ChatGPT "airborne", how did the era of Nvidia come?

Original source: Semiconductor Industry Perspective

Author: Takashi Yunoue

Image source: Generated by Unbounded AI‌

Sales ranking of semiconductor manufacturers in Q2 2023

On August 14, 2023, the American research company Semiconductor Intelligence announced the ranking of the top 15 sales of semiconductor manufacturers in the second quarter (Q2) of 2023.

However, I (Takashi Yunoue) am disappointed with this ranking. The reason is that the ranking of TSMC in Taiwan, China is unknown, so I took a look at TSMC's financial report data and added it to the semiconductor intelligence rankings (Figure 1).

Figure 1 Top 15 semiconductor manufacturers by sales (Q1 and Q2 2023) Source: Author based on Semiconductor Intelligence and TSMC financial report data

Judging from the results, the first place is TSMC, the second place is Intel, the third place is Samsung, the fourth place is Nvidia, the fifth place is Broadcom, and the sixth place is Qualcomm (the fourth and fifth places are predicted of).

Among Japanese manufacturers, Renesas Electronics, which has strong automotive semiconductor business, ranks 15th, but Kioxia, which produces NAND, seems to have fallen out of the top 15.

Notably, Nvidia grew 69.2% from the first quarter of 2023 to the second quarter. NVIDIA uses a fiscal year (FY) rather than a calendar year (CY) for its accounting purposes (Figure 2). Therefore, the time of releasing financial reports is one month different from the time when Intel and Samsung released financial reports in CY. As a result, Nvidia's sales are written as "forecast" in the Semiconductor Intelligence rankings.

Figure 2 The difference between NVIDIA’s FY (fiscal year) and CY (calendar year)

So where is Nvidia's real ranking? Is it really fourth place?

In this article, we first examine Nvidia's quarterly sales figures. Next, I'll try to compare with the sales of the top three semiconductor manufacturers TSMC, Intel, Samsung, and Nvidia. In addition, the explosive spread of generative AI, such as the open AI "ChatGPT" released on November 30, 2022, is the reason behind NVIDIA's rapid sales growth. In addition, I will also talk about the prospect that Nvidia may play a leading role in the sales ranking of semiconductor manufacturers in the future.

Comparison between the three major semiconductor manufacturers and NVIDIA

Figure 3 shows Nvidia's quarterly sales and operating profit trends. As mentioned in the previous section, Nvidia uses its fiscal year (FY) to report its performance. Therefore, the first quarter of 2024 is February-April 2023, the second quarter of 2024 is May-July 2023, and the third quarter of 2024 predicted by Nvidia is August-October 2023.

Figure 3 NVIDIA quarterly sales and operating income (NVIDIA forecasts Q3 in 2024) Source: NVIDIA financial report

Looking at Figure 3 again, quarterly sales have increased from US$7.2 billion in Q1 2024 (February 2023 to April 2023) to US$13.5 billion in Q2 2024 (May to July 2023). You can see the scale doubled.

How should we compare this FY Nvidia with TSMC, Intel, and Samsung that settle in CY? Because it's difficult to make time-consistent comparisons, for CY such as TSMC, sales are plotted in March, June, September and December, which are the last months of quarterly financial results. I created a chart plotting sales for April, July, October, and January (Figure 4).

Figure 4 Quarterly sales of Intel, Samsung, TSMC, and Nvidia (drawn at the end of each company’s accounting period) Source: Prepared by the author based on the financial statements of each company

Intel held the No. 1 spot until March 2017, but Samsung, which grew rapidly due to the memory bubble, jumped to No. 1 between June 2017 and September 2018. However, then came the memory recession and Intel returned to the top position after December 2018.

For a while after that, Intel was number one and Samsung was number two. However, when the new coronavirus special demand began to collapse in 2022, Intel's sales fell rapidly after December 2021, Samsung after June 2022, and TSMC, which had grown steadily since 2019, surpassed that in September 2022 The company jumped to number one. TSMC’s sales have also fallen sharply since December 2022.

In this case, Nvidia's sales will increase significantly from April 2023 to July of the same year. Although the settlement time is one month apart and cannot be directly compared, from Figure 4, around June to July 2023, TSMC will be first, NVIDIA second, Intel third, and Samsung fourth.

In addition, if TSMC's sales continue to decline after September 2023, it is not a fantasy for Nvidia, which is expected to have sales of US$16 billion in October of the same year, to jump to the top of the list. If this happens, Nvidia, founded in 1993, will become the No. 1 semiconductor company by revenue for the first time in history.

But what is the driving force for Nvidia’s rapid progress?

The explosive spread of ChatGPT

ChatGPT was released by Open AI on November 30, 2022, and quickly became popular around the world. The time it takes to reach 100 million active users is 54 months for Facebook, 49 months for X (old Twitter), 30 months for Instagram, 19 months for LINE, 9 months for TikTok, and ChatGPT says there are only two moon.

Its results have also improved rapidly. In January 2023, the answers to the MBA final exam were rated B (passing level). In February of the same year, the correct answer rate for the American Medical Qualification Examination reached the passing line. In March of the same year, it was reported that GPT-4 was in May. In September, he scored in the top 10% of the U.S. Bar Qualification Examination, reaching the level of passing the Japanese National Medical Examination in the past five years.

The generative AI craze that started with ChatGPT is boundless. Since then, high-tech companies have begun to develop generative AI. This generation of AI uses a semiconductor called a Graphics Processing Unit (GPU). Nvidia has a monopoly in the GPU field.

Here, what is generative artificial intelligence and how does it work? What role does the GPU play?

ChatGPT has two steps

Figure 5 We will explain the mechanism of generating AI such as ChatGPT. The steps of generative artificial intelligence are divided into two phases: learning and reasoning.

Figure 5 Generative AI principle and AI semiconductor used (NVIDIA GPU)

First, load big data such as text data and image data on the Internet to a server equipped with AI semiconductors such as NVIDIA GPUs (hereinafter referred to as AI server). At that time, artificial intelligence semiconductors such as GPUs will learn from the data.

Then, when a user writes a question in the chat, the generative AI running on the AI server makes inferences and comes up with an answer. At that time, it was Nvidia's GPU and other AI semiconductors that performed inference on the AI server.

From this we can see that generative AI can be said to be "similar software" that runs on AI semiconductors (such as NVIDIA GPUs) installed on AI servers.

The spread and expansion of generative AI like ChatGPT is limitless. As a result, Nvidia's GPU shortage problem is becoming increasingly serious in the semiconductor market. In this case, high-tech companies developing generative artificial intelligence are racing to collect as many Nvidia GPUs as possible.

Nvidia's GPUs come in many types, but the most sought-after are TSMC's 7nm A100 ($10,000 each) and H100 ($40,000 each). No matter how expensive it is, considering a single DRAM is $10, Apple's iPhone processor is $100, and Intel's PC processor is $200, I've never seen a GPU cost between $10,000 and $40,000, it doesn't exist Ridiculously expensive for tips.

Rapidly expanding data centers demand AI server GPUs

Figure 6 shows NVIDIA's quarterly sales by business area. Originally, Nvidia's GPUs were semiconductors developed for gaming consoles. As you can see from Figure 6, gaming GPU sales are the largest until around FY2020 (actually 2019).

Figure 6 NVIDIA’s Quarterly Sales by Business Area

Under such circumstances, it was found that GPUs capable of processing large amounts of images in parallel are most suitable for AI semiconductors. In my memory, from about 2016 to 2018, Nvidia GPUs were often used in AI semiconductors for autonomous driving in cars.

However, from Figure 6, the sales volume of automotive GPUs is not that large. The reason is that fully autonomous driving of cars has not yet become popular, and autonomous driving pioneers led by Tesla in the United States have begun to develop their own AI semiconductors for fully autonomous vehicles.

From around fiscal year 2023 (actually 2022), Nvidia's internal GPU sales for data center AI servers will grow rapidly. GPU sales for AI servers will explode in FY2024 (actually 2023).

Driven by a surge in demand for GPUs for AI servers, Nvidia's revenue has (roughly) surpassed Intel and Samsung and approached TSMC. If this momentum continues, TSMC may overtake it.

Since 2010, Intel, Samsung, and TSMC, which have the largest sales of semiconductors, have been known as the "Big Three". However, Nvidia has suddenly entered the fray for the top ranks. In the future, Nvidia may become the No. 1 semiconductor sales company. Clearly, Nvidia's time has come.