AI: demand makes Nvidia take 210 days to deliver chips – 02/22/2024 – Market

Companies are racing to develop more powerful AI (artificial intelligence) models than current ones and, therefore, have been anticipating ordering cutting-edge chips from Nvidia. The wait reaches 210 days (or seven months).

The scenario was reported by representatives of companies involved with AI systems and confirmed by the director of the technology giant for Latin America, Márcio Aguiar, in an interview with Sheet.

“The results released yesterday are still small, the AI ​​market involves more than US$ 1 trillion (R$ 4.9 trillion)”, says Aguiar.

The high demand for so-called GPUs (graphics processing units) has caused Nvidia’s market value to more than quadruple since the launch of ChatGPT in November 2022.

This coveted piece sold for more than US$30,000 (around R$147,000) at retail, as it is essential for training generative AI platforms in a viable time. Throughout 2023, there was a shortage in the supply of the component.

This Wednesday (21), Nvidia released its balance sheet and recorded a 126% increase in revenue in 2023, compared to 2022. The numbers made the company once again surpass Google and Amazon in market value and brought the company closer to the valuation of US$2 trillion (R$9.9 trillion).

Brazil represented 5% of the chip developer’s global revenue in the last quarter, with US$1.1 billion (R$5.4 billion) spent with the company. This portion was limited to 3.8% in 2022.

According to Aguiar, the company invests to increase production and create GPUs. Dedicated US$2.9 billion (R$14.3 billion) to chipmaker TSMC’s new plant, which takes Nvidia’s projects off the ground.

Furthermore, in 2021, it developed software that accelerates the chip lithography process — a process that shapes the electronic circuits of GPUs.

Still, Nvidia chief executive Jensen Huang said in an announcement to investors that the expectation that Nvidia will meet all market demand is irrational.

In the balance sheet, the company projected revenue for the first quarter of 2024 of three times that recorded in the same period of 2023.

Nvidia’s datacenter sector, which has GPUs as its main element, represented 83% of the money that entered the cash register in the last quarter.

The chipmaker also provides technology for the automotive sector, robotics and the so-called industrial omniverse — in which industries use virtual reality spaces to carry out simulations.

According to Aguiar, Nvidia should announce more advanced GPUs at its main event of the year, scheduled for March 18th to 21st.

When enthusiasm for AI exploded early last year, H100 GPUs led to a hunt among entrepreneurs engaged in the technology. In November, Nvidia announced the H200, with the same architecture, but with more power.

Nvidia currently controls around 80% of the cutting-edge AI chip market, according to Reuters.

“When we started, 31 years ago, there were 19 chip producers, this number has been reducing because our customers trust us”, says Aguiar.

Nvidia’s leadership challenges the competition, since developing a line of GPUs requires spending around US$2 trillion to US$3 trillion (R$9.9 trillion to R$14.8 trillion), according to the director of Nvidia. “When they finish this project in three years, we will be three generations ahead.”

Advances in data processing capacity are essential for creating even more complex artificial intelligence models. The training process involves massive amounts of data, impossible to manipulate just a few years ago due to technological limits.

On February 8, the Wall Street Journal reported that Sam Altman, the chief executive behind ChatGPT, was moving to raise up to US$7 trillion (R$34.6 trillion) to overhaul chip production around the world. This deal, however, is still limited to a rumor.

For now, Nvidia is at the top of the segment and, at the beginning of the month, reported participation in other companies in the sector such as Super Micro, Arm —protagonist of the biggest initial public offering last year—, Advanced Micro Devices and Marvell Technology. Shares of all rose in value this Wednesday afternoon, after the release of Nvidia’s results.

Another important move for Nvidia was the purchase in 2019 for US$6.9 billion (R$34.1 billion) of the computer networking company Mellanox, which gave the technology giant an advantage in the area of ​​data center projects.

Thus, the company became crucial in setting up networks of GPUs sought after by the largest cloud providers in the world: AWS, Microsoft and Google.

“We had a very large investment from cloud providers, especially after Microsoft announced its Copilot [assistente baseado em IA]”, says the director of Nvidia. Worldwide, there are more than 50 thousand corporations that use this technology. “Copilot runs 100% on Nvidia GPU architectures.”

Aguiar states that demand in Latin America has also grown, with companies interested in setting up their own AI services, after frustration with the hallucinations of generalist models such as ChatGPT and Gemini, from Google. For this, Nvidia has its own AI toolbox, called NeMo.