“Intel Meta Google Is New AI Semiconductors” How big of a threat is it to NVIDIA?! Which companies will benefit?
$NVIDIA (NVDA.US)$ ofcontentionsIt is $Intel (INTC.US)$ and,customersof $Meta Platforms (META.US)$ or Google (parent company $Alphabet-A (GOOGL.US)$ ) isWe are all strengthening the development of new AI semiconductorsI'm doing it. Disgusted by it,NVIDIA's stock price temporarily fell nearly 5% on 4/9I did it. It was warned that the movements of each company would pose a threat to NVIDIA. After 4/10, NVIDIA stocks rebounded due to a bullish outlook from analysts, but trading volume has not increased, and it seems that concerns about each company's strengthening of new AI semiconductors have not completely disappeared.
This time,How much of a threat will the actions of each company pose to NVIDIAI would like to consider it based on the challenge results of AMD, which is more competitive than Intel.
・April 9:Intel launches new AI semiconductorsAnnouncing “Gaudi3”
・April 9:”Google develops new CPU in-house “Reducing the Costs of AI Competition” (Wall Street Journal report, 4/10 in Japanese)
・April 10:AI semiconductors developed in-house by Meta“MTIA” 2nd generation announced
Main products of AI semiconductorsWhen I checked,NVIDIA, which has a market share of 80% or more, is overwhelmingly aheadI'm doing it. The companyThe “H100,” which has the highest performance, was announced in 2022/3It goes back to In other words, it was 2 years before now. Due to the rapid expansion of demand, which benefited from the subsequent generative AI boom,Sales of the company's data center division (the division where sales of AI semiconductors are recorded) for the next fiscal year were 5 times that of the previous yearAs a result, it greatly exceeded market expectations.
In order to capture rapidly expanding demand,2nd place in market share $Advanced Micro Devices (AMD.US)$Will announce the “MI300,” a product that rivals the “H100” in 2023/12I did it.At that timeAlso,Voices concerned about the decline in NVIDIA's market shareIt went up. However, with the announcement of financial results for the first time this yearNVIDIA Announces Good Financial Results That Exceed Market ExpectationsI did it.Demand for NVIDIA products continues to be strongSo,AMD's threats are limited and proven by experienceI did it.
2024 MarchNVIDIA is currently the best-selling AI semiconductor“The most powerful semiconductor in the world” superior to the “H100”(From NVIDIA's announcement)Announced the “B200” and “GB200” equipped with “Blackwell” and “Blackwell”I did it. At that time, the company said, “Generative AI is a technology that defines the present day. Blackwell is the engine that will drive this new industrial revolution,” he explained.“GB200” is LLM compared to “H100”(Large-scale language model)30x for reasoning, 4x for LLM trainingThe performance of has been improved,25 times more energy efficientIt becomes. In other words,Much higher performance than AMD's “MI300”That's it.
When NVIDIA announced “Blackwell,”AWS, Google Cloud, Microsoft Azure, Oracle Cloud InfrastructureIt was introduced that the four major cloud service providers will be the first cloud service providers to provide instances using Blackwell. $Amazon (AMZN.US)$ and Google, $Microsoft (MSFT.US)$ 、 $Oracle (ORCL.US)$ but“Blackwell” customersIt means that it is.
It lags behind AMDIntel April 9, 2024to,“Gaudi3,” a counterpart product to the “H100,” was announcedI did it. Based on the current schedule, IntelWhen shipping “Gaudi3”(scheduled for 2024/7-9),NVIDIA is already expected to ship next-generation AI semiconductorsIt has become.
As mentioned above,NVIDIA is overwhelmingly aheadFurthermore, considering that they are taking the lead and launching the strongest product in the world without giving up on it,It can be said that the threat of AMD and Intel is not that high。
This does not mean that AMD and Intel have no business opportunities at all.The AI semiconductor market pie is growing due to the generative AI boomBecause,Both AMD and Intel have gained a certain amount of marketIt looks like it can be done. It's just,Not enough to threaten NVIDIA's dominant positionSeemably.
Also,AMD and Intel stock prices are soft compared to NVIDIANanohaThe fact that there is a sense that stock prices are expensive compared to performance expectations also has an effectIt is thought that they are doing it.
NVIDIA's AI semiconductorsCustomers such as Meta and Google manufacture AI semiconductors in-houseWhat is progressingIt's not necessarily new news. In fact, not only Meta and Google, but also Microsoft and Amazon are developing AI semiconductorsI've been going there for a few years. For example, Microsoft first announced AI semiconductors developed in-house in 2023/11, and they have already been introduced in some data centers.
In other words,Over the past few years, major tech companies have continued to procure from NVIDIA at the same time as promoting in-house production. In AI semiconductorsThis is probably because our products still fall short of NVIDIA's advantage. That's probably how it should be. Even AMD, which ranks second in the industry for AI semiconductors after NVIDIA, is laps behind NVIDIA.
★ Why are companies strengthening the development of AI semiconductors in the first place?
As a main factor, in addition to tight supply and demand, advantages of in-house production can be cited.
As a main factor, in addition to tight supply and demand, advantages of in-house production can be cited.
1) Tight supply and demand
From the demand side,Demand for AI semiconductors has rapidly expanded since last year due to the generative AI boomDoing it. On the other hand,NVIDIA, which has a market share of 80% or more, can supply AI semiconductors in large quantitiesIt has become.
From the demand side,Demand for AI semiconductors has rapidly expanded since last year due to the generative AI boomDoing it. On the other hand,NVIDIA, which has a market share of 80% or more, can supply AI semiconductors in large quantitiesIt has become.
As a result,A situation where tech giants are unable to procure the required amountIt's in. For example, at NVIDIA's February financial results presentation, analysts asked, “What standards (priorities) are assigned to customers?” In other words, from the customer's point of view, tight supply and demand can become a procurement risk. Tech giants want to acquire AI semiconductors as soon as possible in order to quickly introduce AI-related products and services. From the perspective of tech giants with financial resources,Incentives to embark on in-house development work。
2) Advantages of in-house production
In the case of in-house production, first,So that procurement costs can be reduced. Since NVIDIA has an advantage in terms of position in AI semiconductors, tech giants are forced to accept the high price (also high performance) of NVIDIA products. From the perspective of major tech companies, there may be cases where semiconductors with such high performance are not necessary depending on usage conditions. If that part can be manufactured in-house, the total cost can be reduced. Next, our products are more in line with usage conditions and requirementsCustomized products can be made。
In the case of in-house production, first,So that procurement costs can be reduced. Since NVIDIA has an advantage in terms of position in AI semiconductors, tech giants are forced to accept the high price (also high performance) of NVIDIA products. From the perspective of major tech companies, there may be cases where semiconductors with such high performance are not necessary depending on usage conditions. If that part can be manufactured in-house, the total cost can be reduced. Next, our products are more in line with usage conditions and requirementsCustomized products can be made。
Therefore,In-house production of AI semiconductors by major tech companies does not necessarily mean that they will directly compete with NVIDIA or that they will not procure them from NVIDIALet's say that. In fact, in the article “Google develops new CPUs in-house to reduce costs in AI competition” that appeared at the beginning,”Vice President of Google's In-House Chips DivisionAmin Bardat of“I see this as a foundation for increasing the size of the pieIt says, “I talked about it.” Bloomberg analysts pointed out that “Google may at least think its company can do better than Intel (not NVIDIA)” (NVIDIA's AI semiconductor “H100” is a GPU).
Overall, the threat posed by in-house production by major tech companies is limited in the short termIt is conceivable. On the other hand, tech giantsWhen an AI semiconductor with a certain level of high performance was successfully developedFrom NVIDIAPotential to reduce procurementThere might be one.It is necessary to keep an eye on the movements of major tech companies in the medium to long termThere will be one. Also, is it a message in response to that,NVIDIA CEO Jensen HuangRecently,The total cost can be reduced by introducing our own productsIt was emphasized.
$Taiwan Semiconductor (TSM.US)$
Semiconductor contract manufacturing (TSMC, the world's largest company, can benefit from (foundry)It seems to be.metaThe second generation of the new AI semiconductor “MTIA”TSMCThe 7 nanometer process technology used in the first “MTIA” has been improvedManufactured using a 5 nanometer processIt has been reported.
Semiconductor contract manufacturing (TSMC, the world's largest company, can benefit from (foundry)It seems to be.metaThe second generation of the new AI semiconductor “MTIA”TSMCThe 7 nanometer process technology used in the first “MTIA” has been improvedManufactured using a 5 nanometer processIt has been reported.
Intel is also strengthening its foundry business, and the company announced on 4/3 that it will take several years for sales to decline and surplus conversion for the foundry business. It may still be a long way for Inteln to take on the challenge of TSMC.
Also,TSMC announced on 4/10 that sales for March increased 34% from the same month last yearI did it. In January-March, it was a 16% increase compared to the same period last year, which exceeded market expectations. The Bloomberg article said, “This is the highest growth in the past year or so,The global AI development boom confirmed the expectation that demand for high-end semiconductors and servers would be spurred” it was reported.
$Broadcom (AVGO.US)$
According to the WSJ article dated 4/9 that reported Google's AI semiconductors,”Google produces custom hardwareA major semiconductor product companyClose collaboration with Broadcom since 2016Doing it”. Also, “Broadcom's Chief Executive Officer Hok Tan (CEO) stated in an internal presentation in March that the business of Broadcom's custom chip division has rapidly increased since Google recently rapidly expanded production of TPU (semiconductor developed in-house by Google).”Google's strengthening of in-house development may continue to benefit Broadcom。
According to the WSJ article dated 4/9 that reported Google's AI semiconductors,”Google produces custom hardwareA major semiconductor product companyClose collaboration with Broadcom since 2016Doing it”. Also, “Broadcom's Chief Executive Officer Hok Tan (CEO) stated in an internal presentation in March that the business of Broadcom's custom chip division has rapidly increased since Google recently rapidly expanded production of TPU (semiconductor developed in-house by Google).”Google's strengthening of in-house development may continue to benefit Broadcom。
Created April 12, 2024 Market Analyst Amelia
Source: Created by Moomoo Securities based on company materials, Bloomberg, and various materials
Source: Created by Moomoo Securities based on company materials, Bloomberg, and various materials
Disclaimer: Moomoo Technologies Inc. is providing this content for information and educational use only.
Read more
Comment
Sign in to post a comment
182639895こんさん : Dear moomoo, it was a wonderful analytical skill and document. Thank you very much.
トトロはっぱ : It's easy to understand.
182639895こんさん : The person who wrote this comment to Moomoo is an excellent person who is diverse, flexible, and very hardworking. Thank you very much.
今若俊明 : It's an amazing analytical ability, and even an amateur could understand it.
またぬき : Two years behind is fatal in this industry. Nvidia's superiority of semiconductors targeting generative AI will remain unwavering for quite some time.
tetsu_77 : Intel's ability to handle it will be clear