NVIDIA has been caught in the crossfire of the trade war between the United States and China for more than a year. Its most powerful chips could not be sold to the Asian giant because Washington required export licenses and later Beijing did not give the green light to imports. This week the two fronts have been unlocked simultaneously and Jensen Huang has taken advantage of his annual developers conference to announce it out loud. Your factories they are starting enginesand the future is brighter than ever for her.
H200 chips, unlocked. The H200 chip, NVIDIA’s second most powerful chip today, had become the center of trade and technological tensions between China and the United States. The Trump administration had already granted export licenses but of course, as long as she got her cut. Which was missing it was him approval from Beijingthat according to Reuters It has arrived this week for many of the customers who were demanding access to these chips. Among them are ByteDance, Tencent, Alibaba and DeepSeek.
China is a gold mine. Before the restrictions imposed For the US, China represented 13% of NVIDIA’s total turnover. The export veto was highly criticized by Jensen Huang, who did not stop criticize the measureof try to avoid it and to explain that what the US had done was not protect its technology, but rather shoot itself in the foot. During this blockade, Chinese companies have been advancing both in the development of AI models and in the development of their own chips. They still have room for improvement, but this effort to “become independent” from US technology is already bearing fruit and perhaps would not have occurred if it were not for the US veto.
There will also be Groq chips. NVIDIA will not only export its H200 chips, but is preparing a version of its AI inference accelerators for the Chinese market. Specifically, we are talking about Groq chips, a company in which NVIDIA invested $20 billion to “license” its technology, although In practical terms I have acquired it. These chips are especially interesting because they are not used to train AI models, but to execute and “serve” them. This is the market that is growing the fastest right now, and where competition is toughest.
But China already has inference chips. Companies like Baidu they are already producing its own inference chips, which means that NVIDIA will not enter the Chinese market from a monopoly position, but as another competitor. What is striking here is the fact that Groq chips are not refined versions nor are they adapted to this market according to the cited sources. in Reuters: They will be the same ones that companies in the US or other parts of the world use.
China will continue without access to Vera Rubin. This week NVIDIA presented a new line of products built around its next AI chips, the Vera Rubin. These chips cannot be sold to China due to current restrictions, so at NVIDIA they have found a hybrid architecture: Vera Rubin for markets where it can operate freely, and Groq as an inference component for China.
NVIDIA is brimming with optimism, and with good reason. Jensen Huang spoke precisely in his inaugural conference about how promising the company’s future looks. Previous projections spoke of medium-term revenues of $500 billion for its Blackwell and Rubin chips. Inference solutions and this “opening” to China now mean that this forecast is doubled: Huang hopes to achieve at least $1 trillion in cumulative orders by 2027a simply dizzying figure that makes it clear that NVIDIA’s business seems to be in an enviable state of health.
Image | NVIDIA
In Xataka | DLSS 5: Millions invested in AI graphics improvements so people say it looks like an Instagram beauty filter
GIPHY App Key not set. Please check settings