Interest rate cut expectations fell through, Nvidia (NVDA.US) stopped rising seven times in a row, and “core business” changed again?

As the consumer price index (CPI) heated up beyond expectations in December 2023, the Federal Reserve's path to cutting interest rates changed again. After the US stock pre-market CPI was announced, the price of US Treasury bonds plummeted, and Nvidia (NVDA.US)'s multi-day increase, which reached a record high since the beginning of the year, also dropped slightly by 0.2% on the 12th.

Looking back at the US stock trend in 2023, Nvidia is undoubtedly the “star stock” that received the most attention and truly deserved it. Driven by an unprecedented wave of generative AI, Nvidia surged nearly 240% throughout the year, the biggest annual increase since 2001.

Entering 2024, Nvidia's stock price continues to rise. The stock price soared by about 10% within the first 9 trading days of the year, and the market capitalization surged by 128 billion US dollars, reaching a record high for the first nine days of the year. Institutions such as Baird and UBS have all given optimistic expectations. Among them, Bank of America gave a target price of 700 US dollars in a recent research report, which means that Nvidia still has the potential to rise by more than 40% compared to the current stock price.

Earnings season is approaching, and in addition to the difficult timing of interest rate cuts, the market will also face disturbances caused by disclosure of results. For Nvidia, the continued popularity of the AI chip “carnival” prompted the company to expect to continue to record more revenue than expected in Q4, but the stricter chip ban also cast a shadow over the company's future.

The annual revenue growth rate is expected to reach 118% due to strong supply and demand for AI chips

In 2023, Nvidia's performance trend can be described as promising. According to the financial report for the third fiscal quarter of fiscal year 2024, as of October 29, 2023, Nvidia achieved revenue of US$18.12 billion, up 206% year on year; operating profit of US$10.417 billion, up 1633% year on year, up 53% month on month; net profit of US$9.243 billion, up 1274% year on year and 49% month on month.

Looking at the revenue structure, the data center business has surpassed gaming and has become the company's new revenue pillar. The company's data center business revenue in the third quarter hit a record of 14.51 billion US dollars, an increase of 41% over the previous quarter and an increase of 279% over the same period last year, mainly due to the contribution of cloud infrastructure providers;

The gaming business benefited from the continued popularity of GeForce RTX 30 series video cards. Q3 revenue was US$2.86 billion, up 15% from the previous quarter and 81% higher than the same period last year;

The revenue of the professional vision business in the third quarter was US$416 million, up 10% from the previous quarter and up 108% from the same period last year; the third quarter revenue of the automotive business was US$261 million, up 3% from the previous quarter and 4% from the same period last year.

The main driving force behind the rapid rise in the company's performance comes from the intense demand for generative AI. According to estimates by the US market research agency TrendForce, the GPT-3.5 model, which processes 180 billion parameters, requires 20,000 GPU chips, while the ChatGPT launched by OpenAI uses an AI computing cluster of more than 10,000 Nvidia A100 GPU chips.

According to the Zhitong Finance App, Nvidia unquestionably occupies a leading position in the high-performance GPU market. According to Jon Peddie Research's estimates, Nvidia's GPU shipments will account for 87% of the total market this quarter, followed by AMD (AMD.US) and Intel (INTC.US), with market shares of 10% and 3%, respectively.

The H100 GPU chip is the most popular product this year. The H100 is Nvidia's 9th generation data center GPU. Using the NVIDIA Hopper GPU architecture, the accelerated computing performance of the NVIDIA data center platform has taken another major leap forward.

According to statistical analysis by market tracking company Omdia, Nvidia sold about 900 tons of GPUs in the second quarter of 2023, while about 500,000 H100 and A100 GPUs were sold in the third quarter. Among them, Meta and Microsoft were the biggest buyers, each purchasing as many as 150,000 H100 GPUs.

Under high demand, Nvidia naturally has no shortage of orders; the only thing worth worrying about is production capacity. As of the end of July this year, Nvidia had prepaid supply and capacity agreements of US$3.81 billion on its account, compared to US$3.67 billion at the end of October.

The production capacity of TSMC's CowOS advanced packages is an important factor limiting Nvidia's high-performance GPU shipments. By the end of 2023, the monthly production capacity of CoWoS was only 15,000 wafers, but TSMC revealed that it is expected to reach 17,000 wafers in the first quarter of 2024, and eventually reach 26,000-28,000 wafers per month in 2024.

Currently, Nvidia's balance sheet indicators are also performing well. According to S&P Global Market Financial Intelligence data, Nvidia's total free cash flow for the past four quarters was 17.5 billion US dollars, which is the highest in the chip industry along with Broadcom.

According to a report recently released by Bank of America Securities, Nvidia's leading position in the industry is expected to help it generate an incremental free cash flow of about 100 billion US dollars within this year and next two years, which will further maintain the company's sustainable growth.

Looking ahead, the company revealed that revenue for the fourth quarter is expected to be $20 billion, fluctuating 2% up and down, far exceeding the market's previous expectations. GAAP and non-GAAP gross margins are expected to be 74.5% and 75.5%, respectively. Based on Q4 revenue of 20 billion US dollars, the company's annual revenue will reach 58.819 billion US dollars, an increase of 118% over the previous year.

New product launch giants scramble to add AI chips

Looking at Nvidia's various businesses, in addition to generative AI, the wave of emerging technologies such as autonomous vehicles and the metaverse is also expected to create new favorable factors. However, in 2023, the global game industry continued to weaken, and the game business, which was once Nvidia's main revenue force, declined markedly. Compared with the booming growth in the data center business, the current scale can only be considered an “addition.”

In 2023, Nvidia is undoubtedly the most popular player in the GPU sector that has received the most attention. It has seized the high ground of market share, and its voice in pricing has brought high premiums. Strong supply and demand trends have driven Nvidia's performance to rise steadily like a rocket.

Despite this, Nvidia can't “sit back and relax” on Jinshan; the big cake of AI chips has attracted giants to get a share of the pie.

AMD, which is not far behind, launched the Instinct MI300X AI accelerator and Instinct MI300A (the world's first data center APU) in December. AMD claims that the performance of the Instinct MI300 is 4 times higher than the Nvidia H100 GPU in certain workloads, and its performance per watt is twice that of the H100;

Microsoft launched two self-developed AI chips, Azure Maia 100 and Azure Cobalt 100, at the annual Ignite Technology Conference. Among them, the Azure Maia 100 is specially optimized for generative AI and supports various AI scenarios. The Azure Cobalt 100 is based on ARM architecture and targets similar processors from Intel and AMD. The two chips are planned to be opened to partners and customers in the future;

Google has already released its flagship cloud AI chip TPU v4, an optimized TPU v5e chip created for large-scale and mid-scale AI training and inference. Recently, it also launched the TPU v5p AI chip for cloud acceleration, showing a 2.8 times increase in LLM (Big Language Model) training speed;

According to the third-generation AI chip roadmap released by Intel, GAudi3 using a 5nm process will be launched next year, and its computing power will reach twice that of Gaudi2. Additionally, Intel has introduced AI chips for the automotive sector, which can enable more than 12 advanced workloads.

Facing intense competition, Nvidia is also intensively preparing new products. According to the Zhitong Finance App, Nvidia released its new flagship chip H200 in November. According to official data, the H200's memory and bandwidth increased by 76% and 41%, respectively, compared to the H100, and delivery is expected to begin in the second quarter of next year. Nvidia also said that this is only a transition product for the currently popular H100. The next generation GPU product B100 will be released in March next year, and the performance is expected to greatly exceed the H100. To maintain its leading edge in the market, Nvidia announced that the future release pace will change from once every two years to annual updates.

Did Nvidia's $5 billion order fall through?

On what appears to be a bright and easy path, the newly announced chip ban and the expected decline in sales in the Chinese market will be an uncertain factor dragging down Nvidia's performance. In October 2023, the US government issued a new ban to further restrict the export of Nvidia's high-performance AI chips. This new regulation may cause Nvidia's data center revenue from China and other affected regions to drop sharply in the fourth quarter. According to reports, the products involved in compliance requirements continue to contribute about 20% to 25% of data center revenue.

In response, according to market sources, Nvidia is expected to launch a series of three new artificial intelligence chips specifically targeting the Chinese market, namely HGX H20, L20 PCle, and L2 PCle, all based on improvements to the Nvidia H100 to meet the latest US technology export control policies.

According to the Zhitong Finance App, these chips were originally scheduled to be launched at the end of 2023, but this was extended until early 2024 due to policy changes. According to the “Wall Street Journal”, quoting unnamed sources, AI chip orders from major Chinese manufacturers exceeded 5 billion US dollars in 2024, and Nvidia planned to deliver part of it in advance before the ban came into effect, but due to the early entry into force of the ban, Nvidia will lose this 5 billion US dollar order.

In response, Caixin Securities pointed out that the performance of Nvidia's special Chinese version of the AI chip has been drastically reduced, and it is not as cost-effective as before, and it is difficult to match the increasing parameter scale of AI models. However, domestic AI chips are already close to aligning with mainstream overseas products in terms of hardware parameters, and downstream customers are expected to increase their tendency to choose domestic AI chips.

Overall, Nvidia's overall growth trend in 2024 is likely to continue, but interest rate cuts or internal variables, the arms race among chip manufacturers is becoming more heated, and there are still many uncertainties compounded by the international situation, all of which will add some twists and turns to Nvidia's future.