During CES 2026, Wu Jun shared his perspectives on artificial intelligence. The event attracted 148,000 global entrepreneurs, investors, analysts, and media, all enthusiastically discussing AI opportunities. However, Wu remained calm and relaxed amid the fervor.

Wu Jun possesses a multifaceted background: a computer science PhD from Tsinghua University and Johns Hopkins, former Vice President at Tencent and research scientist at Google, founder of venture capital firm Fenyuan Capital, chairman of the Silicon Valley Innovation and Entrepreneurship Forum (SVIEF), and bestselling author of books including "The Wave of the Top" and "The Beauty of Mathematics." This diverse experience grants him broad perspective across both geographical and temporal dimensions.

He measures current trends against historical patterns. While many cheer for embodied AI, he calls it a "false proposition." When AI entrepreneurs dream of unicorn status, he predicts that "the vast majority of AI companies will inevitably fail." This may sound harsh, but Wu's views stem from historical cyclical patterns and business realities.

The discussion covered why robots need not resemble humans, a sober assessment of Chinese AI talent growth, advice for AI companies expanding globally, and the possibility of an AI bubble bursting around 2028.

Wu stated that bubbles will eventually burst and most current AI companies will disappear. However, this is not alarming—the internet era emerged through a similar process.

When asked about CES 2026 highlights, Wu noted that autonomous vehicles and robotics were key focuses, as physical products allow direct comparison. Traditional consumer electronics lacked innovation, with many features achievable over a decade ago. The industry urgently needs revolutionary form factors.

Regarding AI companion hardware, Wu believes it is premature and fundamentally misguided. He argues humans seek companionship from other humans, and machines may even cause irritation. The most valuable interactions will remain human-to-human.

On embodied intelligence, Wu dismisses it as another false proposition. He suggests humans narcissistically want robots to mimic human form, despite human senses and physical capabilities being inferior to many species. Robots should be designed for specific scenarios—like patrol dogs or wheeled transporters—rather than imitating human morphology. Current robots also lag significantly in sensory perception; human skin can detect one-gram touches, while robots require kilogram-level force, increasing risk of accidents.

Why then does the industry persist with humanoid robots? Wu explains they serve as technical "exams" to demonstrate flexibility and simulation capabilities, but practicality and cost remain major hurdles. A robot dog can cost 200,000 yuan, while humanoid robots reach 400,000–500,000 yuan. Simple wheeled devices for tasks like hotel delivery cost under 10,000 yuan and suffice for the purpose.

Comparing robotics industries, Wu says China exhibits a "pack of wolves" effect—producing multiple quality products consistently, indicating technological maturity. The U.S. lacks this advantage. China's edge stems not from past industrial foundations but from diligent professionals and efficient supply chains, like Shenzhen's next-day 3D printing parts delivery.

Despite 50 billion yuan in embodied AI funding during 2025's first three quarters—2.5 times year-over-year—Wu considers this modest compared to U.S. raises, such as Databricks' $4 billion or OpenAI's $6.6 billion. High R&D and talent costs justify current funding levels.

On robotics demonstrations like dancing, Wu says they only prove joint mobility without practical value. Real applications require scenario-specific designs.

In AI competition, the U.S. leads in foundational models, with Google's Transformer architecture underpinning most large language models. Chinese firms like Alibaba and DeepSeek match second-tier U.S. companies but cannot rival Gemini. China excels in Chinese language processing and model efficiency optimizations. However, most domestic models still rely on Meta's open-source Llama, and the U.S. dominates the platform layer. Robotics is China's stronger domain.

Regarding Chinese talent prominence in Silicon Valley AI giants, Wu attributes this to solid computer science and math foundations from top Chinese universities. U.S. "political correctness" has admitted unqualified students, while capable students are steered toward leadership roles, reducing technical talent proportion. AI development requires strong math and programming skills, which many Americans previously avoided. Additionally, over half of Silicon Valley tech employees are immigrants, with Chinese expertise becoming more visible in AI, similar to their historical roles in semiconductor firms like Intel and AMD. Figures like Jensen Huang (NVIDIA), Lisa Su (AMD), and Hock Tan (Broadcom) highlight Chinese leadership.

Wu distinguishes between executing theoretical programming tasks at large firms and groundbreaking creativity. All major language models build on Transformer and deep learning research pioneered by Western scientists like Hinton, Bengio, LeCun, and Hassabis—none Chinese. Optimizing existing technology differs fundamentally from original invention.

On AI startups targeting global markets from day one, Wu stresses founder vision: targeting 20% of the global market via China or the entire world. Silicon Valley firms internationalize early, while post-2010 Chinese companies like TikTok, Cheetah Mobile, and iHealth achieved vast markets through global strategies.

For AI companies expanding overseas, Wu advises designing products for simplicity beyond Chinese user habits. WeChat's complexity hindered its globalization, whereas U.S. social media and Apple products need no manuals. Multilingual and multicultural support is critical; adapting a product for one country requires effort level 1, but global expansion demands 10 times the input. Without early preparation, later internationalization becomes extremely difficult.

Addressing AI's high costs and commercialization challenges, Wu notes bubbles are inevitable in technological waves—from British railways and U.S. electricity to internet booms. Most companies vanish, leaving few survivors like Bell in telephony or three auto giants in the U.S. He told Luo Zhenyu that most internet or AI companies would die within 1,000 days—a certainty—though who remains is unknown. However, industry progress continues regardless; internet bubble losses pale against current single-company revenues. Even 50 billion yuan in robotics losses would be negligible if one or two firms reach Xiaomi's scale. AI development remains robust, and Chinese venture capital should increase AI investments.

Describing 2025 AI development, Wu chooses "rapid growth." For 2026, he says "still rapidly growing" or "rising steadily"—AI is just beginning, not yet peaking. If a bubble bursts, 2028 is most likely, as U.S. election years often spur market frenzy, presenting a critical test for AI sustainability.