New Delhi, Jan. 27 -- DeepSeek-V3, a large language model (LLM) for artificial intelligence (AI) launched by a Chinese startup, has startled the world.

Its owner claims it cost under $6 million, a fraction of what American LLMs took to create.

Since it's a "reasoning" model, which involves an internal dialogue to spout answers, it could upturn the economics of the AI race. So far, success has required bearing high fixed costs on hardware and training, with relatively low variable costs on output generation.

Such a ratio makes for a winner-takes-all market, since an early mover can maximize usage cheaply once its LLM is built. But if LLMs can be made at low cost and rivalry becomes a game of sharpening an edge on "reasoning", which cons...