Community Discussion · Policy

Compute Costs Drop to 2% Off-Peak: Rewriting AI Model Pricing Logic

Yelin Does Not Eat Sponsored MealsJul 192026/07/19 61 views

I noticed an interesting detail: Alibaba Qoder's pricing for Qwen3.8-Max-Preview this time isn't just a "massive discount," but rather "starting at 2% off during nighttime hours." This time-dimension-based discount reveals deeper industry signals—AI compute is transitioning from a "scarce resource" to "infrastructure," and infrastructure pricing has always followed peak-valley patterns.

Policy

1 replies

Ctrl + Enter to reply

HuangCFOJul 29(edited)

[quote="xia_yelin, post:1, topic:1088"]

I noticed an interesting detail: Alibaba Qoder's pricing for Qwen3.8-Max-Preview isn't just a "massive discount," but rather "starting from 0.2x off during nighttime." This time-dimension discount reveals deeper industry signals—AI compute is shifting from a "scarce resource" to "infrastructure," and infrastructure pricing has always followed peak/off-peak patterns.

With 2.4T parameters, what does that mean? Compared to GPT-4's rumored 1.8T, Qwen3.8-Max-Preview has already reached the first tier in terms of parameter scale. But...

[/quote]

From a financial perspective, the 0.2x nighttime discount is essentially marginal cost pricing. As long as nighttime utilization exceeds the break-even point, every additional inference is pure profit increment, which is positive for return on assets.