来源：环球网 【环球网科技综合报道】3月24日消息，《The Information》报道，英伟达在GTC 2026大会上推出的 MGX ETL开放标准化机架，首次实现自家芯片与第三方AI芯片的混合部署。这一设计不仅允许客户集 成谷歌、AMD等厂商的AI XPU，还通过标准化架构化解了捆绑销售争。 据悉，混合部署与高效互联MGX ETL机架属于英伟达Vera Rubin POD系统的一部分，该系统包含五款 专用机架级扩展系统，均采用第三代MGX设计。其中，Vera Rubin NVL72基于MGX NVL机架，支持最 高256颗芯片，采用Spectrum-X以太网或Groq 3 LPU直接芯片到芯片链路铜质脊线连接方案；其余四款 则基于MGX ETL机架，若客户使用第三方芯片，则仅支持Spectrum-X以太网连接。 《The Information》报道称，英伟达方面不会限制合作伙伴在MGX中使用替代组件，这意味着客户可将 谷歌、AMD的AI XPU与英伟达网络芯片混合部署于MGX NVL中。（青山） ...

Core Viewpoint - Broadcom's transformation from Avago to a successful hybrid software and hardware company is attributed to strategic acquisitions and a focus on artificial intelligence (AI) opportunities, positioning it to compete effectively in the semiconductor market [4][5]. Financial Performance - For Q1 FY2026, Broadcom reported sales of $19.31 billion, a year-over-year increase of 29.5%, with operating profit rising 36.8% to $8.56 billion and net profit increasing by 33.5% to $7.35 billion, representing 38.1% of total revenue [8]. - The semiconductor solutions segment saw a revenue increase of 54.2% to $12.52 billion, driven primarily by AI chip business [8]. - The infrastructure software group generated $6.8 billion in sales, with VMware contributing approximately $5.2 billion, reflecting a 13% growth [10]. AI Business Growth - AI chip and system revenue is projected to exceed $8.44 billion in Q1 FY2026, with significant contributions from AI networking chips and AI accelerators [15]. - Broadcom has established partnerships with six major AI clients, including Google and OpenAI, to develop custom XPU solutions, with expectations of strong demand for AI products in the coming years [16][17]. - The company anticipates AI revenue to reach a minimum of $100 billion by FY2027, with projections indicating substantial growth from $19.3 million in FY2022 to $127.4 billion in FY2024 [17]. Market Position and Competition - Broadcom aims to enhance its market position against competitors like Nvidia and AMD by focusing on custom AI hardware solutions, potentially capturing significant market share in the AI and networking sectors [5][14]. - The competitive landscape is intensifying, with companies like Astera Labs and Marvell also vying for market share in the semiconductor space [14]. Future Outlook - For Q2 FY2026, Broadcom expects sales of approximately $22 billion, a 47% year-over-year increase, with semiconductor solutions projected to grow 76% to $14.8 billion [18].

Core Viewpoint - Arm Holdings plc (NASDAQ:ARM) is recognized as a promising stock by hedge funds, with a recent upgrade from Susquehanna from Neutral to Positive, setting a price target of $150 due to significant initiatives in AI and custom server CPU development [1] Group 1: Company Initiatives - Arm is collaborating with SoftBank and Broadcom to produce an AI XPU ASIC for OpenAI, which is considered a transformative initiative for the company [1] - The development of a custom server CPU, potentially for Meta, marks Arm's first full silicon product, further enhancing its product offerings [1] Group 2: Financial Projections - Susquehanna estimates that the AI XPU program could generate over $1 billion in annual royalty revenue, contrasting with the projected total company royalties of $2.5 billion for the current year [2] - Despite anticipated challenges in the Mobile and PC markets leading to a royalty default for 2026 models, the stock's recent 40% drop is viewed in light of potential growth opportunities that could exceed Arm's current modest royalty contribution of approximately 10% [3] Group 3: Market Position - Arm Holdings designs and licenses CPU architectures, system IP, and software utilized across various sectors, including automotive, computing, consumer, and IoT applications [3]

Core Insights - Marvell Technology (MRVL) is currently trading at a forward 12-month price-to-earnings (P/S) ratio of 7.67X, which is above the Zacks Electronics - Semiconductors sector's average of 7.44X, indicating a premium valuation [1] Group 1: AI and Data Center Growth - Marvell Technology is benefiting from the AI wave, leading to increased demand for custom silicon and interconnect semiconductor products, including AI XPU and digital signal processors [2] - The custom silicon business is expected to contribute approximately 25% to overall data center revenues, with a projected growth rate of at least 20% next year [3] - In Q3 fiscal 2026, Marvell's data center segment revenues reached $1.52 billion, reflecting a year-over-year increase of 37.8%, driven by strong demand for custom XPU silicon and next-generation switches [4][7] Group 2: Product Innovations and Partnerships - Marvell's Ara 200G/lambda 1.6T PAM4 optical DSP enhances AI networking capabilities, while the Teralynx switch telemetry API provides real-time analytics for network performance [5] - The acquisition of Celestial AI is expected to accelerate Marvell's momentum in the interconnect space, positioning the company at the forefront of next-generation AI data center architectures [8] - Strategic partnerships with Amazon Web Services (AWS) and NVIDIA are enhancing Marvell's offerings in AI and data center workloads, integrating advanced technologies into its custom silicon solutions [9][10] Group 3: Competitive Landscape and Market Position - Marvell faces competition from semiconductor giants like Broadcom and Advanced Micro Devices (AMD) in the AI accelerator market, but its rapid portfolio expansion helps it keep pace with market trends [11][13] - The company is developing advanced networking solutions to support memory-intensive AI workloads, which is crucial for maintaining its competitive edge [13] - Zacks Consensus Estimates indicate a revenue growth rate of 42% and an earnings growth rate of 81% for fiscal 2026, with upward revisions in earnings over the past 60 days [14] Group 4: Stock Performance and Recommendations - Marvell's shares are trading above the 200-day and 50-day moving averages, indicating a bullish trend [17] - The company is recommended as a buy due to its strong fundamentals, deep AI portfolio, and strategic partnerships, making it a key player in the AI ecosystem [18][19]

Core Insights - Marvell Technology's AI XPU offerings are gaining significant traction among hyperscalers, AI data centers, and high-performance computing workloads, with data center revenues reaching $1.52 billion in Q3 FY26, a 37.8% year-over-year increase [1][10] Group 1: Revenue and Growth - The custom silicon business is expected to contribute approximately 25% of overall data center revenues and is projected to grow at least 20% next year [2] - Marvell has announced 18 XPU and XPU-attach socket design wins, many of which are already in volume production, contributing to a revenue pipeline exceeding 10% of a $75 billion lifetime revenue funnel [3][10] Group 2: Technology and Competitive Position - Marvell develops custom ASIC designs for AI and cloud data center customers using advanced 5nm and 3nm processes, which include various high-speed SerDes technologies, positioning the company ahead of competitors in the custom AI silicon market [4] - Marvell's deep partnership with NVIDIA provides essential support to maintain its competitive edge against other players like Broadcom and Advanced Micro Devices [7] Group 3: Valuation and Earnings Estimates - Marvell Technology's shares have increased by 13.1% over the past six months, compared to the Electronics - Semiconductors industry's growth of 22.3% [8] - The company is trading at a forward price-to-sales ratio of 7.31X, which is lower than the industry's average of 8.46X [11] - The Zacks Consensus Estimate for Marvell's fiscal 2026 and 2027 earnings indicates year-over-year growth of 81% and 26%, respectively, with upward revisions in estimates over the past 30 days [12]

Core Insights - Marvell Technology's AI XPU offerings are gaining significant traction among hyperscalers, AI data centers, and high-performance computing workloads, with 18 XPU/XPU-attach sockets and over 50 new pipeline opportunities estimated to generate $75 billion in lifetime revenue potential [1][10] Group 1: Custom Silicon Development - The company has heavily invested in custom silicon programs to scale production rapidly, introducing a 2.5D advanced packaging platform to reduce power consumption and overall product costs for customers [2] - Marvell Technology has developed a first-of-its-kind 64 Gbps/wire Bi-Directional die-to-die interface IP in 2nm, offering over 3x UCIe bandwidth density while reducing die area usage to 15% [3] - The company also designs custom application-specific integrated circuits (ASICs) for AI, cloud data centers, and OEM customers using advanced 5nm and 3nm processes, enhancing its competitive edge in the custom AI silicon space [4] Group 2: Financial Performance - Marvell Technology's data center segment led all segments with a 69% year-over-year revenue growth in Q2 of fiscal 2026, driven by strong demand for custom AI accelerators and high bandwidth memory chips, with expectations for continued double-digit growth in the upcoming quarter [5] - The company trades at a forward price-to-sales ratio of 8.48X, which is higher than the industry's average of 7.78X [12] - The Zacks Consensus Estimate for Marvell's fiscal 2026 and 2027 earnings indicates year-over-year growth of 78.3% and 19.3%, respectively, with the estimate for fiscal 2026 remaining unchanged and the estimate for fiscal 2027 revised upward in the past 60 days [13] Group 3: Competitive Landscape - Broadcom is a leading supplier of custom silicon solutions for data centers, with its advanced 3.5D XDSiP packaging platform designed to enhance the performance of custom AI XPUs [6] - Advanced Micro Devices (AMD) competes in the custom silicon solutions and AI accelerator space with its semi-custom SoC offerings and Instinct Accelerators, which are utilized in numerous data centers [7] - Despite competition from Broadcom and AMD, Marvell Technology's strong partnership with NVIDIA provides essential support to maintain its position in the custom silicon market [8]

Core Viewpoint - Morgan Stanley identifies Broadcom (AVGO.US) as a top investment choice in the semiconductor sector due to its diversified business model, high profit margins, and strong long-term growth trends in AI/data centers and infrastructure. The firm maintains an "overweight" rating with a target price of $400 by December 2026, indicating a potential upside of approximately 16% from the recent closing price of $346.17 [1] Group 1 - Strong AI demand continues, with Broadcom's team making expected progress on potential collaborations, confident that clients will transition to million-unit XPU AI clusters [2] - If CEO Hock Tan increases AI revenue to $120 billion by 2030, related incentives from his recent employment contract will be maximized, with the team having a history of exceeding targets [2] - Broadcom's team is expected to complete the first 2nm/3.5D packaged AI XPU product this year with a potential client, establishing a leadership position in technology, even ahead of independent GPU peers [2] Group 2 - Broadcom is moving to a higher level by not only delivering chips but also providing clients with complete AI rack-level deployments, resulting in increased gross and operating profit despite lower margins due to cost pass-throughs [2] - There are strong growth opportunities in scale-up, scale-out, and scale-across networking, with Ethernet architecture remaining the preferred technology [2] - Orders in non-AI business are on the rise, with signs of recovery in broadband, server storage, and enterprise networking, although revenue recovery is expected to follow a "U-shaped" pattern, potentially accelerating after entering fiscal 2026 [2] Group 3 - The transition of VMware VCF platform is expected to be completed by the end of fiscal 2026, maintaining strong double-digit year-over-year growth in revenue, with infrastructure software growth projected to return to mid-single-digit percentages thereafter [3] - Broadcom plans to focus on reinvesting in business, particularly in large AI growth opportunities, while continuing to return 50% of free cash flow and prioritizing debt repayment [3] - Broadcom is a leader in wireless, data center networking, AI/deep learning ASICs, storage, and infrastructure silicon, with broad positive exposure across these end markets, ensuring its leadership position in a diversified market [3]

Core Viewpoint - Morgan Stanley's report highlights strong demand trends in artificial intelligence (AI) driving growth for Broadcom's custom AI XPU business and high-performance networking product portfolio [1][3] Group 1: AI Demand and Market Potential - AI demand remains robust, with inference workloads creating additional demand that was not included in the previous 2027 serviceable available market (SAM) forecast, indicating potential upside [2] - The execution in product technology development is strong, with 2nm 3.5D AI XPU products expected to complete tape-out this year, and the first generation AI XPU products also anticipated to complete tape-out, with partnerships established with key potential clients [2][3] Group 2: Networking and Non-AI Business Recovery - AI computing workloads are driving strong networking demand, significantly enhancing unit value in expanded networks, providing substantial revenue capture opportunities [2] - Non-AI semiconductor business is experiencing a "U-shaped" gradual recovery, with observed order/reservation inflection points expected to drive earnings per share (EPS) upgrades next year [2] Group 3: VMware and Overall Financial Outlook - The VMware Cloud Foundation (VCF) platform transition is expected to continue driving strong growth for VMware business, projected to reach an annual revenue of $20 billion by the end of 2026 or 2027, with growth rates returning to mid to high single digits thereafter [2] - Despite some margin pressure in AI XPU business, overall semiconductor operating profit margins are expected to continue expanding due to operational leverage, with management focusing on reinvestment in AI growth opportunities rather than acquisitions [3] Group 4: Market Position and Valuation - Broadcom is positioned as the second-largest AI semiconductor supplier globally and the largest custom chip (ASIC) supplier, with a strong presence in wireless communication, data center networking, AI/deep learning ASICs, storage, and infrastructure silicon chips [3][4] - Morgan Stanley maintains a price target of $325 for Broadcom by December 2025, representing approximately 18% upside from the closing price of $275.4 on July 10 [3]

Investment Rating - The investment rating for Broadcom Inc is Overweight (OW) with a price target of $325.00 by December 2025 [2][17]. Core Insights - Strong AI demand is driving growth in Broadcom's custom AI XPU business and high-performance networking portfolio, with incremental demand from inference workloads not previously accounted for in forecasts [1][13]. - The non-AI semiconductor business is showing signs of recovery, which could lead to positive EPS revisions next year [1][15]. - VMware's revenue growth is expected to be sustained until late 2026/2027, with an anticipated annualized revenue base of $20 billion [1][15]. - Semiconductor operating margins are expected to expand due to operating leverage, despite potential gross margin headwinds from the AI XPU business [1][15]. - Broadcom is focusing on reinvesting in AI growth opportunities rather than pursuing M&A, which could dilute resources [1][15]. Summary by Sections AI Demand and Product Development - AI demand remains strong, particularly in inference workloads, which could lead to upside in previous forecasts [1][13]. - The company is on track to tape-out first-generation AI XPU products this year, collaborating with leading prospects like Arm/Softbank and OpenAI [1][13]. - AI compute workloads are driving significant networking demand, presenting opportunities for higher dollar content capture in scale-up networking [1][13]. Non-AI Business Recovery - A gradual "U" shape recovery is observed in the non-AI semiconductor business, reflected in increasing bookings and orders [1][15]. - This recovery could contribute to a positive EPS revision cycle next year [1][15]. VMware Business Growth - The VMware VCF platform is expected to sustain strong revenue growth until late 2026/2027, with projected annualized revenues reaching approximately $20 billion [1][15]. Financial Performance and Margins - Semiconductor operating margins are anticipated to continue expanding due to strong operating leverage, despite lower gross margins from AI XPU demand [1][15]. - AI revenues are projected to grow at 60% year-over-year in FY26, with operating expenses not expected to grow at the same elevated levels [1][15]. Strategic Focus - The company plans to reinvest in business growth over the next 1-2 years, focusing on AI opportunities and customer engagement rather than M&A [1][15].

Core Viewpoint - Nvidia may be prominent in AI discussions, but Broadcom is recognized for its critical role in building the infrastructure necessary for AI, being the "2 global AI semiconductor supplier" and a leader in custom ASICs [1] Group 1: AI Infrastructure and Strategy - Broadcom's AI strategy is focused on its custom AI XPU business, which is currently thriving, with plans to tape-out first-generation AI XPU products this year [2] - The company is collaborating with four AI customers, including OpenAI, to develop massive AI clusters of "1 million XPU" units, positioning Broadcom at the center of AI development [2] Group 2: Networking and Performance - AI models require not only computing power but also ultra-fast networking, where Broadcom excels, offering 5–10 times more dollar content per unit compared to traditional networks [3] - Broadcom's leadership in next-generation interconnects and silicon photonics strengthens its status as a preferred vendor for high-performance infrastructure [3] Group 3: Software Growth Potential - VMware's Cloud Foundation (VCF) platform is projected to sustain strong growth until late 2026/2027, with revenues expected to reach $20 billion before transitioning to mid- to high-single-digit growth [4] - Despite lower gross margins on custom AI chips, Broadcom is anticipated to expand operating margins due to business leverage [4] Group 4: Business Focus - The company is committed to reinvesting in its business rather than pursuing distractions such as mergers and acquisitions [5]