TLDRs;
- Nvidia B300 servers surge to $1M in China amid supply squeeze
- Export restrictions and smuggling crackdown tighten availability significantly
- Chinese firms pay steep premiums due to limited AI chip access
- Rentals also rise as companies seek alternative compute access
Nvidia’s B300 AI servers have reached an unprecedented price level in China, now selling for around 7 million yuan, or approximately $1 million per unit.
The sharp rise reflects a tightening supply environment driven by both regulatory restrictions and sustained demand from Chinese tech firms racing to secure high-performance computing capacity.
Just months earlier, late last year, the same systems were priced at roughly 4 million yuan ($585,000). In the United States, the equivalent hardware sells for about $550,000, highlighting a widening regional price gap fueled by trade restrictions and market distortions.
Export Controls Tighten Supply
A key driver behind the surge is the continued restriction on Nvidia’s most advanced AI hardware. The B300 servers are not permitted for sale in China under current US export rules, limiting official supply channels.
As a result, availability has become increasingly constrained. Even as demand for AI computing power rises sharply, legitimate access to top-tier Nvidia systems remains restricted, creating significant pricing pressure within unofficial and secondary markets.
Smuggling Crackdown Reshapes Market
The price spike has also been amplified by a sustained crackdown on illegal chip trafficking networks. Authorities have intensified efforts to disrupt large-scale smuggling operations that previously helped move restricted AI hardware into China through complex supply chains.
These networks often relied on third-party intermediaries in Southeast Asia and elaborate shipment structures to bypass inspection systems. As enforcement tightens, black market supply has been squeezed, reducing available inventory and pushing prices higher.
Reports have also highlighted the scale of past diversion schemes involving billions of dollars in high-end server equipment routed through indirect channels before ending up in China.
Rentals Rise as Alternative
With outright purchases becoming increasingly expensive, some Chinese firms are shifting toward renting Nvidia-powered computing infrastructure instead. Rental prices for B300-based systems have reportedly climbed to around 190,000 yuan ($28,000) per month for one-year contracts.
This reflects a broader adaptation strategy among companies that still require access to advanced AI compute but cannot reliably source or legally import the hardware.
The shift toward rentals also underscores how critical Nvidia’s hardware remains for large-scale AI model training, despite growing domestic competition in China’s semiconductor and AI infrastructure space.
AI Demand Keeps Pressure High
Underlying the price surge is persistent demand for high-performance computing in China’s rapidly expanding AI sector. Even as local chipmakers scale up, they still lag behind Nvidia’s top-tier GPUs in training capability for frontier models.
As a result, many firms continue to rely on US-designed hardware, even at significant premiums. This dependence has helped maintain strong demand pressure despite export controls and geopolitical friction.
At the same time, China’s domestic ecosystem is accelerating. Companies like Huawei and Baidu are expanding their AI infrastructure offerings, while local suppliers continue gaining market share. However, industry analysts note that the most advanced AI workloads still largely depend on US hardware architectures.
Market Outlook
The widening price gap between official and restricted markets suggests continued volatility ahead. As export controls remain in place and enforcement tightens, Nvidia’s high-end AI systems are likely to stay scarce in China.
For now, the combination of strong AI demand, constrained supply, and smuggling disruptions continues to push B300 server prices into record territory, cementing their status as some of the most sought-after computing assets in the global AI race.


