Nvidia: search for AI chips surprised, says director – 9/1/2023 – Tech

Trillionaire Nvidia’s graphics processing units (GPU) are the main engine of artificial intelligence. The demand is such that smaller startups struggle to assemble the necessary structure to train AIs, and the hardware manufacturer has been choosing who to sell to.

This situation has elevated the computer graphics giant to the position of the company that has gained the most from the race for generative AI. Revenue of US$ 13.5 billion (R$ 65.7 billion) in the second quarter of 2023 was twice the amount registered in the same period of the previous year.

If, on the one hand, the strategic position seems favorable, the company’s plans are to expand the production of its graphic processing units, according to the director of the business area for Latin America, Marcio Aguiar.

“We were taken by surprise by this ChatGPT boom”, he says.

The technology company invested US$ 2 billion (R$ 9.74 billion), out of a total of US$ 2.9 billion (R$ 14.2 billion), in the construction of a new plant focused on AI chips for the TSMC, Taiwanese chip factory that takes Nvidia’s projects off the ground.

“We have already committed to buying surplus production, but TSMC is expanding the factory and it is not something done overnight”, says Aguiar. He says the company’s focus is on generating technology, and revenue numbers are just part of the business.

A great example of this high demand was the investment by Nvidia and Microsoft of US$ 1.3 billion in the startup of former Google engineers Inflection AI. Of this amount, 95% was reinvested in the purchase of Nvidia’s own H100 GPUs for the expansion of a datacenter specialized in training generative AI. Each chip sells for more than US$30,000 (about R$147,000) at retail.

In addition to chip design, Nvidia develops computing platforms aimed at graphics processing that optimizes the performance of GPUs. The technology giant’s team of engineers and researchers is global and includes Brazilians, according to the executive.

The high-performance, low-power graphics cards from Nvidia attract not only startups, but also large cloud service providers such as Amazon Web Services, Microsoft, Oracle and IBM. The latter buy chips to set up data centers.

Amazon and Microsoft develop their own chips for specific activities, but they still depend on Nvidia items to build an efficient infrastructure.

But not all of Nvidia’s business is based on the H100 GPUs used to run language models like ChatGPT.

In Brazil, the main customer, according to Aguiar, is a large state-owned oil company — without mentioning the name of Petrobras. The company runs fluid analysis and oil well mapping programs and does not need H100 GPUs for this.

Another important niche for the company is the gamer audience. Nvidia started developing specialized game processing cards in 1993.

The company’s current CEO and co-founder, Jensen Huang, says he chose video games because of the constant challenge imposed in the area of ​​computer graphics, demanding in research and implementation. According to Huang’s public speeches, this difficulty made the company move forward.

In 2007, Huang led Nvidia’s expansion beyond the gaming industry, which enabled it to dominate the AI-focused hardware market today. Nvidia introduced graphics processing units (GPU), which had functions other than simulating complex video game graphics.

At the time, this movement weighed on the accounts, but allowed the company to occupy the current position, in which it flirts with a market value of around US$ 1 trillion (about R$ 5 trillion).

Being at the forefront of technology is always a concern for Nvidia, according to Aguiar. Therefore, although it does not have a government relations arm like other technology giants, the chip developer develops partnerships with universities.

In one such case, it offered GPUs and training in accelerated computing to the Artificial Intelligence Center of Excellence (CEIA). This contact, says Aguiar, is important for training, getting to know and attracting talent around the world.

In March, Nvidia became the first company in the semiconductor industry to reach a $1 trillion valuation. It is a branch where the starting ingredient is sand. All other value is added by people, former Intel chairman Andy Bryant once said.

---

NVIDIA X-RAY

Foundation: April 1993
Net profit*: US$ 6.2 billion (R$ 30.2 billion)
Market value: US$ 1.208 trillion (R$ 5.883 trillion)
Employees: 26,196
Occupation area: chips, semiconductors and software
competitors: AMD, Intel, Qualcomm and Samsung Electronics

*In the second quarter of 2023