NVIDIA B300 sells for $1M

AI servers are supposed to get cheaper and easier to buy as production ramps. But Nvidia’s new B300 is doing the opposite in China. Prices there have jumped to about 7 million yuan — roughly $1 million — for a single server, nearly double a comparable U.S. price. That is the news. The reason matters more than the sticker shock: this is what happens when export controls, smuggling crackdowns, and still-ravenous AI demand all collide at once. (money.usnews.com) ### What exactly is the B300? The B300 is Nvidia’s latest top-end AI server platform — basically a dense box of compute, memory, and networking built for training and running large AI models. This is not a gaming chip story. It is infrastructure. Companies buy (money.usnews.com)roduct launch. (money.usnews.com) ### Why is China paying so much more? Because the legal supply is constrained and the gray-market supply just got riskier. The B300 cannot be sold into China under current U.S. export rules, so buyers have been relying on indirect channels. Reuters’ reporting says anti-smuggling enforcement has dried up part of that flow. When that happens, scarcity stops being theoretical and turns into a cash premium almost overnight. (money.usnews.com) ### Is this just a black-market blip? Probably not. The telling detail is that rentals have surged too — to about 190,000 yuan a month on a one-year contract. Rental prices matter because they capture real operating demand, not just collector-style panic buying. If customers are willing to lease at those levels, they are signaling that getting compute now is worth more than waiting for a cleaner supply path later. (techinasia.com) ### Why does the U.S. price matter? Because it shows how distorted the market has become. A similar B300 server in the U.S. is around $550,000, so China is paying something like an 80 percent scarcity premium. That gap is not coming from better performance in China. It is coming from access. In other words, geography is now part of the bill of materials for AI infrastructure. (tradingpedia.com) ### What does Nvidia have to do with this now? Nvidia is not just watching demand pile up. The company said on April 29 that it will report first-quarter fiscal 2027 results on May 20, with the quarter having ended April 26. So investors are heading into earnings with a live example of how constrained supply can(tradingpedia.com)revenue, but it does show how strong demand remains for the highest-end AI gear. (nvidianews.nvidia.com) ### Why are Samsung and SK hynix in the picture? Because AI servers are not just about GPUs. They also depend on advanced memory, especially high-bandwidth memory, and Nvidia’s ecosystem runs through suppliers in South Korea. A senior Nvidia executive, Madison Huang, met Samsung Electronics and(nvidianews.nvidia.com)ms. Different story on the surface — same underlying point: the bottlenecks in AI are spreading across the whole stack. (koreaherald.com) ### So what should boards and investors take from this? The old assumption was that AI hardware was expensive because it was new. The new reality is harsher — it can be expensive because it is politically gated, geographically trapped, and operationally scarce all at once. That changes procurement math. It changes how cloud capacity gets valued. And it raises t(koreaherald.com) (money.usnews.com) ### Bottom line A $1 million Nvidia server in China is not just a wild pricing anecdote. It is a stress signal from the AI supply chain. The hardware race is no longer only about who can build the fastest systems. It is also about who can actually get them.