How did a small developer of graphics cards for gamers suddenly become the third most valuable firm on the planet? | John Naughton

AI SaaS

A funny thing happened on our way to the future. It took place recently in a huge sports arena in San Jose, California, and was described by some wag as “AI Woodstock”. But whereas that original music festival had attendees who were mainly stoned on conventional narcotics, the 11,000 or so in San Jose were high on the Kool-Aid so lavishly provided by the tech industry.

They were gathered to hear a keynote address at a technology conference given by Jensen Huang, the founder of computer chip-maker Nvidia, who is now the Taylor Swift of Silicon Valley. Dressed in his customary leather jacket and white-soled trainers, he delivered a bravura 50-minute performance that recalled Steve Jobs in his heyday, though with slightly less slick delivery. The audience, likewise, recalled the fanboys who used to queue for hours to be allowed into Jobs’s reality distortion field, except that the Huang fans were not as attentive to the cues he gave them to applaud.

Still, it made for interesting viewing. Huang is an engaging speaker and he has built a remarkable company in the years since 1993, when he first sketched his idea for Nvidia in a Silicon Valley diner. And the audience were in awe of him because they regard him as a man who saw the future long before they did, and hoped to catch a glimpse of what might be coming next.

And in this they were not disappointed. What’s coming next is Nvidia’s Blackwell B200 chip, complete with its 208bn transistors, and the family of monster machines that it will enable, including a formidable supercomputer that fits into a rack and has almost two miles of copper cabling neatly intertwined in its innards. Cue wild applause.

Watching this spectacle, the thought that came to mind was this: how did a small company specialising in graphics cards for gamers come to be the third most valuable company on the planet? And how did it happen so quickly at the end? After all, Nvidia was only worth $278bn in October 2022 and is now worth $2.3 trillion, trailing only Apple and Microsoft.

It’s a good story, and no doubt someone is already working on the screenplay. But even a cursory account produces a picture of a company that from the beginning was good at anticipating the needs of a particularly demanding class of users – gamers – and eventually realised that in developing processors that could address their needs it had produced a new kind of computer: a graphics processing unit (GPU) that could perform many calculations in parallel, as opposed to conventional CPUs that did everything serially.

A pivotal moment came in 2013, when Huang decided that GPUs could be useful for an emerging technology called machine learning, and that henceforth the company would focus on that. It was a bold bet at the time, and initially Wall Street thought it foolish. But when machine learning really started to take off and desperately needed parallel processing machines to handle the associated heavy computation, Nvidia hit the jackpot. If you wanted to do this kind of AI then you needed Nvidia GPUs – lots of them. And, more importantly, you needed a way to enable them to work seamlessly together – a kind of operating system. A Stanford software genius named Ian Buck created one for Huang. They called it CUDA (for compute unified device architecture), and from then on buying Nvidia kit became a no-brainer for anyone aspiring to get into the AI business.

Which is how Huang found himself virtually the only trader with a supply of ready-made shovels in the biggest gold rush in tech history. In February his company reported record quarterly revenue of $22.1bn, up 22% from the previous quarter and up 265% from a year ago.

The Blackwell B200 chip is his latest super shovel. And in his keynote, Huang unveiled what can be done with it. He rolled out the DGX GB200 NVL72 (Nvidia doesn’t do human-friendly labelling), which is a powerful supercomputer with 72 Blackwell processors in a single water-cooled rack. I couldn’t find any information about pricing, but then I guess that if you have to ask you can’t afford it.

Google, can, though. So can Microsoft, Meta, Oracle, Tesla, Amazon and Dell. From what their bosses say, they’ve already joined the queue for supplies. In this case, Huang has indeed seen the future. And it works for him. Whether it works for the rest of us, though, remains to be seen.

What I’ve been reading

Forget me not
A thoughtful essay by René Walter in his Substack magazine, Good Internet, on the intrinsic hostility to cultural memory in a digital world.

The politics of technology
A terrific review by Bill Janeway on the online opinion site Project Syndicate of Daron Acemoglu’s and Simon Johnson’s great book, Power and Progress.

Doubting Thomas
Jia Tolentino in the New Yorker (paywall) doesn’t like the Fat Controller in the railway stories of Thomas and Friends. Neither do I.

AI SaaS

Leave a Reply

Your email address will not be published. Required fields are marked *