Beyond the sometimes impressive real advances, the considerable hype and the legitimate concerns, it seems that artificial intelligence, seen through the narrow lens of semiconductors, is on the verge of a third phase. After years of study and preliminary work, and an initial boom that made Nvidia’s fortune in the cloud, a certain democratisation is underway. This is initially the case for the ‘big’ processors and accelerators used in servers, with the rise of heavyweights like AMD and the emergence of credible start-ups like Cerebras, spurred on by clients anxious not to be tied to a single supplier free to set its margins as it pleases – in short, the aim is not to reproduce with Nvidia and AI what benefited Intel so much with PCs. Of course, this will be a long process and for the time being, Nvidia remains a safe choice, like the old saying “nobody gets fired for buying IBM“.
But democratisation is also taking place at the edge of networks. Increasingly powerful embedded processors and microcontrollers, equipped with AI accelerators, are beginning to be able to handle fairly complex AI algorithms without the help of the cloud. Here too, the competition promises to be fierce. Already, the major microcontrollers manufacturers have realised that their salvation lies in providing sophisticated development tools, sometimes free of charge, in order to attract developers – not all of whom are familiar with the complex charms of AI. Will these chip manufacturers make money out of it ? Nothing is less certain. Not only do we not yet really know what this AI could be used for on the consumer side, but, above all, it’s a safe bet that if these consumers ever agree to pay for an AI service, it will be through a software application (for games, photo editing, office automation, etc.) and not by overpaying for a smartphone or another physical device. For the time being, ST, Renesas, NXP, Microchip and others are groping their way through the artificial intelligence investment process in order to stay in the race, the course of which is still uncertain, and to avoid losing market share rather than to make an immediate profit.