TrendForce’s latest investigations have revealed that the recent release of DeepSeek-V3 and DeepSeek-R1 underscores an industry-wide shift toward more cost-effective AI infrastructure. This development is expected to prompt end users to conduct more rigorous evaluations of AI infrastructure investments, focusing on adopting more efficient software computing models to reduce reliance on hardware such as GPUs. CSPs are also likely to expand the adoption of in-house ASIC infrastructure to lower deployment costs. Consequently, actual demand for GPU-based AI chips and semiconductors could see notable changes from 2025 onward.
TrendForce notes that the global AI server market has experienced rapid growth since 2023. By 2025, AI servers are projected to account for over 15% of total server shipments; by 2028, they will be nearing 20%. Major CSPs have aggressively expanded their AI infrastructure in response to escalating AI training demands.
Starting in 2025, the focus will shift toward edge AI inference. Companies will adopt next-generation GPU platforms such as NVIDIA Blackwell and accelerate the development of proprietary ASICs, as seen with AWS. This strategic move aims to enhance cost efficiency and meet the needs of specialized AI applications. Meanwhile, Chinese CSPs and AI firms like DeepSeek are prioritizing the development of more efficient AI chips and algorithms to foster diversified AI applications in the face of U.S. chip export restrictions.
Historically, the AI industry has relied on scaling models, increasing data volume, and enhancing hardware performance for growth. However, escalating costs and efficiency challenges have prompted a shift in strategy. DeepSeek has adopted model distillation techniques to compress large models, improve inference speed, and reduce hardware dependencies. By optimizing the performance of NVIDIA Hopper downscaled chips, DeepSeek maximizes computational resource utilization.
DeepSeek’s competitive advantage stems from its high-performance hardware selection, innovative distillation techniques, and an open API strategy. This approach balances technological innovation and commercial viability while reinforcing the AI industry’s push toward greater efficiency.
TrendForce notes that China's AI market is expected to develop in two key directions in light of ongoing U.S. chip export restrictions. First, AI-related companies will accelerate investments in domestic AI chips and supply chains. Large Chinese CSPS, for instance, will continue procuring available H20 chips while also ramping up the development of proprietary ASICs for deployment in their data centers.
Second, China will leverage its existing internet infrastructure to compensate for hardware limitations with software-based solutions. DeepSeek exemplifies this approach by breaking from conventional methods and adopting model distillation technology to enhance AI applications.
Overall, as the U.S. government potentially tightens AI and semiconductor restrictions on China, domestic AI firms will be compelled to accelerate the development of proprietary AI chips and HBM hardware. While these solutions may not match the performance of NVIDIA’s GPUs, they are primarily designed to support China’s domestic data center infrastructure, where individual chip performance is no longer the sole priority. Additionally, companies like DeepSeek are advancing AI multimodal models, aiming to achieve similar performance in specific application areas at lower training costs to expedite commercialization.
For more information on reports and market data from TrendForce’s Department of Semiconductor Research, please click here, or email the Sales Department at SR_MI@trendforce.com
For additional insights from TrendForce analysts on the latest tech industry news, trends, and forecasts, please visit https://www.trendforce.com/news/
Subject
Related Articles
Related Reports