Semiconductor giant Nvidia became the world’s most valuable company this week after making a smart move a few years ago to corner the AI chip market.
Long before it rose to the top of the corporate world, NVIDIA was a startup first conceived by CEO Jensen Huang in a Denny’s booth with co-founders Chris Malachowski and Curtis Prime. Shortly after founding in 1993, the company invented one of the first GPUs (graphics processing units) developed for use in video games and graphic design.
Since then, the company has upped its ambitions, living up to its name, which means “envy” in Latin, and dwarfing tech giants like Apple, Microsoft and Google with a market capitalization of $3.34 trillion. Its value has more than doubled since January. But how did it get there? Analysts say the key to its success is luckstarted decades ago as an early adopter of the AI boom that is now sweeping the market.
CPUs, the most common computer chips that emerged in the 1950s, are best suited to perform complex calculations one by one, but as research into deep learning and AI gained steam in the 2010s, they weren’t perfectly suited to the needs of data scientists. In contrast, Nvidia’s GPUs were ideal for AI because they could perform many simple calculations at once. In 2012, Ilya Sutskever, former chief scientist at OpenAI and co-founder of AI startup Safe Superintelligence, was already using Nvidia’s chips for an early convolutional neural network called AlexNet.
Nvidia’s chips have advanced rapidly in recent years, and its GH200 Grace Hopper superchip, released last August, can now perform 2 quintillion (that’s 200 followed by 18 zeros) calculations per second.
But Nvidia’s dominance of the AI market owes much to a timely bet made by CEO Jensen Huang several years ago, said Tristan Guerra, senior research analyst at Baird Semiconductor. luckOne of the company’s visionary moves was the creation of CUDA in 2007, a high-level programming tool that the company built to unleash the full power of GPUs in a simple way.
“NVIDIA co-founder and CEO Jensen was a visionary who early on identified the trend for GPU adoption in the data center and aligned the company’s strategy with that vision,” Gera said.
CUDA is so widely used now that it’s hard to imagine companies building large-scale language models like OpenAI’s ChatGPT using any other technology, added John Abbott, infrastructure analyst at 451 Research, a unit of S&P Global Market Intelligence.
“Training large models can take months, and reducing that time requires large clusters. We had no choice because the mature software tools and the skills needed to use them are readily available for Nvidia GPUs, which have become the de facto standard,” Abbott said in an email.
In addition to its first-mover advantage, Gera said the company also has a technological advantage.
“NVIDIA offers a complete supercomputer solution that includes the highest performance hardware (chips) and software suite. Our competitors only offer AI chips,” Gera said.
Still, Abbott warned that Nvidia faces several threats to its dominant position in AI chips. Nvidia controls about 90% of the AI chip market, but other big tech companies such as Meta and Google have started making their own chips to compete.
The company also faces geopolitical obstacles in China: The U.S. has restricted Nvidia’s expansion there, and the Chinese government is working hard to develop alternatives to Nvidia’s equipment. The threat of war could also upend the company’s business.
“Taiwan, where NVIDIA currently sources all its GPUs, is under political threat. Ongoing supply chain issues are also a major risk,” Abbott said.
But for now, Nvidia is doing well. Its shares have risen so much that it forced a 10-for-1 stock split earlier this month. The company’s gains account for a third of the total value added in the S&P 500 index since January.