Nvidia Shares Climb to $237 as Blackwell Chips Power OpenAI's Faster GPT-6 Astra Ultrafast Inference Mode

Nvidia shares rose 1.4 percent on Monday, extending a rebound that has put the chipmaker back near a $6 trillion valuation after it said its Blackwell processors are running OpenAI's faster GPT-6 Astra Ultrafast service.

The stock traded at $237.11, up $3.16, or 1.35 percent. On Oct. 1 it closed at $230.86, up 1.09 percent, after trading as high as $232.29. By Oct. 2 the shares had touched an intraday high for the first time since May and closed up 1.3 percent, still short of the May record, as the company's market value approached $6 trillion. The stock had rallied about 23 percent from a late-July low after a selloff that erased more than $1 trillion of value.

The product claim is inference, not a new training cluster. Nvidia said GPT-6 Astra Ultrafast runs on Blackwell graphics processors and is available through the OpenAI application programming interface and to eligible ChatGPT Work and Codex users. The company said the Ultrafast mode can generate tokens up to eight times faster than Astra Standard. The target is work in which delay stacks: coding agents that write code, call a tool, read the result and decide again. Training built the last two years of Nvidia's revenue. Serving those models, repeatedly, is the bill that has to keep growing if the stock is to hold a valuation near $6 trillion.

OpenAI's chief technology officer for compute, Uday Ruddarraju, said in the announcement that the company's internal models were used to optimize inference on Nvidia chips. That is a loop: OpenAI uses its own models to speed the hardware, and Nvidia supplies the hardware. It is not a disclosure of how many processors the service uses or what revenue Nvidia books from it.

A separate, larger commitment is still only a plan. Nvidia has said it intends to invest up to $100 billion in OpenAI as infrastructure is deployed. Monday's move is the narrower item, a named model on named chips. The $100 billion figure is capacity and capital that have not all been spent.

The Monday gain is small next to the July repair. A 1.4 percent rise on a stock this large is tens of billions of dollars of market value, and it is also a session that did not need a new contract. Buyers have been putting money back into the name since the late-summer low, on the view that Blackwell demand for inference will follow the demand that filled the training buildout. The OpenAI deployment is one public example. It is not a forecast.

The risks are the ones already in the price. Export rules can cut the China piece of the market. Customers including Google, Amazon and Microsoft are designing their own accelerators. A service that runs eight times faster can also mean fewer chips per query if the speed gain is efficiency rather than volume. Nvidia did not say which. Power and data-center construction remain the constraint on how fast any of this ships.

At $237 the stock is trading the claim that Blackwell is the default engine for the fastest commercial models, and that inference will be large enough to justify a value near $6 trillion. The OpenAI line supports the first half. The second half is still a buildout, a $100 billion intention, and a market that sold off $1 trillion of this stock once already this year.