- Chinese AI centers disassemble 4090D GPUs and sell them by
- To resell inactive GPUs provide faster profits than waiting for rental returns
- Some say it’s overcapacity others claim preparation for newer tech
Some Chinese AI data centers disassemble and resell China-specific NVIDIA RTX 4090D GPUs, have reported claimed.
A report from Digator Asia Says these 48 GB cards, designed to bypass US export restrictions on the flagship AD102 Gaming GPU, and originally deployed as part of China’s AI infrastructure push, are now deducted from racks, renovated and sold on the open market.
Data Center Operators reportedly finds that this provides a faster and more profitable return than waiting three to five years to recover their investment through GPU rental.
An alternative theory
Each RTX 4090D sells for between CNY20,000 and CNY40,000 (around US $ 2,735 to US $ 5,470), and although the cards are easily used, the cards require changes to consumer resale.
This typically means converting them from fan-style to fan-style coolers, which are better suited for dense server environments, but less effective for single-GPU use.
Digator Asia Says this step reflects a deeper financial pressure, with many AI data centers struggling to keep their heads over the water in the middle of low demand.
According to the report, data centers need the utilization rate of at least 70% to break evenly – but the current rates are often below 20%. It leaves expensive infrastructure that is idle while repaying loan winds weaves.
This is not an isolated case. As we recently reported, China’s rapid AI infrastructure enlargement – has been encouraged by state policy – led to superstructure.
Hundreds of data center projects were launched across the country in 2023 and 2024, but the actual use is hung behind expectations. It is therefore no surprise that developers are now reading hardware to reduce losses.
Although overcapacity is probably at least part of the reason behind the sale, some operators can simply clear room for newer technology.
With interest moving from large -scale model education to real -time inference, older training -focused systems may no longer be as relevant as they once were.
Combined with the latest US export restrictions affecting chips such as NVIDIA’s H20, Chinese data centers will look at relieving old hardware and switching to inferable setups.
