蜜豆视频

C1 Podcast: Opportunities in alternatives amid trade uncertainty (12:24)
C1 Video: Quick view with CIO's Sundeep Gantori on US AI vs China AI (2:01)
C1 Video: CIO鈥檚 Kelvin Tay on expectations for Trump鈥檚 2 April tariffs (1:27)
Podcast:

Technology and artificial intelligence-related names have led US equity declines in recent weeks amid an overall risk-off mood due to tariff uncertainty and growth concerns. The Nasdaq and the Philadelphia Semiconductor index are dow n 9.6% and 14% year-to-date, respectively, underperforming the S&P 500鈥檚 4.2% slide.

The introduction of low-cost models like DeepSeek has weighed on investor sentiment in leading AI stocks, while changes in data center spending have raised questions over the sustainability of big tech鈥檚 AI capex. These have prompted investors to reassess an AI rally that has run for more than two years.

However, our recent analysis shows that industry leaders with frontier models are better positioned to capitalize on the AI monetization trend, while demand for leading AI chips should remain strong.

Performance plays a more important role than costs for large enterprise customers paying for AI applications. It is worth highlighting that despite many low-cost alternatives in operating systems and office productivity tools, Fortune 500 companies are known for prioritizing performance and security over cost. Based on AI model benchmarking results compiled by the Arc Prize Foundation, a non-profit organization co-founded by prominent AI researcher Fran莽ois Chollet, frontier models have a clear performance advantage over low-cost models. This doesn鈥檛 mean there is no place for low-cost models鈥攚e believe the market is big enough for both frontier and low-cost models to grow. But in the practical world, we think expensive frontier models are likely to be adopted more among large enterprises while low-cost models will likely be more concentrated in consumer AI applications and among smaller enterprises.

AI profitability is largely decided by performance. As large enterprises are the most profitable segment for AI companies, there is a clear monetization advantage for leading US cloud platforms given their focus on the expensive frontier models. We anticipate that China鈥檚 AI leaders will focus largely on integrating AI into consumer technologies, where these companies are already dominant, such as e-commerce, gaming, and electric vehicles. Our analysis also shows that while top US cloud companies on average spend 6-8x more on capex than their Chinese counterparts, their revenue sizes are around 12x more given their large enterprise customers care less about costs and more about performance.

Leading AI chips should retain their competitive edge. The importance of performance over price is also evident in the AI compute industry, where leading chips are able to generate a much higher number of tokens per second鈥攁 measure of how fast AI models can generate a response to human requests. A token in the AI field is a fundamental unit of data that AI models use to understand and generate human language, and more tokens are required to create images relative to text-based requests. In real-world applications, the performance gaps between chips affect the way AI platforms choose them. For example, NVIDIA鈥檚 most advanced chips can generate 4x more tokens in a second compared to some of its peers.

So, while low-cost AI models are a good addition to the AI industry, we believe demand for frontier models and leading AI chips will remain strong. We continue to prefer AI semi stocks and leading cloud platforms as the technology advances further.

For more detailed comparison, refer to Intelligence weekly #33: Are low-cost AI models really cheap after adjusting for performance?