For the past few years, whenever Jensen Huang, boss of Nvidia, has taken the stage, he has followed a familiar script, unveiling ever more powerful semiconductors, software and systems for running artificial intelligence in data centres. No wonder: the AI boom has sparked a spending spree in giant server farms, and much of that money has flowed straight to the giant chipmaker. In the past four years, annual revenue at Nvidia’s data-centre division has grown from $11bn to $194bn, pushing the firm’s market value past $5trn, more than any other company ever.
But on June 1st at Computex, an annual tech-industry jamboree in Taiwan, Mr Huang did more than refresh Nvidia’s data-centre wares. He unveiled RTX Spark, a chip to be launched later this year for personal computers (PCs), built in collaboration with Microsoft, whose software they will run. Nvidia is taking on Intel and AMD, the chipmakers that dominate the segment—and is betting that the next phase of AI will play out not just in data centres, but on devices at the edge.
In recent years PCs have been one of tech’s backwaters. Evercore, an investment bank, estimates that in the past decade unit sales of chips for desktop computers have fallen by 4% a year, while those for laptops have been roughly flat. What is drawing fresh interest is the rise of “agentic” AI, software that can carry out complex tasks on its own. As it spreads, the number of tokens (the chunks of text processed by AI models) is expected to rise sharply. Model-makers and cloud providers are already straining to keep up. Offloading some work from the cloud to local devices could be cheaper and more efficient.
Nvidia argues that this will require a different sort of machine. PCs rely on central processing units (CPUs), general-purpose chips that handle everything from word processing to web browsing. CPUs may co-ordinate the work of AI agents, but the models those agents rely on need another type of chip: graphics-processing units, or GPUs, the market for which Nvidia dominates. With RTX Spark, Nvidia is combining the two types into a “superchip”. The upshot, according to Mr Huang, is that the PC is being reinvented for the first time in 40 years, replacing the old model, in which humans did most of the clicking and typing, with one in which AI agents do much of the work.
Success is far from assured. Nvidia is not a stranger to PCs; before the AI boom, it made a big chunk of its money selling GPUs for gaming machines. But in CPUs for PCs, it is a newcomer. Intel and AMD together supply more than four-fifths of all CPUs sold for PCs. As for software, Nvidia says it has worked with Microsoft for over two and a half years on the new chip. Even so, some analysts remain sceptical that developers and users will quickly embrace a new class of AI-first PC.
Yet Nvidia has plenty of advantages. For a start, it has piles of money to invest; analysts expect the company to generate roughly $200bn in free cashflow this year. It also has the clout to draw in partners and a brand that may help catch the attention of consumers. A number of the biggest PC-makers, including HP, Lenovo and Acer, have already signed up to offer the new chip in their devices.
Nvidia’s move is another sign of how fiercely contested the chip business has become, with firms venturing ever further from their original turf. In March Arm, a British designer that licenses chip blueprints to others, announced its own CPU for AI data centres; RTX Spark uses Arm’s designs for its CPU. A day after Mr Huang spoke at Computex, Lip-Bu Tan, Intel’s boss, presented his own plans. Among them was an AI chip due later this year, aimed at running models in data centres. Intel, Mr Tan said, was “going to do really well”. Mr Huang is equally confident.
To track the trends shaping commerce, industry and technology, sign up to “The Bottom Line”, our weekly subscriber-only newsletter on global business.