Alibaba's Joe Tsai says the AI race has no winner — and the US has to learn from China (www.businessinsider.com)

At the All-In Summit 2025, Alibaba chairman Joe Tsai argued that the AI contest is "a long marathon," not a winner-takes-all sprint: success will come to those who deploy and diffuse models fastest, not necessarily those who build the biggest. Tsai urged the U.S. to shift emphasis from pouring billions into ever-larger models and data-center builds toward faster real-world adoption—highlighting that models rise and fall weekly, while practical integration into products determines who actually benefits. His point underscores a growing technical divide: U.S. firms are investing heavily in hyperscale infrastructure (Meta’s multibillion forecasts, OpenAI/Oracle’s Stargate) to push frontier models, whereas China is prioritizing lean, cheaper, often open-source models optimized for edge and mobile deployment. Examples like DeepSeek’s R1—reportedly competitive yet low-cost—illustrate how model efficiency, quantization, distillation and hardware-software co-design can enable rapid proliferation. For the AI/ML community, the implication is clear: advances in inference efficiency, deployment pipelines, and integration into everyday devices may be as strategically decisive as raw model size, shifting priorities for research, product strategy, and national AI policy.