DeepSeek V4 once again sparks "internal competition upgrade" in the AI circle? Is this market trend taking off or just a false move!

This time, DeepSeek released a preview version of V4, which is a bit like "the fitness fanatic of the tech world": not only more powerful but also more cost-efficient. 1 million tokens of long context + a more memory-efficient attention mechanism directly lower the barrier for "AI that can't be affordably used." The question is—will this ignite a new wave of AI market activity?
My judgment: there will be a structural revaluation, but not a full-blown celebration.
The market has already been driven by NVIDIA once, now it’s more about "who can reduce costs." DeepSeek’s significance lies in rewriting the formula "high performance = high cost" into "high performance ≈ affordable," which is a reshaping of valuation logic for the entire industry.
Who is more attractive: domestic computing power or application side? In the short term, focus on computing power; in the long term, focus on applications. Computing power is like selling shovels—those who sell them profit first when the market heats up; applications are like gold prospectors—ultimately deciding who can sustain profitability. Especially as the Ascend ecosystem gradually takes shape, the combination of "domestic computing power + open-source models" is moving from concept to closed loop.
In one sentence: the AI market is not gone, but it has evolved from "competing on computing power" to "competing on efficiency."