2023-04-29 20:05:40
Behind the ChatGPT phenomenon lie tens of thousands of expensive and sophisticated electronic chips, which have become a big problem for tech giants in the race for artificial intelligence (AI), from Silicon Valley to Taiwan.
Since the generative AI interface conquered millions of people with its ability to create all kinds of text with a simple request in a simple language, the start-up OpenAI, responsible for the language model, and its main investor, Microsoft, are to the front.
“But the only one making money is Nvidia,” says Alan Priestley, a Gartner analyst, referring to the US manufacturer. Microsoft has integrated generative AI into many online services, but “hasn’t yet figured out how to generate additional revenue,” he adds.
Production of the vast majority of the semiconductors used to drive artificial intelligence models in the world depends on two companies, the Californian Nvidia, for the design, and the Taiwanese TSMC, for the manufacture.
Nvidia is known for its graphics processing units (“GPUs”), which enable you to play high-resolution video games or conduct video conferencing. These chips capable of processing large amounts of data have enabled the rise of next-generation AI.
“The versatility and power of generative AI has triggered a sense of urgency in companies around the world to develop and implement AI strategies,” Jensen Huang, chief executive officer of Nvidia, said at an analyst conference in February.
“There is a huge appetite” for these chips, confirms independent analyst Jack Gold. “Nvidia says it can supply all the chips people want. I’m not sure that’s quite the case.”
Although there is supply, not all organizations have the means to buy them. “Machines powerful enough to train generative AI models can cost hundreds of thousands of dollars,” says the expert. Not to mention power consumption.
Intel, America’s number one semiconductor company, which designs and manufactures the components, has lagged behind: “They’ve rested on their laurels,” says Gold. “They are trying to produce chips that can compete with Nvidia, but this is a new area for them, they haven’t made GPUs for at least ten years.”
The Californian group continues to produce chips dedicated to AI, less sophisticated but essential in many computer equipment, especially for the cloud.
“We are working to democratize AI,” Intel CEO Pat Gelsinger told analysts on Thursday. According to him, some AI capabilities will migrate from servers to computers, where Intel “has strengths.”
Tech giants are also working on their own components.
The specialized site The Information reported that Microsoft has been developing a chip called Athena since 2019, which is supposed to save money. Contacted by AFP, the group did not confirm it.
“Competition is getting more intense in this ecosystem,” summarizes Gold.
Last September, US President Joe Biden said that the manufacture of these increasingly sophisticated chips is a matter of “national security”, particularly in the face of China’s ambitions.
Biden had gone to inaugurate the construction of a semiconductor factory in Ohio (east) and praise a law adopted at his initiative, which released 52,000 million dollars in subsidies to revive semiconductor production.
Europe has also allocated a similar amount of funds to foster that industry. Intel has invested heavily in production and research on both continents.
Western countries seek to guarantee the sources of supply of these electronic components present in smartphones, cars, refrigerators, weapons…
“God forbid, but if China attacked Taiwan and TSMC stopped production, global chip supply would drop by 80%,” Gold warns.
But diversification will take time.
“It takes 3-4 years to build a new specialized factory,” stresses the analyst.
Eager to preserve its leadership in next-generation chips synonymous with advances in AI, the United States has also moved to restrict China’s access to essential technologies for semiconductor manufacturing.
#war #chips #race #artificial #intelligence