OPPO as an example: the company reportedly operates nine data centers globally, with its Dongguan facility hosting nearly 20,000 servers.

As one industry source noted, “A single large-scale training cluster costs billions. Startups can barely afford the electricity bill, let alone the hardware.”

By 2025, with open-source models lowering entry barriers, the focus began shifting away from raw infrastructure toward algorithm optimization, real-world scenarios, and practical on-device deployment.

The strategic goal also evolved—from bundling isolated AI features (image erasing, fraud detection, AI office tools) to transforming phone assistants into personal AI agents capable of understanding intent and acting proactively.

Manufacturers pursued this along two main paths: interaction paradigm upgrades and ecosystem construction. Assistants are moving from passive response to proactive service, combining vision, voice, location, and sensor data to interpret context and execute complex tasks.

Yet even executives acknowledge the difficulty. Cross-app execution, sensitive data handling, and payment-related scenarios raise privacy, security, and regulatory challenges that cannot be solved overnight.

A Short-Lived Disruptor

Just as consumers questioned whether AI assistants could truly “work for them,” an unexpected outsider briefly reignited public excitement.

In December 2025, ByteDance’s Doubao team partnered with ZTE to launch the nubia M153 Doubao AI Phone at RMB 3,499. Despite clear warnings that it was an experimental product, the first 30,000 units sold out quickly, with resale premiums appearing on secondary markets.

Most buyers were developers and tech enthusiasts eager to explore cutting-edge AI. Early users praised its dedicated AI button and system-level automation—but soon ran into limits. Major apps restricted automated operations for security reasons, gradually undermining the phone’s core value proposition.

Industry insiders were blunt: the product packaged existing AI features into a cohesive system experience, but relied on high-level permissions and simulated taps rather than fundamental breakthroughs. As app restrictions increased, the hype faded just as quickly.

The Unexpected Blow: Memory Prices

While manufacturers struggled with AI ecosystems, a harder problem emerged upstream.

From the second half of 2025, global memory prices entered a sharp upward cycle. DRAM prices surged dramatically, with LPDDR5X—the backbone of high-end smartphones—more than doubling within a year.

The cause was structural: AI training and inference consume massive amounts of high-performance memory, prompting suppliers to prioritize data-center-grade products over consumer electronics. New capacity for standard memory is unlikely to ease pressure until late 2027 or beyond.

Memory accounts for 15–20% of BOM costs in mid-range phones and up to 15% in flagships. As prices rise, manufacturers face a stark choice: raise prices, cut specifications, or both.

The irony is clear. When smartphone giants invested heavily in AI infrastructure years ago, few expected that the biggest bottleneck for AI phones would be something as mundane as a memory chip.