Alibaba Cloud and Baidu AI Cloud Raise Prices by 30%: End of Compute Price Inversion, AI Industry Enters New Phase

On April 18, Alibaba Cloud and Baidu AI Cloud announced price increases almost simultaneously.

Alibaba Cloud’s AI compute and storage prices rose up to 34%; Baidu AI Cloud up to 30%. This officially marks the end of the “expensive compute but cheap models” price inversion era of the past two years.

Honestly, my first reaction was: finally.

How absurd was the price inversion?

Over the past two years, large model API prices kept dropping. GPT-4 was frighteningly expensive at launch; now there are “bargain basement” models everywhere. Some small vendors even advertise “free unlimited calls”—of course, funded by burning investor money.

But what about compute costs? NVIDIA H100 prices haven’t dropped. Data center electricity, operations, depreciation—none of these costs decreased.

This created a distorted value chain: upstream compute vendors made huge profits, midstream cloud vendors sold at a loss, and downstream applications got used to cheap APIs.

Obviously unsustainable.

What the price hikes mean

First, the subsidy period is over. Cloud vendors are no longer using low prices to capture market share; they’re pursuing profit.

Second, industry consolidation. Small applications living off “free APIs” may need to reconsider their business models. Without core value, businesses that simply repackage APIs will struggle.

Third, opportunity for domestic chips. NVIDIA chips are too expensive. One implicit logic behind Alibaba Cloud’s price increase: replace with domestic alternatives to reduce costs. Huawei Ascend, Cambricon, Hygon may see new opportunities.

The DeepSeek variable

On the same day as the price hikes, more details about DeepSeek V4’s architecture were revealed. 1.6 trillion parameters, Mega MoE architecture, active experts jumping from 256 in V3 to 512.

If DeepSeek can offer GPT-4-level capabilities at lower cost, will cloud vendors’ pricing strategy be undermined?

I don’t think so. Because DeepSeek itself also needs compute. Model efficiency improvements can reduce unit costs, but can’t eliminate costs entirely.

Impact on developers

If you’re an AI application developer, what does this mean?

Simple: the good times are over. The era of “unlimited API calls without worrying about costs” is gone. You’ll need to manage token consumption more carefully and seriously consider ROI.

But from an industry perspective, this is healthy. Only when costs truly reflect value can the industry develop sustainably.

Are you affected? How much will the price increase add to your project costs?