When Colette Kress was appointed CFO of Nvidia in 2013, only gamers were familiar with the company that supplied the motherboards for their computer games. Almost a decade later, Kress is EVP and CFO of the company, which made history last May by increasing its market share by $200 billion in just one day. This happened after the artificial intelligence (AI) chip giant updated its positive guidance. This development also made the company a new member of the trillion-dollar club - double its market cap at the start of 2023.
Kress, 56, owes a lot to her sons. A decade ago, they were only eight and ten years old, but even then, they "advised" her to join the company that was the Holy Grail for gamers. Today, she is one of the most influential women in Silicon Valley, and the most senior woman in the sector underlying the growing AI trend. Her estimated net worth of about $250 million in Nvidia shares also makes her one of the world’s most highly paid CFOs. "This is my fourth company in a 35-year career. I still can’t believe it’s been 10 years at Nvidia." she tells Globes, in her first interview with the Israeli media. It is also her first interview granted to the global media since reaching the $1 trillion capitalization mark.
Overtaking Meta and leaving Intel behind
Prior to Nvidia, Kress served as SVP and CFO at Cisco’s Business Technology and Operations Finance organization . Then, Nvidia 's charismatic CEO and founder, Jensen Huang, went out of his way to bring her to the company. He sketched his vision out on a whiteboard: a mega-corporation supplying both the fuel and the engine to the giant companies leading the AI revolution. To a great extent, this vision has been fulfilled over the last year, and today Nvidia serves the generative AI needs for all of the giants: OpenAI, Microsoft, Google, Amazon, and Oracle.
Even a decade ago, Huang forecast that graphics processing units (GPUs) - usually used to process a small number of tasks such as calculating the movement of a 3D character in a video game - could be used in high-performance servers for more complex tasks, such as training computers to converse like humans, or create images in any requested artistic style.
The company’s first eye-opening moment, she says, happened even before she joined, when in 2012 the University of Toronto, (where the majority of today's most influential AI scientists got their start), first used GPUs in training neural networks in the classification of dogs . The second moment, of course, was the emergence of ChatGPT - OpenAI's chatbot.
Kress tries to downplay the company’s’ entry into the trillion-dollar club. "One of the things we don’t focus on is the stock price. We focus on our investors and how we would like to innovate. On the day it happened, we said ‘Okay, it’s here. But let’s get back to work’. So, in all honesty, I think more investors really now understand who we are - because 10 years ago, probably few could articulate our long-term vision.
I call it the 'supermarket test'. Years ago, if you walked into the grocery store wearing a green Nvidia shirt, some people wouldn't understand what that was. Today, I believe that people know our company, and are learning what graphic processors can do for artificial intelligence."
Kress' restrained, measured approach comes in stark contrast to the upheavals experienced by Nvidia investors, shareholders, and employees. From last October until now, the stock has jumped more than 300%, and since the beginning of the year, the stock has been almost exclusively on an upward trend. It trades at a $1.16 trillion market cap - the only chip company apart from Apple to be in the trillion-dollar club, not far from Amazon with a $1.39 trillion market cap, or Google which trades at a $1.55 trillion market cap. It has already surpassed Meta (at $809 billion), and left rival Intel far behind (at just $143 billion).
A monopoly dominating the world's fastest-growing market
Nvidia's financial performance at the end of May was instructive for both the semiconductor industry and the capital market in the last quarter: not only did revenue exceed all forecasts, but the company finished the first quarter at $ 7.2 billion, $700 million above the forecast, and earnings-per-share (EPS) increased by 26% year-on-year. Kress also updated second quarter guidance upwards by no less than 53% over the previous forecast.
Kress and Huang told investors they believe that next month’s second quarter financials will report peak revenue of $11 billion, instead of $7.5 billion as initially estimated. Hours after the presenting this data, Nvidia added about $200 billion to its market cap, proving to all that the demand for GPUs - even during an economic crisis - is the most important economic phenomenon of the year.
The main reason for the upward correction is, of course, the global hunger for AI, and technology giants are rushing to purchase supercomputers, graphics processors and graphics cards, which are Nvidia 's bread and butter. "This significant growth is mainly driven by data centers," Kress says explaining the steep update. "And it represents a sharp increase in demand for large language models in generative AI."
The second reason, no less important, for Nvidia 's performance is that it is almost the only player in the field, making it a monopoly that dominates the fastest growing market in the world. On the whole, what Intel did for semiconductors for personal computers for about 20 years at the end of the previous millennium, Nvidia is doing today. Its development language, CUDA (Compute Unified Device Architecture), which allows AI software developers to work on its chips, has become a market leader. Competitors like Intel and AMD have also launched their own artificial intelligence processors, but their success is far more limited
GPUs are the chips driving the generative AI revolution. Nvidia has an 84% share of the gaming market, and an even higher share of the AI segment - an estimated 90% or more. Accordingly, Nvidia's prices are high: while Intel’s latest supercomputer core processor is priced at about $17,000, the second-hand price for a similar Nvidia processor might even hit $40,000. Ultimately, however, Nvidia’s strategy favors its customers: the more cloud providers optimize data center activity, and, as a result, reduce their use of CPUs - like those made by Intel and AMD - the more they will choose GPUs, by Nvidia.
One of the reasons behind the higher costs is the fact that Nvidia is the dominant player in her sector, some may say a monopoly. Is being a monopoly a burden or an advantage?
"When you think about our products and about our gross margins, I think you have to understand that we don’t necessarily have an equal partner from a competition standpoint. When we say we are a full-stack company, a very big part of what we are selling is software. We are not selling a piece of hardware.
"We have written the system software, the development software, and more than 300 SDKs in order to help use that hardware. So that value and margin, incorporates software that acts as a major driver, it's not just about the cost of hardware. We haven’t found a competitor out there that has established the same development platform. So, there’s never been an apples-to-apples comparison."
"My day starts with Israel meetings"
Intel is a rival trying its luck in launching AI chips and is constructing a fab plant to serve companies like ARM and Qualcomm. How attractive would it be to manufacture at a company that is also your competitor?
"Would be happy to add another foundry partner. We currently are a multi-founder customer, working with Samsung and TSMC. If Intel wants to establish their business as a service provider - we’d be happy to be a part of that. And they know that’s a different part of their company. We have no problem being a partner with them in their foundry as they continue building their service opportunity."
Interaction with Intel was also played out in the difficult battle the two companies waged over the acquisition of Israeli tech company Mellanox. In the end, Nvidia made the acquisition, and can now sell high-performance computing solutions for data centers and servers that enable optimized communication performance between interconnected processors - a very important factor in supercomputer and/or AI server operations. The field of inter-server communication, also known as networking, also plays a central role in the market potential, in Nvidia’s view. The potential revenue identified for Mellanox by Nvidia may reach approximately $60 billion, out of a future total revenue opportunity of $1 trillion for the company's products.
"Being able to continue our vision of working on full data center computing solutions, not just with GPUs and networking as components, but with the ability to say we can build ourselves almost the entire data center for our customers - all of that is because of the combination of NVIDIA with Mellanox. The acquisition of Mellanox was one of the best acquisitions I saw in the industry."
Following this acquisition, 12% of Nvidia's employees - roughly 3,200 - are in Israel, distributed between centers in Yokneam, Tel Aviv, Ra'anana, Tel Hai, Kiryat Gat and Beersheva. "My day starts with Israel meetings, and only then continues to California."
"Most applications today do not leverage AI"
The sharp jump in AI stocks and the meteoric rise in the value of AI-related startups might indicate an AI bubble. However, Kress sees things differently.
"When we think about our future and what is important for Nvidia, there are two things: accelerated computing and AI. They are different but sometimes looked at together. Focusing on using accelerated computing to process the massive amounts of data and the efficiency that comes from accelerated computing is an essential part of our future and it will continue.
It is no bubble. Many companies will need to focus on sustainability and will be needing to move to accelerated computing, because they need to get energy efficient at what they are doing. If they continue their path of just buying standard CPU servers, they are spending money and getting no significant performance improvement. That doesn’t seem to be the right direction to go, and accelerated computing will be very important in that future."
According to Kress, "an outcome of accelerated computing is AI. It will probably be incorporated in all kinds of computing. Using AI can speed up computing, it can improve performance, productivity, efficiency, and it can lower costs. You can already visualize future AI applications for decades to come. Using recommendation systems, large language models, the generative AI applications after LLMs.
As for the contention that every company might buy AI chips and saturate the market - which could affect the hype overall, Kress offers an alternative line of thought: "That is one way to look at the situation, but let me try a different perspective. With every technology revolution, there is always a little bit of ‘side-hype’ that gets weeded out fast. I think we’ve heard the word ‘AI’ for several years right now and some may have started shrugging at the word - but there is a lot of real work that is being able to do, thanks to AI."
"Now, is this a period of time that it's only over a period of time of enthusiasm and soon everyone will have it? No, because technology progress, new types of workloads, application solutions, are still going to be a lot of work to complete over the next decades. Most applications today do not leverage AI. They will be re-routed to use accelerated computing, and will be re-written to build on AI."
Will the Biden administration tighten China restrictions?
There is one major albatross around Nvidia's neck to hinder its boundless growth trend: in recent months, the Biden administration has been intensifying its fight against China's escalating power in AI and semiconductors. This chip arms race could lead China to develop its own guided missile weapons, and code-breaking systems that could hack encrypted classified intelligence.
The direct casualties of this conflict are US chip companies such as Intel, which needs Chinese regulatory approval to complete its acquisition of Israel’s Tower Semiconductor, and Nvidia, which, although it launched a dedicated chip with limited capabilities for the Chinese market, may be forced to export them under a special license that will limit its sales, according to the "Wall Street Journal."
Reacting to the reports for the first time, Kress maintains that, "There are rumors that the US government is working to continue their work on export controls. We don’t know what that will entail - but we are aware of the rumors. We believe helping the government to understand the impact of export controls to the US industry or the world’s industry has been one of our important things to do. It’s not just about us, it will impact the industry as a whole."
Kress also adds that the Chinese market "is a very important region, and not having an ability to serve them will impact us and the industry. In the short term, we don’t believe it will, but in the long term, we do believe it will. If you are unable to sell products to them, it will have an impact."
Published by Globes, Israel business news - en.globes.co.il - on July 23, 2023.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2023.