As the world competes for the future of artificial intelligence, not only innovation strategies but also market prices are shifting in ways few anticipated only a year ago. This is particularly evident in China, where Nvidia’s B300 servers have reportedly soared to almost $1 million. This staggering price point is part of a larger narrative of supply constraints, geopolitical restrictions and an ever-growing demand for computing resources that outstrips supply.
Underlying this trend is the increasing reliance on powerful AI computing hardware. Nvidia B300 servers, which contain eight high-performance GPUs, are built to support demanding AI applications, such as training sophisticated models and large-scale data processing. Not only are these servers costly, but they are essential tools for businesses looking to participate in industries such as generative AI, large language models, and automation. With growing demand, particularly from Chinese technology companies, the supply-demand imbalance has grown.
In comparison, in the US the same B300 servers cost about $550,000, already up from last late last year. But in China, the price has almost doubled to around 7 million yuan. This stark contrast is not due to regional market dynamics. It reflects the impact of rapidly tightening US export controls that curtail the supply of advanced AI chips to China. These restrictions have tighten over time, leaving it harder for businesses to access the latest equipment.
An interesting feature of this story is how channels have shifted. A grey market has long filled the supply gap, enabling Chinese companies to get their hands on banned chips. But recent efforts to crack down on chip smuggling have had a major impact. With tighter enforcement, the availability of such supply channels has become limited, leading to increased prices. Once a backdoor option, this network has become a risky and unpredictable alternative, reducing options for businesses.
Nvidia has been adamant about compliance. In a statement to the media, the company reiterated that the B300 cannot be sold in China, and partners are required to comply with the regulations. It said, “As systems grow in size and complexity, illegal diversion is a formula for disaster.” It added, “Nvidia does not support the operation of such systems and the enforcement of such actions is very effective.” These assurances show not only the legal dangers at play but also the risks of using unsupported hardware in critical systems.
Despite the limitations, the market has remained strong. Chinese tech firms continue to heavily invest in building AI capabilities, for both immediate and long-term gains. Notably, many companies reportedly have reservations about putting Nvidia equipment on their books; if they did, they would risk being caught in the crossfire of US sanctions. This is prompting more innovative strategies for accessing and using the equipment, such as joint ventures, indirect ownership, and outsourcing to third parties.
Another potential development is the adoption of AI leasing. Leases are an important option for those companies that can’t or won’t pay the high upfront costs. The monthly leasing cost for B300 servers has risen to as much as 190,000 yuan for short-term leases. This allows for more flexibility but it also highlights the fact that access to best-of-breed AI computing power is increasingly a service, not an asset.
The scarcity premium on these servers not only reflects their rarity but also highlights the strategic value of AI computing resources. Chinese internet giants are in search of the most economical means to create tokens, the smallest unit of text that AI processes. Tokens are crucial for commercialising AI services, such as chatbots and business applications. This has made the availability of fast and reliable computing power a critical element of differentiation.
Legal developments have also contributed to the supply crunch. In March, the arrest of Yih-Shyan “Wally” Liaw, co-founder of Nvidia partner Supermicro, by US authorities further heightened the focus on the ecosystem. These steps reinforce a focus on export control enforcement and deterring efforts to work around regulations. For businesses in this industry, the takeaway is that noncompliance is not an option and the stakes are high.
More broadly, this case study demonstrates the growing convergence of technology, policy and market dynamics. The increased cost of B300 servers isn’t just a price increase, it’s a control issue, a power struggle, and a game of chess in the global AI ecosystem. This presents a challenging environment for companies to navigate when making decisions.
