For most of its life Nvidia was a company gamers loved and the rest of the world ignored. Its graphics cards rendered the explosions in your favourite shooter, and that was the business. The clearest sign of how completely that era has ended came at CES 2026, where chief executive Jensen Huang took the stage and announced no new GeForce gaming card at all. Not one. The omission was deliberate, and it spoke louder than any product could have. Nvidia is now the first company in history valued above 5 trillion dollars, and gaming is no longer what it is about.
The numbers behind the pivot are almost hard to credit. In early 2014, data centres and AI made up just 5 percent of Nvidia's revenue, and gaming was its single largest segment. Twelve years on, the ratio has flipped: data centres and AI now account for more than 90 percent of sales, a segment that has grown 1,300-fold, from 57 million dollars a quarter to over 75 billion. Since ChatGPT's debut in late 2022, that revenue has doubled roughly every eleven months. The company that once sold to teenagers now sells to hyperscalers and governments, and it captures something like 85 percent of the global market for AI chips.
What Huang spent his keynote on instead was "physical AI," his term for systems that do not just generate text and images but act in the world. The idea is to train models inside simulated environments, feeding them synthetic data until they learn how physics behaves, then drop them into real machines. He showed Cosmos, a foundation model that simulates environments and predicts movement, and Alpamayo, a reasoning model built for autonomous driving, with a Mercedes-Benz running Nvidia-defined driving on stage. Most striking, Nvidia said it will test its own robotaxi service as soon as 2027, a move from quiet supplier to active competitor in the self-driving race.
The other half of the strategy is to make Nvidia unavoidable in the data centre. The star was Rubin, pitched not as a chip but as a whole system, GPUs, CPUs, networking and storage co-designed to handle models past a trillion parameters. That same week, at HPE's Discover conference, the two companies expanded a joint "AI factory" line aimed at enterprises moving so-called agentic AI into production, complete with a new CPU built specifically for the constant tool calls and orchestration that autonomous agents demand. The message is consistent: if a workload can be trained, simulated or automated, Nvidia wants to sell the entire stack it runs on.
There is a strategic elegance here, and a familiar risk. By selling systems rather than parts, and by pushing its own open models that happen to run best on its hardware, Nvidia keeps customers inside its world from chip to software. But a valuation that assumes AI demand keeps doubling year after year is a tall bet, and the same energy and supply bottlenecks Huang cites as reasons to buy Rubin are also the constraints that could slow the whole industry he depends on. The gamble that physical AI, robots and self-driving will be the next leg of growth is real money placed on a future that has not arrived. For now, the company has made one thing clear. It is no longer driven by the people who play games, but by everyone trying to automate the things that move.