The rapid expansion of computing power demand caused by the new wave of AI has not only put the king of GPUs on the altar, but also made Intel, the CPU overlord, a little lonely.
According to the financial report, in the first quarter of 2024, Intel's revenue will continue to grow, but its profit will not improve significantly.
It also gave second-quarter guidance: revenue of $12.5 billion to $13.5 billion, below analysts' expectations of $13.63 billion, adjusted EPS of $0.2, below analysts' expectations of $0.24, and gross margin of 43.5%, below analysts' expectations of 45.3%.
This seems to show Intel's lack of confidence in development, but in the face of competition, Intel still has momentum, judging from its big bet on AI chips and chip foundry, Intel is finding the lost throne in the new AI era.
Under the good performance, what worries does Intel have?
According to the financial report data, in the first quarter of 2024, Intel's revenue was US$12.724 billion, a year-on-year increase of 8.61%, and its net loss was US$437 million, a year-on-year decrease of 84.21%.
It can be seen that Intel's overall business situation is significantly better. In response, Intel CEO Pat Gelsinger also said in a statement: "Strong innovation in our client, edge, and data center portfolios has driven Intel's revenue growth. ”
However, judging from the performance of enterprises that are still in the red, Intel's business still has a lot of room for optimization. Take Intel's two major business segments that have high hopes - data center and artificial intelligence business, and foundry business as an example.
According to the financial report, Intel's data center and artificial intelligence business revenue is about 3 billion US dollars, although it increased by 5% compared with the same period last year, but under the new round of data center investment peak driven by the AI boom, such a growth rate does not seem to have significantly benefited from industry growth.
Specifically, in order to consolidate the foundation of the competition for AI, giants such as Amazon, Microsoft, and Google have invested heavily in data center business, such as Google's capital expenditure in the fourth quarter of last year increased by 45% year-on-year to $11 billion, of which nearly half was used for data infrastructure construction.
Driven by strong demand, Nvidia's data center business revenue reached about $18.4 billion in the fourth quarter of fiscal 2024 (ending January 28), a year-over-year surge of 409%.
In contrast, the growth potential of Intel's related businesses seems to be unleashed.
In fact, for a long time, whether it is the PC market or the data center market, Intel's advantages have been concentrated in the CPU (central processing unit) field, but since 2021, artificial intelligence technology has developed rapidly, and because Al servers are usually equipped with acceleration chips based on GPUs (graphics processing units), GPUs have begun to become the protagonists of the global data center incremental market.
According to data from Verified Market Research, the global GPU market size will be 33.5 billion yuan in 2021, and the global GPU market size is expected to reach 477.4 billion yuan in 2028, and the dominant player in the data center market has also become NVIDIA.
However, in this round of transition, Intel also faces certain opportunities, which involves the development of its foundry business.
In 2021, Intel proposed the IDM2.0 strategy, breaking the tradition of "self-made chips" and operating chip production independently to meet market demand.
Specifically, with the rapid expansion of demand for GPU-based AI chips, the bottleneck of chip production capacity is becoming increasingly prominent, for example, Nvidia mainly uses TSMC foundry, but due to TSMC CoWoS advanced packaging capacity limitations, in the second half of 2023, Nvidia will have a situation where the delivery cycle of high-performance computing cards is generally extended to 12-16 months. In this context, the importance of Intel's layout of chip foundry is significant.
In addition, the United States and the European Union have introduced chip bills to provide huge subsidies for European and American companies to expand their wafer foundries. All these benefits have provided a rare driving force for Intel to increase its wafer foundry.
However, judging from the first-quarter financial report, Intel's foundry business does not seem to have really entered the harvest period. According to the financial report, in the first quarter of 2024, Intel's foundry business achieved revenue of $4.4 billion, a year-on-year decrease of 10%, and an operating loss of $2.5 billion.
Such a performance is not difficult to understand. First of all, the cost of fab construction is extremely high. According to agency estimates, the cost of building a 2nm factory with a monthly output of 50,000 wafers is about 28 billion US dollars (about 199.844 billion yuan), while the cost of a 3nm factory with the same capacity is about 20 billion US dollars (about 142.746 billion yuan).
In order to open up space for business development, Intel is currently strengthening the construction of foundry infrastructure. It is reported that since 2023, Intel has successively announced plans to build semiconductor manufacturing plants in the United States, Europe and Israel, with a total investment of up to 100 billion US dollars with subsidies from local governments.
Secondly, it is understood that the Intel foundry sector is formed by the merger of the original chip manufacturing related departments in the enterprise system, which has been serving internal orders for a long time and does not need to independently bear profits and losses, which also leads to its service to external customers, open up a larger market, or need a certain adaptation period.
On the whole, Intel is still in the stage of deep adjustment, and from the latest layout, Intel will still accelerate its bet on data center and foundry business: at present, the company has spent a huge amount of money to purchase the most advanced High NA EUV lithography machine, and in April, it released a new generation of AI chip Gaudi 3, benchmarking NVIDIA H100.
And in the fierce market competition, is Intel's big attack enough to be confident?
Fighting back on two fronts, can Intel return to the throne??
At present, the rapid development of AI technology is constantly driving the expansion of the intelligent computing power market, and the data shows that the average annual compound growth rate of the scale of intelligent computing power in mainland China is expected to reach 33.9% from 2022 to 2027.
In the face of explosive market demand and increasingly competitive situation, Intel's aggressive strategy is still reflected in the product side, such as the release of Gaudi 3, which is seen as a challenge to NVIDIA's leadership in the AI chip market.
Intel claims that the Gaudi 3 has an average of 50% better inference power, an average of 40% more energy efficiency, and 100 times faster than the H100 compared to the Nvidia H3 chip. At the same time, the Gaudi 3 also has an extremely important advantage, "the cost is only a fraction of the NVIDIA H100", which undoubtedly makes the downstream demand side full of expectations.
In fact, the current price rise caused by the difficulty of finding a GPU card has also affected the layout of AI by downstream technology companies. It is reported that the price of H800, which can be bought in China, is as high as 200,000 yuan/piece, and if the computing power demand reaches 2000P, the cost of the card alone will reach 200 million yuan. In this case, more cost-effective products are undoubtedly more popular in the market.
The price advantage of Gaudi 3 may be due to the fact that Intel does not use all the most cutting-edge technology. For example, in terms of process, Gaudi 3 uses TSMC's 5nm process, while in terms of memory, Gaudi 3 does not use the latest HBM3 (High Bandwidth Memory).
To a certain extent, this has also caused the market to doubt the competitiveness of Intel's products. Because the H100 benchmarked by Gaudi 3 is a product released by Nvidia based on the Hopper GPU architecture two years ago, and last month, Nvidia released the B100 based on the BlackWell GPU architecture, which has undergone a round of significant upgrades, far better than Intel's Gaudi 3 in terms of performance. In this case, can Intel really break through with new products?
In fact, judging from Intel's current performance, it may be a smarter product strategy not to pursue cutting-edge configurations, but to greatly improve chip performance under the premise of controlling costs.
At present, NVIDIA has an absolute technological and ecological leadership in AI accelerated computing, in this situation, if Intel catches up at all costs, the cost pressure is bound to rise sharply (in the first quarter of this year, Intel's R&D expenses were 4.382 billion US dollars, accounting for 34.4% of total revenue), and it will take a long time to return to positive profits. Therefore, through insight into market pain points, we can create differentiated products or more secure choices.
In order to explore greater growth space, Intel also has a major strategy to promote the development of foundry business. As mentioned above, the layout of this field requires a lot of investment, and in the case that it is difficult to reduce fixed expenses such as expanding production capacity, another means to control costs is to improve cost efficiency. Therefore, it can be seen that Intel has successfully won the world's first high-NA EUV lithography machine, which may continue to reduce Intel's manufacturing costs over time.
In addition, according to the financial report, Intel is about to complete the "four years and five process nodes" plan, of which Intel 7, Intel 4 and Intel 3 have achieved mass production. In this way, Intel may be able to win more orders in the future.
In this regard, Intel CEO Gelsinger also said: "With the development of Intel foundry, we are expected to regain the leading position in the process next year, and we are confident in the plan to promote sustained growth throughout the year." ”
However, judging from the continuous competition in the industry and the industrial attributes of chip foundry "heavy assets and long cycle", Intel, a veteran chip company, still has many tough battles to fight.
Author: Jianbai
Source: U.S. Stock Research Agency