The AI boom began, in a way, in a Canadian bedroom, in 2012.
Alex Krizhevsky, a promising graduate student at the University of Toronto, was conducting an experiment that many saw as futile. It was about the programming technique "artificial neural networks", which can be simply described as a way to imitate the nerve cells of the brain. At this time it was freezing cold. Few thought it would lead to any meaningful AI development.
But Alex Krizhevsky wanted to give it a chance. He built an image recognition system—software that can tell whether an image is a bicycle or a cat—and began training it by analyzing tens of thousands of images.
The computer that did it was in his bedroom, buzzing and drawing electricity.
To cope, he had done something odd: He had bought hardware from a company best known for selling graphics cards to computer gamers: Nvidia. The graphics card's chip could also be used for that kind of calculation, he reasoned.
The result was an explosion. When Alexnet, as the system was called, entered an annual competition for machine learning enthusiasts, it completely outclassed everyone else. No one came close to its ability to guess what a picture represents.
The tech world lost its breath and drew two conclusions: Partly artificial neural networks were the way to artificial intelligence. Partly because it was chips like those from Nvidia that would be used, not processors of the type that are in ordinary computers.
14 years later, Nvidia's founder Jensen Huang stands outside Münchenbryggeriet on Södermalm in Stockholm. As always dressed in a leather jacket and exclusive sneakers, now next to Volvo Cars CEO Håkan Samuelsson and a well-polished Volvo EX60. Jensen Huang talks about exactly the same thing as Alexnet: Image recognition using Nvidia chip. The difference is that now he is not talking about experiments to recognize cat photos, but about how self-driving cars can use it to understand their surroundings.
Another difference is that since the success of Alexnet, Nvidia has become the world's highest valued stock exchange company, ahead of Apple, Google's owner company Alphabet and Microsoft.
Worth over five thousand billion dollars.
And Nvidia has done it by betting everything on exactly what Alex Krizhevsky tested: Computer chip, with the same technological roots as the video cards of computer gamers, for AI calculations.
It is said that the people who made the most money in the Klondike gold rush were not those who looked for gold, but those who sold the shovels. Nvidia is selling the spades to the AI gold rush.
- All companies will become AI companies, says Jensen Huang.
Earlier in the day, he celebrated with Marcus Wallenberg that the Wallenberg sphere is building a Swedish AI supercomputer with an Nvidia chip for the computing needs of Saab, Astra-Zeneca and Ericsson. Then he spoke in particular about how he sees the future of pharmaceutical companies such as Astra-Zeneca.
- All these companies will turn into companies where robots discover new drugs.
The demand is huge when AI companies open up their giant data centers around the world and want to fill them with Nvidia machines. At times it has not been a question of whether anyone wants to buy the products, but whether Nvidia can deliver quickly enough.
This means that Jensen Huang and Nvidia have extreme power: Those who get to buy their chip can join the AI journey. Those left without risk falling behind. The question becomes bigger than competition between companies – for the US it is a hot potato which AI chips China gets to buy. The goal is that the most powerful will not be allowed to reach Beijing. That's why the world's stock markets and the AI world were horrified when Chinese Deepseek appeared last year with impressive AI performance, even though it had been trained with a simpler chip.
In other cases, access literally becomes a matter of life and death.
On July 6 this year, a Russian drone flew in over the Ukrainian city of Zaporizhzhya. It appeared set to attack a gas station, possibly a gas tank located there, the New York Times reported. When it slammed into a wall and exploded, three people were killed.
The drone was controlled by an Nvidia chip, which Ukraine said likely meant it operated autonomously, could navigate based on what it saw on the ground and make decisions about what to attack. A literal killer robot.
In Nvidia's defense: The computer on the drone was a consumer variant that can be bought just about anywhere. They are not exported to Russia, but should not be difficult to obtain via detours.
How does he see what happened? First he gives up.
- We sell these computer systems for general purposes. They are used for everything from shopping to logistics systems and self-driving cars. And sometimes drones. It is impossible for us to know how they are distributed and used, he replies.
Then he sharpens the tone, leans more towards the arms race being the way to go.
- The technology will be used in autonomous systems such as drones. We need to make sure that the good side has all the autonomous systems and all the drone technology that the others have. I don't know all the details of the case you mention but I am in favor of Nvidia's technology being used for autonomous systems for human safety and to create a safer world.
That is another of Jensen Huang's characteristics: He flatly refuses any hint of a slowdown in AI development and dismisses warnings about the dangers of AI development. This at the same time as many others have begun to advocate a slowdown and warn of the consequences of an artificial superintelligence: A series of defectors from the major AI companies. Actually the leaders of some of them, although one can sense a PR strategy in the background.
And Alex Krizhevsky's supervisor at the University of Toronto, Professor Geoffrey Hinton, who has warned of artificial intelligence far exceeding human. He was one of those who laid the foundation for artificial neural networks in the 1980s and thus for much of today's AI development, earning him the 2024 Nobel Prize in Physics.
Of course, Jensen Huang's historically unique business success relies on the AI world hitting the gas and continuing to buy his chip—perhaps unsurprisingly, he opposes a brake—but the heads of Open AI and Anthropic also have everything to gain from rapid AI development. Still, they open up to more regulation.
You are known to dismiss doomsday scenarios and seem far more optimistic than others in the industry. Why did you come to such a different conclusion?
- I am a responsible optimist, says Jensen Huang.
- If we are not optimistic about the future, then we should shut down completely. We continue because we are optimistic, but we must do so responsibly. It's not that different from what we're talking about now, he says, pointing, first at the Volvo CEO, then at the electric car.
One can object that the car industry is highly regulated, with a lot of requirements for safety and environmental considerations. Jensen Huang prefers to rely on corporate self-imposed liability.
- As CEOs, we have to say: If the product we create is not safe, then don't deliver it. It is very simple.
