If there is a company that has capitalized on the artificial intelligence boomthat was NVIDIA. Your chips have become a fundamental piece to train and run many of the models that underpin the current rise of generative AI. At the GTC conference held in San Joséits CEO, Jensen Huang, went so far as to project at least $1 trillion in backlogged orders for its chips. Meanwhile, a new map of competitors begins to form around the company.
The message. Huang put figures on what the company is going through. The executive explained that the expected demand for Blackwell chips, like the B300and Rubin architecture could reach at least one trillion dollars in accumulated orders by 2027. Just a year ago that estimate was around 500,000 million, as he recalled during his speech.
The transformation. For decades it was known primarily for its gaming GPUs, but its architecture ended up fitting the needs of machine learning perfectly. This turn has transformed the company: according to data cited by the Associated Pressits annual revenue went from $27 billion in 2022 to $216 billion last year, driven largely by demand for infrastructure for artificial intelligence.
Changes are coming. Much of the growth of artificial intelligence in recent years has been based on model training, a process that requires enormous amounts of calculation and where NVIDIA GPUs have clearly dominated. However, the sector is increasingly beginning to look towards inference, the stage in which already trained systems produce answers for users. According to analysts cited by Reutersthat change could expand the number of competitors capable of executing those loads.
Big tech companies move. Companies that for years have purchased large quantities of NVIDIA GPUs are simultaneously investing in their own artificial intelligence accelerators. Amazon has developed its Trainium family of chips to train and run models in your cloud, Google continues to expand its line of TPUs and Goal works on several generations of its MTIA accelerator to sustain your AI loads.
The Chinese front. The pressure on NVIDIA is not limited to the United States or Europe either. Trade restrictions imposed by Washington have reduced Chinese companies’ access to some of their most advanced chips, accelerating the search for local alternatives. Huawei, for example, prepares mass shipment of its 910C chip for artificial intelligence customerswhile several companies in the country promote alliances to reinforce a domestic ecosystem that encompasses chips, models and infrastructure.
The general photography. There is no doubt that NVIDIA remains a fundamental piece of the infrastructure that fuels the rise of artificial intelligence, and its demand forecasts reflect that dominant position. But at the same time, the market is beginning to move in several directions: large technology companies, new companies and national ecosystems are working to build alternatives that reduce their dependence.
Images | NVIDIA
GIPHY App Key not set. Please check settings