Thirty thousand euros, that’s how much the specialized graphics cards that Nvidia build cost. You will not find them in a normal PC, but you will find them in large data centers that invest in artificial intelligence applications.
The American chip company Nvidia is taking full advantage of the AI hype, whipped up by the introduction of the smart chatbot ChatGPT at the end of 2022. Since then, the demand for the specialized chips has grown so fast that Nvidia can barely handle that demand.
This week, the company announced that quarterly revenue doubled from last year. Last quarter, Nvidia sold 13.5 billion dollars (12.5 billion euros) worth of chips and the company expects the growth to continue for some time. Investors reacted euphorically and boosted the stock market value to USD 1,200 billion, more than three times as much as a year ago. By way of comparison: industry peer Intel has a market capitalization of USD 137 billion.
Processing of images
Nvidia has 27,000 employees and annual revenues of $27 billion. The company started in the 1990s with chips specialized in image processing. These so-called GPUs (graphics processing units) can handle many simultaneous computational tasks, which is a good feature for the development of artificial intelligence or AI. The processing of large amounts of data for fine-tuning the algorithms is fastest on such processors. Without Nvidia’s chips, ChatGPT and related products wouldn’t be like this clever are.
Microsoft, Google and Amazon, providers of cloud services, are preparing their data centers for AI applications. They want to sell their services to AI startups that use generative AI just like ChatGPT. This is software that creates new, derived content based on existing examples. These apps can belch credible texts, or other media such as audio, photos, or videos. It just depends on what data you let the algorithms chew on.
Read also: How Microsoft with 350 people beats the lies he ChatGPT and BingChat.
Meta, owner of Facebook, does not want to be left behind in the AI race and is investing in its own variant. Chinese tech giants such as Alibaba, Tencent and Baidu are also developing generative AI and want to buy the Nvidia chips, but due to US export restrictions, the fastest processors are not allowed to go to China. Nvidia designed a slowed down AI processor for Chinese customers that does comply with export regulations.
According to Nvidia CEO Jensen Huang, a ‘new era’ is dawning: software development in the cloud. Competitors such as AMD and Intel also make AI chips, but their market share is limited when it comes to generative AI. Nvidia also built up a lead in the software with which you control those specialized processors.
Huang isn’t just targeting the AI hype. Its specialized processors are also in demand in the automotive industry, which uses artificial intelligence to improve driver assistants or allow cars to drive fully autonomously. GPUs are also in vogue in the crypto world. Not so long ago there was a lack of Nvidia processors, because you can also ‘mine’ crypto coins with them. After 2021, the bitcoin price collapsed and with it the price of graphics cards.
Read also: Reportage TSMC: Why the whole world counts on the chips from Taiwan
Fine-meshed technology in-house
Thanks to ChatGPT, Nvidia is once again struggling to keep up with demand. At the moment, the chip designer has its processors made at TSMC in Taiwan. Only that ‘foundry’ currently has the fine-meshed technology to manufacture processors with 80 billion transistors each. That capacity is not easy to expand; that would be at the expense of TSMC’s other customers. As an excuse, Nvidia could have its chips produced by South Korean Samsung. Another option is Intel, which also wants to work on behalf of other manufacturers – even competitors. Intel had hoped to acquire the Israeli foundry Tower Semiconductor as a launching pad for this new strategy. But that also requires approval from Chinese regulators, which refused, after which Intel canceled the acquisition last week.
A version of this article also appeared in the newspaper on August 26, 2023.