Nvidia is one of the biggest winners of the AI boom, and now its founder and CEO Jensen Huang is presenting the next generation of its superchips. But competition is growing – and so are geopolitical tensions.
Nvidia is one of the leading companies in the field of AI.
Given Ruvic / Reuters
The AI boom in Silicon Valley has rarely been as tangible as it was this week in San Jose. The local ice hockey team and superstars like Taylor Swift usually play in the SAP Center in the city center. On Monday, 11,000 visitors came there to see the head of a semiconductor company.
“I hope you know that this isn’t a concert, we’re talking about science,” said Nvidia’s founder and CEO Jensen Huang as he took the stage. Dressed in all black and his iconic leather jacket, the 61-year-old actually looked a bit like a rock star.
Can Nvidia continue its amazing success story?
Nvidia is currently the star of Silicon Valley. The company’s last developer conference took place in 2019 on the premises of the local university with around half as many visitors. Back then, hardly anyone talked about high-performance chips, but today they are the engine for almost all innovations in Silicon Valley. Without them, companies can no longer train their AI models.
With a share of more than 70 percent, Nvidia is the most important player in this market: Meta, Google, Amazon, they are all tearing the latest chips of the so-called Generation Hopper out of Nvidia’s hands. Despite unit prices of around $25,000, these have been sold out for months.
The 30-year-old Nvidia is one of the biggest beneficiaries of the AI boom; In the current quarter, the company is expected to have sales of $24 billion, about three times as much as a year ago. With a market valuation of $2.24 trillion, Nvidia has become the third largest listed company in the world within just a few months, behind Microsoft and Apple.
Nvidia continues to boom
Market value in billions of dollars
“There is an AI war going on out there and Nvidia is the only arms dealer,” said Wall Street analyst Srini Pajjuri from the financial services provider Raymond James.
The big question that concerns investors and corporate customers in San Jose this week is: How does CEO Huang want to continue this enormous success story for his life’s work, Nvidia?
Nvidia’s sales continue to rise rapidly
Quarterly figures in billion dollars
Blackwell is the name of the company’s new superchip
With Blackwell, is the short answer. That’s the name of Nvidia’s latest superchip and the architecture behind it, named after mathematician David Blackwell, who was the first African American to be inducted into the US National Academy of Sciences.
What the new chip can do is so unprecedented and difficult to imagine that in his speech Huang repeatedly uses comparisons with the previous superchip of the Hopper generation: There should be twice as many transistors on a Blackwell chip, namely around 208 billion. The computing speed will be about five times as fast as Hopper: While the H-100 calculated at a speed of 4 petaflops, the new Blackwell GB 200 chip will have 20 petaflops. In other words, the chip can carry out 20 quadrillion – i.e. 20,000 trillion – calculations per second. “It is the largest chip that is physically possible,” said Huang.
This is good news for companies that are working on ever larger and more complex AI models. While AI models with 1.7 trillion parameters could be trained on the Hopper generation chips, the Blackwell system can now operate a model with 27 trillion parameters.
In addition, the models can now be trained more cheaply. Huang calculates this using an example: To train OpenAI’s GPT-4 model, it took 8,000 chips from the Hopper generation, which consumed 15 megawatts of energy over 90 days. Blackwell could train GPT-4 in the same time with just 2,000 chips and would only need to use 4 megawatts of energy. This means enormously reduced energy costs for companies – a point of criticism that has been mentioned repeatedly to date.
The customer base should become broader
The Blackwell chips and the associated architecture are scheduled to come onto the market at the end of the year. According to Nvidia, Amazon’s AWS cloud division alone will buy 20,000 of the new high-performance chips. Google, Microsoft and Oracle have also already announced that they will sell the new Blackwell chip via their cloud services. The selling price is said to be between $30,000 and $40,000, below what analysts had expected.
The market grows with every chip generation, Huang says in an interview with analysts, “and we want the entire market to be able to afford a Blackwell.”
The reason for this is that Nvidia wants to open the market for AI chips to a broader range of customers, as Huang explained to analysts on Tuesday. To date, Nvidia has a real cluster risk here: In the last fiscal year, a single corporate customer accounted for 19 percent of Nvidia’s sales, as the company announced in February. Nvidia did not disclose who this customer was.
But it is known that Microsoft, Meta, Alphabet and Amazon make up a large part of Nvidia’s sales – a whopping $9 billion in the last quarter. At the same time, all four companies are working on their own high-performance chips in order to make themselves less dependent on third-party companies like Nvidia. It is therefore in Nvidia’s own interest to reduce its own dependence on the four major customers.
Robots as the next big AI revolution
In his two-hour, largely technical speech, Huang made it clear that he sees Nvidia at the forefront of the next industrial revolution. Weather models, drug research, healthcare, robotics, self-driving cars – every industry will be revolutionized by the new wave of AI, Huang said with certainty.
He also sees the next computer revolution on the horizon, when AI teaches itself the laws of physics using video recordings. Then the big breakthrough in robotics will come, Huang is convinced, and brings up images of humanoid robots on stage to emphasize his statement: “The Chat GPT moment for robotics may be just around the corner.”
Jensen Hung with a mini robot on stage in San Jose.
Eric Risberg / AP
Nvidia sees itself at the controls of this revolution: instead of just producing chips, the company is continuing to expand its ecosystem around them. In particular, in the future they want to offer a software interface with which every company can build its own AI system.
The competition is growing – and the geopolitical tensions
The markets seemed to have already priced in the euphoria surrounding Nvidia’s new super chip. The shares were quoted unchanged on Monday and up one percent on Tuesday. Analysts at JP Morgan wrote that with its new products and strong ecosystem, Nvidia is further distancing itself from its competitors. “Nvidia is still one or two steps ahead of the competition.”
Huang presents his descriptions of the AI-driven future so confidently that one could forget that Nvidia does have competition – for example from the Taiwanese chip company AMD, from Intel or, more recently, from startups like Groq and Cerebras. The fact that Nvidia has an enormous gross profit margin of more than 70 percent attracts competitors like moths to light. At the press conference, Huang waves off questions about new challengers like Grow. “I don’t know much about her,” he claims simply.
There are also geopolitical threats to Nvidia’s business model. American high-performance chips are subject to ever-widening export bans to China, a huge market for Nvidia. Huang also downplays that threat, saying they play by the rules and follow what Washington policy dictates.
But the tensions with China not only endanger Nvidia’s sales, but also production: The Taiwanese company TSMC produces the chips designed by Nvidia, but there is a threat of Chinese invasion in Taiwan, which would mean war with the USA – and probably destruction the fabs, i.e. the chip factories there.
“We are doing everything in our power to create a resilient supply chain,” says Huang, and is naively optimistic: He is confident that the goal of countries is not to fight each other. In order to spread the geopolitical risks, TSMC is currently building factories in the USA, among other places. But it will still be years before Nvidia’s chips can be produced there. But no one in San Jose wants to talk about these dangers.