Nvidia B300 servers hit $1M

Nvidia’s newest AI servers are becoming a pricing story, not just a technology story. In China, some B300 systems are now going for about 7 million yuan — roughly $1 million — which is far above comparable U.S. pricing. That jump did not happen because the hardware suddenly got twice as good. It happened because the supply path got squeezed just as Chinese demand for top-tier AI compute stayed intense. (money.usnews.com) ### What is a B300 server? A B300 server is one of Nvidia’s highest-end Blackwell systems for AI training and inference, typically built around eight B300 GPUs in a single box. This is the kind of machine companies want when they are training fron(money.usnews.com 1)(money.usnews.com 2) ### Why did the China price jump so hard? The short version is scarcity. U.S. export controls already limited China’s access to Nvidia’s most advanced AI hardware, but a big part of the market had still been getting equipment through gray channels(money.usnews.com)systems. When a product is both restricted and essential, the price stops behaving like a normal hardware price. (money.usnews.com) ### How big is the premium, really? It is huge. Reuters says a B300 server in the U.S. is around $550,000, up from about $500,000 late last year. In China, the same class of system is now around $1 million, versus about 4 million yuan late last year. So this is not just a mild markup for import risk — it is a near-doubling in a matter of months. (y94.com) ### Why are companies still paying it? Because top-end compute is still the bottleneck. If you are trying to train competitive models, launch AI products fast, or win customers before rivals do, waiting can cost more than overpaying. That is the ugly math here — a server that looks irrat(y94.com) making that trade. (money.usnews.com) ### What does this do to China’s AI market? It pushes the market in two directions at once. Big firms with cash can keep chasing scarce Nvidia gear. Smaller startups have to stretch hardware longer, rent more compute, delay training runs, or r(money.usnews.com 1)(money.usnews.com 2)ay)) ### Does this help Nvidia? Sort of, but not cleanly. The headline price spike shows Nvidia hardware is still the gold standard. But these China sales are happening in a constrained, unofficial environment where Nvidia says it does not provide service or support for such systems. So the brand stays dominant, while the company has less control over how its most coveted hardware moves. (y94.com) ### What is the real takeaway? The real story is not just that one server got absurdly expensive. It is that advanced AI hardware has become a strategic scarce asset, and scarcity now sets the rules. When the best box in the market costs $1 million in one country and roughly half that in another, the gap is no longer about shopping around. It is about power, access, and who gets to build fastest. (money.usnews.com)