Cloud Giants Jack Up AI Compute Prices: The End of Inverted Pricing

Honestly, when I saw the news that Alibaba Cloud and Baidu AI Cloud announced price increases on the same day, my first reaction was—it’s about time.

On April 18, both cloud providers released pricing adjustment announcements. Alibaba Cloud raised prices by up to 34%, while Baidu AI Cloud went up to 30%. The increases targeted AI compute and storage products, with timing that felt strategic—right before Q1 earnings season.

The signal is clear: the past two years of “models getting cheaper while compute gets more expensive” is over.

Why the Inversion?

From 2024 to 2025, the price war for large language models was brutal. GPT-4’s API price dropped from $0.03 per 1K tokens to $0.005, and Chinese models drove prices down to “cabbage levels.” For a while, everyone had this illusion—AI is getting cheaper.

But compute? GPU prices didn’t drop. Instead, they climbed higher due to exploding demand. H100 rental rates once hit $8 per hour, and you still couldn’t always get one.

This created a bizarre situation: cloud providers were losing money to gain market share, while model companies were slashing prices to grab users. The profit gap? All captured by hardware vendors, mainly NVIDIA.

I chatted with a friend in cloud services last year. His exact words: “We’re subsidizing model price wars with compute profits. This model isn’t sustainable.”

Why Now?

April is an interesting timing choice.

First, Q1 earnings are coming up. Cloud providers need to answer to shareholders—they can’t keep crying “strategic losses” forever.

Second, demand for large models has cooled since early 2026. Not fewer users, but most early adopters have already boarded. The remaining ones are either still观望 or have already burned out.

Cooling demand means less room for price wars. Raising prices now faces less resistance than last year.

There’s also a technical reason: new inference frameworks (like optimized vLLM versions) have significantly improved GPU utilization. Cloud providers’ compute costs are actually dropping. Raising prices now means better margins.

What Does This Mean for Developers?

If you’re a small team or indie developer, this price hike will hit you directly.

First, API costs. While this increase targets compute prices, API providers (like Zhipu AI and Moonshot) will likely follow suit. Their underlying costs just went up.

Second, deployment costs. If you’re using cloud GPU clusters, elastic compute will get more expensive. If you’re on serverless, the impact will be even more noticeable.

My advice: before prices go up across the board, audit your compute consumption. Can a smaller model handle your tasks? Can prompt optimization reduce token usage?

A friend switched their customer service system from GPT-4 to Claude Haiku last month. Costs dropped 70%, and performance actually improved. That kind of optimization is worth exploring now.

What’s Next?

My personal take: this is just the first wave.

Compute prices returning to normal means the entire AI supply chain’s pricing logic needs to be rebuilt. Models won’t keep dropping prices indefinitely, and services won’t stay free forever.

For the industry, this is a good thing. The “lose money to gain users” model isn’t sustainable. Everyone needs to make money for the industry to grow healthily.

For developers, it means we need to start counting costs seriously. The days of “APIs are cheap, let’s just call them freely” might really be over.

Then again, this might not be a bad thing. Normalized compute prices will let truly valuable AI applications shine, rather than surviving on subsidies. Long-term, this is a sign of industry maturity.

Just wonder—who’s raising prices next?