Nvidia exec: ‘The market still sees us as a one-product type of company’

Fueled by continued demand for powerful AI processors, Nvidia delivered another blockbuster quarter in its fiscal Q2, ended in July, with sales more than doubling year-on-year to US$30bn and profits soaring 168% to US$16.6bn.

However, the company’s shares on US markets fell on the post-financial disclosure, as investors expected even stronger results from what has become the world’s most valuable publicly traded company.

Meanwhile, demand remained high across the board, including in Latin America, where firms other than big cloud and datacenter companies, such as banks and service providers, lined up to receive Nvidia's chipsets, as Márcio Aguiar, head of corporate sales in the region, tells BNamericas in this interview.

BNamericas: In any other company results or growth like this wouldn't necessarily lead to a drop in the share price. Has the market become spoiled by Nvidia or are investors looking beyond the AI frenzy?

Aguiar: We're used to all this movement, these expectations, which are always quite high. We have been delivering results, even surpassing them. Nvidia is a public company, we have a commitment to our investors, but, honestly, it's difficult to control market expectations.

We're working, as always, concerned with continuing to innovate, continuing to meet the computing demand that exists today, which was created by us.

We have a very strong commitment to more than 40,000 companies that are developing their products on top of our software and hardware platforms.

The market has become spoiled. The market is the market. Obviously, it can lift you up and it can bring you down, but our commitment here remains unchanged. And we're heading towards closing a record fiscal year.

I would say that the market, even though it understands that we're one of the protagonists of this new modern AI, still sees us as a one-product company. One that sells GPUs [graphics processing units].

BNamericas: The company has become synonym of genAI chipsets.

Aguiar: All the news that comes out is about GPUs, the market doesn't talk about the platforms that Nvidia delivers to several other players.

And, again, it's a somewhat volatile week. The US government is under a lot of pressure to reduce interest rates – the announcement will be made tomorrow [Friday]. Will they reduce rates? If they do, stocks will rise. If interest rates remain the same, it will inhibit the growth of the US economy, and this could also mean consumers will have less money to invest or buy. 

It's the global economy. And it's also an election year in the United States. We can't be worried about that, and we're not.

BNamericas: What is the focus of attention?

Aguiar: Our concern is to have growing demand for the current generation [of GPUs], which is the Hopper, especially the H200 model.

We'll now start delivering the new generations, the Blackwell generation, which is where we'll see quite good growth. I can't say exponential, because we haven't given guidance on what coming quarters will be like.

[Editor's note: Blackwell is Nvidia's next-generation AI superchip, which combines several generations of Nvidia technologies to advance generative AI and accelerated computing. Taiwan Semiconductor Manufacturing (TSMC) is the manufacturer of Blackwell.]

BNamericas: The production of Blackwell chipsets was delayed, right?

Aguiar: There were some setbacks, yes, so to speak. But demand is already high and we're trying to keep all of our customers satisfied, because we can't please everyone. Everyone wants something, and they want a lot.

The Blackwell generation will be highly targeted at CSPs, which are cloud providers, who already placed their orders a long time ago, while Hopper continues to see growing demand among corporations.

BNamericas: What would you highlight in Latin America?

Aguiar: There has been a wave in Latin America where companies are using GPUs in the cloud a lot, so all the major cloud providers have invested heavily in more GPU offerings in the region.

This continues, and it has given us a position to reach customers with more authority and say, ‘You can run this locally’. Because we're beginning to enter an era where not only countries, but also companies, want to start running their workloads locally.

We have seen this investment by cloud service providers with this GPU offering, and smaller companies, but with many customers in the cloud, are also beginning to emerge, offering almost a ‘private cloud’ with GPUs for local customers.

BNamericas: Which companies?

Aguiar: Banks, other service providers, among others. It's no longer just the big cloud providers like Amazon and Oracle. Companies are starting to offer GPUs as a service because it's very difficult for customers to buy new servers.

BNamericas: I recently heard from an executive in the datacenter sector that the boom of generative AI, from the point of view of long-term, scalable investments in the data structure and ecosystem, hasn't yet reached Latin America. Do you agree?

Aguiar: We have to remember the following: generative AI became public at the end of 2022. 

And everyone now uses AI, uses some generative AI technology every day, but not necessarily yet applied. It is customized to the client's business because they have to have people, they have to have knowledge, they have to know what they can take advantage of from their work with this technology.

I hear the words 'boom', 'hype' a lot... If you stop to think about it, chatGPT is in version 4 in two years. 

The trend, and what we have also been working on, is to expand this ecosystem.

We've added many more software companies, such as ServiceNow, Salesforce, Hugging Face, Adobe, thousands of companies from different segments, investing and migrating their applications to bring more generative intelligence concepts to their applications.

It is a change, a revolution that we're going through and will go through for many years.

BNamericas: The point the executive was referring to wasn't necessarily the hype itself, but the investments in data infrastructure for generative AI. This spending, in Latin America, would still be very specific and concentrated among a few large companies. This executive even pointed at costs, not only the price of Nvidia, but costs of imports.

Aguiar: This is work that we're also doing on another front, at the government level. 

Even though the Brazilian government has announced an investment plan in artificial intelligence, the way these products are still being taxed... it's very focused on the end-consumer. We have to think about the industry. 

Datacenter companies are making the investments that are relevant to help this Brazilian economic growth. Now, we're coming with new GPU architecture, which is RUBI, and Blackwell itself, whose manufacturing already includes water cooling concepts.

No datacenter that is installed today or that is being installed now [in Latin America] is considering this type of cooling. So, this will have to be modernized too

BNamericas: How are demand and delivery times looking?

Aguiar: There are several datacenter companies in the region that say 'I want to get into this business', 'I want to buy the GPUs'. I say, 'okay, but if you really want to buy, you have to buy now, because you will receive it in nine months'.

There's also the matter of profile, of use. We're not interested in making a big sale to a partner and this partner remaining with that technology for five or six years, because there's no one to use it. In the meantime, I have other companies that already have a focus on AI. 

This hype drives lots of companies that aren't specifically focused on offering this technology but want to get in because AI is in fashion. You're going to lose money.

BNamericas: Ultimately, with high demand and long delivery times, non-specialist clients will receive a technology that could soon become outdated.

Aguiar: You have to start slowly. A datacenter company can start leasing other datacenters that already have this base installed to see what the clients' consumption is. In the meantime, you draw up the business plan.

Because the moment you stop to draw up a business plan and include GPUs in it, that business plan will only be executed in nine months. In nine months, Nvidia has already announced two new generations.

In 2018, when we came up with the first hardware architecture with eight GPUs, no manufacturer wanted that type of architecture. And we warned that there would be a huge computational demand.

We started licensing to companies, for example, the Chinese company Supermicro, which is a major partner of Nvidia. After two years, US companies said, 'Oh, I want it too', and we passed on the 'recipe' so that they could operate in a model as a service. The same thing with datacenters.

The first company to invest in this, in GPU as a service, was Amazon. That was back in 2014. Then several others came along, and now those that weren't focused on GPUs are starting to come and are now buying GPUs and selling services to these giants.

It’s a revolution. Even for us, who were focused on working with that one end-customer. Now we’re working with partners who will sell to all of these end-customers.

BNamericas: You mentioned Amazon. In addition to buying GPUs from Nvidia, the company started developing its own architectures. This is also due to supply and delivery issues. Other big tech companies are likely to follow suit. How do you see that competition?

Aguiar: I would say it's more of a co-opetition. 

But everything we're doing today isn't something we started doing two years ago. We've been making GPUs since 1999. Other companies will have the same skills we had to develop this type of hardware, but this doesn't happen overnight because it's not just the hardware, it's the ecosystem that needs to be developed.