AI Investment Shifts to 'Soft Infrastructure' and Longevity, VCs Say

At the 2026 Smart Computing Infrastructure Innovation Conference held in Q2 2026, multiple venture capital investors outlined a significant shift in how they evaluate AI-related opportunities, moving away from narratives centered on matching Nvidia’s scale toward identifying specialized solutions and supporting long-term company survival.

Infrastructure: From Scale to Specialization

While global AI demand continues to surge, investor focus in the infrastructure space has fundamentally changed. Wang Xin, partner at Dongfang Fuhai, stated that the AI industry has entered a positive cycle where infrastructure remains dominant. However, he emphasized that the narrative of purely “benchmarking against Nvidia” no longer applies; the market increasingly favors “hidden champions” that solve specific technical bottlenecks.

Zhang Qian, founder of Tianjin Technology Investment, presented a contrarian view: once global computing infrastructure profits are captured by leading manufacturers, venture capital should focus on vertical-sector applications that achieve high ROI without consuming massive computational resources. She argued that China, as the world’s most comprehensive application economy, should pursue deep penetration across thousands of industries rather than chasing generic large language models.

Jiang Chun, managing partner at Puhua Capital, highlighted the importance of “soft infrastructure” alongside traditional hardware. He specifically identified AI safety as an emerging investment priority. As video generation technologies become more realistic, establishing data authenticity and building trustworthy data systems will become critical, he said. “We must establish a trusted data system to prevent malicious attacks and data fraud,” Jiang stated, noting that his firm focuses on underlying projects capable of building “data flywheels” and “fault tolerance mechanisms” to ensure healthy AI ecosystem expansion.

Hard Tech: Long-Cycle Adjustments

Wang Meng, partner at Junshan Capital, revealed that his firm had heavily invested in embodied AI in 2024 but shifted strategy upon entering 2025 after discovering physical bottlenecks in hardware breakthroughs. “We quickly realized that investment focus must shift from the ‘body’ to the ‘brain,’ beginning to deploy at the model and ‘world model’ layers,” he explained.

Wang also noted that as industrial deployment completes in 2026, investment interest is turning toward home robotics. The logic differs: industrial robots prioritize efficiency and cost, while home robots resemble consumer electronics requiring extreme product quality and emotional value. This shift from “hard” to “soft” reflects market recognition of technology adoption timelines.

Jiang Chun added that AI is driving radical changes on the production side. While the internet era focused on consumer-side digitization, the AI era emphasizes production-side intelligence. Extracting data from industrial systems like distributed control systems (DCS) and computer-integrated manufacturing (CIM) to form new data flywheels represents the most promising area for hard tech investment, he argued.

Revised Investment Criteria

Investors’ evaluation standards have undergone fundamental restructuring under the new industrial narrative. Qiao Yuting, general manager of OPPO Xingxing Investment, stressed that application-layer investment cannot rely on previous-generation internet thinking. AI-native teams should be “small and beautiful,” with minimal labor costs and rapid iteration. “Traditional software companies struggle to transition to AI because their organizational structures and processes are designed for labor-intensive models,” she noted, adding that investors now prioritize whether teams possess “native” AI thinking and can rapidly find product-market fit globally.

Zhang Qian stated that certain sectors have entered the 3.0 era of AI investment, where the goal is “longevity.” While first-round market valuations appear inflated, the real challenge lies in surviving market cycles. “Investors must not only survive themselves but help portfolio companies build long-term survival capabilities,” she said. “AI application potential may only be 5-10% tapped; AI’s disruption of industries is just beginning.”

Wang Meng expressed concern about “FOMO (fear of missing out) sentiment” in frontier fields like quantum computing. Although quantum computing is viewed as the next frontier, commercial viability may not arrive until after 2030. “Investors need to balance chasing trends with respecting technology timelines, avoiding premature valuation depletion on long-cycle technologies,” he concluded.