Alibaba and DeepSeek Intensify the AI Price-Performance Race
Alibaba and DeepSeek introduced new AI systems that increase pressure on model developers over performance, openness and operating cost.
Why it matters: Declining model costs strengthen the case that enterprise value will shift away from raw models and toward the harness layer.
Alibaba and DeepSeek have introduced artificial-intelligence systems that increase pressure on model developers over performance, openness and operating cost.
Alibaba unveiled a large Qwen model using a mixture-of-experts architecture intended to reduce response latency and computing costs compared with activating an entire model. The system supports multimodal inputs and large context windows.
DeepSeek released a model designed around extremely low operating costs, reinforcing the price-performance competition among Chinese and U.S. AI developers.
Price comparisons require caution because cheaper models may require more steps or perform inconsistently across tasks. Even so, many enterprise workflows do not require the most capable model available. They require a model that is sufficiently accurate, inexpensive, adaptable and deployable within existing systems.
If capable models become widely available at rapidly declining prices, more economic value may move toward systems that manage proprietary data, workflows, evaluation, identity and governance.