NVDA
NVIDIA
-- 114.060 TSLA
Tesla
-- 284.650 TSM
Taiwan Semiconductor
-- 172.970 PLTR
Palantir
-- 83.420 AMD
Advanced Micro Devices
-- 98.230 Aiming for the throne of NVIDIA! The high-tech giant intensifies the battle for hegemony in self-developed AI chip, focusing on the noteworthy "Custom Chip" stocks selection!
There are two main trends in the current AI chip market. One isCommon GPUs (Graphics Processing Units)and the other isIt is a specific Application Specific Integrated Circuit (ASIC) specialized for specific purposes.NVIDIA holds an overwhelming market share in GPUs for datacenters. Specifically, NVIDIA holds over 90% of the share, effectively in a monopoly state. Although AMD also manufactures GPUs, the structure of NVIDIA's monopoly remains unchanged.
According to the earnings report for the third quarter of 2025 announced at the end of November. $NVIDIA (NVDA.US)$It is expected that the GPU shipments in 2025 will reach 6.5 million to 7 million units, with most of them being the latest.The upcoming NVIDIA GPU accelerator product "Blackwell", including the dominant H100 in the AI semiconductor market, has resolved the previously identified design flaws...NVIDIA's next-generation GPU accelerator product, Blackwell, starting with H100, which has almost a monopoly in the AI semiconductor market, has resolved design flaws previously pointed out and is already in mass production at TSMC. According to last week's report, $Taiwan Semiconductor (TSM.US)$It has been revealed that Apple is in talks with NVIDIA to manufacture NVIDIA Artificial Intelligence Blackwell Chips at a new factory in Arizona.
However, currently, technology giants such as Amazon and Google are increasingly developing their proprietary AI chips, posing challenges to NVIDIA along with other tech giants.
米アマゾン・ドット・コムのクラウドコンピューティング部門アマゾン・ウェブ・サービス（AWS）は3日、自社設計の半導体「トレーニウム（Trainium）」数十万個を搭載する強力な人工知能（AI）スーパーコンピューター「ウルトラクラスター（Ultracluster）」の開発計画と、新たなサーバーを発表した。ブルームバーグによると、米アマゾンは野心的な計画を進めており、現在エヌビディアが支配するAIチップ市場への挑戦を試みている。テキサス州オースティンの研究所で、新型AIチップ「Trainium2」の開発に力を入れている。同社の目標は、このチップを早急にデータセンターに導入し、年末までにテストと配布を完了させることだ。
アマゾンのCEOアンディ・ジャシー氏は第3四半期の決算説明会で自社設計のAIチップ「Trainium2」について、「このチップに対する市場の関心が非常に高く、製造パートナーと何度も協議し、生産計画を大幅に上方修正した」The company states that they are positioning their Veru Inc as a viable alternative to the image processing semiconductor (GPU) developed by the US semiconductor giant NVIDIA. $Adobe (ADBE.US)$such as AI startups Poolside, data platform Databricks, and semiconductor giants are training AI models using 'Trainium2'. $Qualcomm (QCOM.US)$are also training AI models using 'Trainium2'.
● $Meta Platforms (META.US)$
in April of this year, introduced their self-developedsecond-generation AI chip 'MTIA'. 'MTIA' is designed to achieve AI for ranking and advertising recommendations on Meta's Facebook and Instagram, enhancing processing efficiency by incorporating large-capacity SRAM. The second generation of 'MTIA' has improved processing performance threefold compared to the first generation developed last year.in April of this year, introduced their self-developedsecond-generation AI chip 'MTIA'. 'MTIA' is designed to achieve AI for ranking and advertising recommendations on Meta's Facebook and Instagram, enhancing processing efficiency by incorporating large-capacity SRAM. The second generation of 'MTIA' has improved processing performance threefold compared to the first generation developed last year.。
in April of this year, introduced their self-developedsecond-generation AI chip 'MTIA'. 'MTIA' is designed to achieve AI for ranking and advertising recommendations on Meta's Facebook and Instagram, enhancing processing efficiency by incorporating large-capacity SRAM. The second generation of 'MTIA' has improved processing performance threefold compared to the first generation developed last year.in April of this year, introduced their self-developedsecond-generation AI chip 'MTIA'. 'MTIA' is designed to achieve AI for ranking and advertising recommendations on Meta's Facebook and Instagram, enhancing processing efficiency by incorporating large-capacity SRAM. The second generation of 'MTIA' has improved processing performance threefold compared to the first generation developed last year.。
● $Tesla (TSLA.US)$
Electric car manufacturer Tesla is also focusing on the development of AI accelerator chips. In order to improve self-driving technology, so far.Full Self-Driving (FSD)Chips andDojo D1 chiphave been announced. AMD is also gaining attention in the AI chip market. The next generation of
Electric car manufacturer Tesla is also focusing on the development of AI accelerator chips. In order to improve self-driving technology, so far.Full Self-Driving (FSD)Chips andDojo D1 chiphave been announced. AMD is also gaining attention in the AI chip market. The next generation of
● $Advanced Micro Devices (AMD.US)$
AMD is also attracting attention in the AI chip market. The next generation The "AMD Instinct MI350 Series"is expected to debut in the second half of 2025, based on the AMD CDNA 4 architecture. This design aims to improve inference performance by 35 times compared to the previous generation AMD CDNA 3-based accelerators.It is designed to achieve a 35-fold improvement in inference performance.
AMD is also attracting attention in the AI chip market. The next generation The "AMD Instinct MI350 Series"is expected to debut in the second half of 2025, based on the AMD CDNA 4 architecture. This design aims to improve inference performance by 35 times compared to the previous generation AMD CDNA 3-based accelerators.It is designed to achieve a 35-fold improvement in inference performance.
● $Broadcom (AVGO.US)$
Broadcom is positioned as a key supplier of custom AI ASIC chips for global large-scale AI data center Ethernet switch chip and AI hardware domains, and has been focusing on AI investments since 2023.a major supplier of custom AI ASIC chips.In the field of custom AI ASICs, $NVIDIA (NVDA.US)$ and $Marvell Technology (MRVL.US)$They are the main competitors. Because these companies operate on a large scale, it is difficult for other manufacturers to establish a competitive advantage in the market.
Broadcom is positioned as a key supplier of custom AI ASIC chips for global large-scale AI data center Ethernet switch chip and AI hardware domains, and has been focusing on AI investments since 2023.a major supplier of custom AI ASIC chips.In the field of custom AI ASICs, $NVIDIA (NVDA.US)$ and $Marvell Technology (MRVL.US)$They are the main competitors. Because these companies operate on a large scale, it is difficult for other manufacturers to establish a competitive advantage in the market.
Broadcom is currentlyAs a customer of AI chips,including ByteDance, the parent company of TikTok, and OpenAI, $Alphabet-C (GOOG.US)$ 、 $Meta Platforms (META.US)$Broadcom has acquired six companiesincluding ByteDance, the parent company of TikTok, and OpenAI.しており、経営陣は今後5年間でAI半導体は150 billionドル以上のビジネスチャンスがあると見込んでいます。また、水曜日のインフォメーションの報道によると、 $Apple (AAPL.US)$がＡＩチップでブロードコムと協力と伝わった。新たなチップのコードネームは「Baltra（バルトラ）」As a result, it is expected that mass production preparations will be completed by 2026. This move highlights Apple's long-standing position of avoiding chip purchases from NVIDIA, which demonstrates overwhelming strength in the AI processor market. In response to this news, On the 11th in the US stock market, Broadcom's stock price rose by more than 6.63%. The company is scheduled to announce its fourth-quarter earnings after the close of trading on the 12th, and the results are also eagerly anticipated.
ブロードコムのカスタムASICとNVIDIAのGPUの競争はよく話題になるけど、アナリストのMosesmann氏は「比較するのはリンゴとオレンジみたいなもの」と言ってる。彼は、「今後数年でカスタムASICがGPUコンピュートを上回る成長を見せるだろう」とも話していてる。これがブロードコムのAIアクセラレーション、生成AI、クラウドやエッジでのネットワーキング分野での成長の可能性を示しているんだ。
The stock price of the US semiconductor giant Marvel Technology hit a record high on December 4th,Its market capitalization surpassed $100 billion (approximately 15 trillion yen) for the first time, surpassing the long-established semiconductor manufacturer Intel.Marvel Technologyhas risen by 85% year-to-dateand demonstrated strong performance in both performance and growth. Marvel announced sales of $1.520 million for the third quarter, with an adjusted earnings per share of $0.43, exceeding analyst estimates of sales of $1.450 million and earnings per share of $0.41. The datacenter division, which accounts for 70% of revenue, performed well, driven by robust AI demand. The company expects an increase in revenue due to the rapid growth in AI demand and raised its performance outlook for the fourth quarter, anticipating revenue to reach $1.710 million to $1.890 million and adjusted earnings per share to reach a maximum of $0.65.
In addition, Marvel recently agreed with Amazon on a 'multi-generational' contract to support Amazon in designing its own artificial intelligence (AI) semiconductors for a period of 5 years."Contracts spanning multiple generations".Amazon, the world's largest player in the cloud computing services sector with Amazon Web Services (AWS), is significantly expanding its in-house chip development business, partly due to its desire to reduce its dependency on NVIDIA for important AI components. Recently, at the annual developer conference, Amazon announced the next-generation model of its self-designed semiconductor, 'Trainium.'According to analysts, Trainium is expected to be a key factor in more than doubling the sales of Marvel's custom AI division in January 2026.
Marvell's CEO Matt Murphy emphasized during the third-quarter earnings call that the contracts Marvell has signed with customers for custom chips span 'multiple generations.' It is predicted that by 2028, Marvell will capture 25% of the market share of data center acceleration chips, potentially bringing the company over $40 billion in revenue.
Mark Lipacis, an analyst at Evercore ISI, predicts that the total revenue of the custom AI chip industry is expected to reach $30 billion to $50 billion by 2030. In a recent note to clients, he pointed out that Marvel could 'capture a third of that market.'
Source: MOOMOO, Dow Jones, Bloomberg
This article uses automatic translation in part.
- MOOMOO News: Alicia
This article uses automatic translation in part.
- MOOMOO News: Alicia
Disclaimer: Moomoo Technologies Inc. is providing this content for information and educational use only.
Read more
Comment
Sign in to post a comment
株の慶次 : We are certainly developing our own software, but there is a question of whether the total cost of software, personnel, and development required is really lower than purchasing from NVIDIA, and whether we can surpass the quality of NVIDIA.
やっぱテスラは最高 : When considering the disadvantages of depending on the NVIDIA H100 and H200, price and electricity will become a hindrance to the future. I think I'll overcome the funds to buy large amounts of GPUs with customizations that specialize in the company, and now purchasing companies also have financial power, and I see actions that rely on NVIDIA in the future as a risk.
Amazon AWS has decided to collaborate on ASIC development for the next 5 years because of its design and development capabilities based on Marvel's 16,000 patents, and I'm looking forward to seeing other amazing products preserved in the product group currently under development.