The tech industry entered 2025 convinced it was on the verge of another leap forward. New devices promised lighter hardware, smarter AI, and more “revolutionary” user experiences. Yet as the year unfolded, a different pattern emerged: products once celebrated at launch slid rapidly from praise to public frustration.
The turning point came when IT Times released its 2025 “Red and Black List,” separating products that genuinely earned user approval from those that failed to meet expectations. The most striking lesson was not that innovation stalled, but that many companies mistook marginal tweaks and ambitious narratives for real progress. The market, increasingly transparent and data-driven, pushed back.
Where the promise collapsed
Several products illustrate the problem clearly.
Apple’s iPhone 17 Air, launched in 2025 as a lighter and more affordable flagship, leaned heavily on AI as its selling point. In practice, users found that the much-advertised AI photography amounted largely to color adjustments. Under low-light or backlit conditions, images showed noise and lost detail, lagging behind similarly priced Android rivals. Battery-life “AI optimization” also disappointed: the system consumed significant power itself, leaving overall endurance slightly worse than the iPhone 16. What was marketed as innovation felt like minor tuning wrapped in new branding, and sales fell short of expectations.
Google’s Galaxy MR mixed-reality headset aimed to challenge the XR market with AI-driven spatial interaction. Instead, users encountered inaccurate environmental recognition, lagging gesture controls, frequent crashes during multitasking, and a reliance on Wi-Fi streaming for its so-called “unlimited computing power.” Once untethered, the device’s battery lasted only about 1.5 hours. Weighing 545 grams, the headset caused discomfort, dizziness, and fatigue, directly contradicting launch claims that it was lighter and more comfortable than Apple’s Vision Pro. Many users concluded that simpler AI glasses or all-in-one headsets were more practical.
The Shanji AI Paipai Glasses A1, marketed as an entry-level, lightweight first-person camera for beginners, also struggled in real use. Although it supported AI night enhancement, noise reduction, and voice control, autofocus was slow—sometimes taking five to six seconds even for static subjects. Voice commands often lagged or misfired, testing users’ patience. Design issues compounded the problem: heavy components packed into thick temples caused imbalance and slippage, making the glasses uncomfortable to wear. Improving image quality required higher-end hardware, raising costs and undermining its beginner-friendly positioning.
Fujifilm’s X Half camera highlighted a different misstep. At just 240 grams with a 1-inch sensor and built-in filters, it appeared ideal for street photography. However, priced at 4,999 yuan, it competed with smartphones that already offered similar features. Fujifilm argued that its custom 32mm f/2.8 fixed lens was the real appeal, but critics pointed out that such limitations reflected old technological constraints, not modern creative freedom. As many users put it bluntly: they wanted a camera with a retro feel, not a retro machine.
ByteDance’s much-discussed Doubao phone, developed around the Nubia M153, revealed how ecosystem realities can undermine bold AI concepts. The phone’s AI assistant could theoretically handle cross-app tasks such as price comparison and purchasing with a single command. In reality, many major apps—WeChat, JD.com, Taobao, Meituan—restricted data access to protect advertising revenue. Even when it worked, results were inconsistent: user preferences are subjective, and the system struggled to reproduce them reliably. Performance issues such as overheating and slowdowns further eroded trust, making manual app use faster than automation.
Perhaps the most symbolic disappointment was GPT-5. Sam Altman had teased the model in 2024 as superior to all predecessors. When OpenAI released it online on August 8, 2025, the company claimed major advances in programming, mathematics, writing, and healthcare, with improved “active reasoning.” Users instead reported slow responses, basic errors in math and spelling, and fabricated information in open-ended answers. Many called for a return to GPT-4o. Analysts, including researchers at MIT, argued that GPT-5 represented a refined product rather than the revolutionary leap promised, constrained by the growing scarcity of high-quality internet training data.
A market that no longer buys narratives
Across these cases, the failures share a common thread: expectations rose faster than genuine capability. Consumers no longer accept incremental fixes repackaged as breakthroughs. In an era of abundant information and comparative reviews, shortcomings surface quickly.
The backlash also reflects a broader shift. Users increasingly judge technology by whether it saves time, reduces friction, or clearly outperforms existing tools—not by parameter counts, concept demos, or launch-stage storytelling. AI, in particular, has entered a phase of sober reassessment. It remains powerful, but its limits are now visible, and overreliance without critical judgment carries risks.
As 2026 approaches, the lesson for manufacturers is stark. Products that fail to align innovation with real user needs will not survive on hype alone. The next cycle will reward companies that deliver tangible improvements—and punish those that mistake marketing noise for progress.
