AI芯片市场暗流涌动。 巨头之一谷歌正加快自研AI芯片TPU的商业化步伐，有报道称谷歌正与Meta等科技大厂谈外采合作。在 外界看来，如果合作落地，TPU将进入谷歌体系之外的超大规模数据中心，或对英伟达GPU 主导的算 力市场带来冲击。 相关消息一出，英伟达股价随即震荡。周二美股早盘，英伟达股价一度下滑7%，最终收跌约2.6%。而 自10月29日以来，英伟达市值从5.03万亿美元跌至11月25日收盘的4.32万亿美元，不到一个月时间市值 缩水已超过7000亿美元（约合人民币5万亿元）。 11月26日凌晨，英伟达在社交平台上正面回应谷歌的竞争："我们对谷歌的成功感到高兴——他们在人 工智能领域取得了重大进展，而我们仍将继续向谷歌供货。英伟达领先行业整整一代，是唯一能够运行 所有AI模型，并可在所有计算场景中部署的平台。" 作为全球GPU市场的主导者，英伟达用"领先一代"与"全场景优势"回应这场自研芯片带来的挑战。而即 便谷歌TPU得以进入Meta等巨头的数据中心，也并不意味着GPU会在短期内被替代。事实上，谷歌也表 示，自家定制的TPU和英伟达GPU的需求都在加速增长。 记者丨倪雨晴 编辑丨张伟贤 更多业内观点 ...

记者丨倪雨晴 编辑丨张伟贤 AI芯片市场暗流涌动。 巨头之一谷歌正加快自研AI芯片TPU的商业化步伐，有报道称谷歌正与Meta等科技大厂谈外采合作。在外界看来，如果合作落地，TPU将进入 谷歌体系之外的超大规模数据中心，或对英伟达GPU 主导的算力市场带来冲击。 相关消息一出， 英伟达股价随即震荡。 周二美股早盘，英伟达股价一度下滑7%，最终收跌约2.6%。而自10月29日以来，英伟达市值从5.03 万亿美元跌至11月25日收盘的4.32万亿美元，不到一个月时间市值缩水已超过7000亿美元（约合人民币5万亿元）。 11月26日凌晨，英伟达在社交平台上正面回应谷歌的竞争："我们对谷歌的成功感到高兴——他们在人工智能领域取得了重大进展，而我们仍 将继续向谷歌供货。英伟达领先行业整整一代，是唯一能够运行所有AI模型，并可在所有计算场景中部署的平台。" 作为全球GPU市场的主导者，英伟达用"领先一代"与"全场景优势"回应这场自研芯片带来的挑战。而即便谷歌TPU得以进入Meta等巨头的数据 中心，也并不意味着GPU会在短期内被替代。事实上，谷歌也表示，自家定制的TPU和英伟达GPU的需求都在加速增长。 更多业内观点认为 ...

Everyone on Wall Street has the same Nvidia Corp (NASDAQ:NVDA) debate — "how big is AI demand?" — and almost no one is asking the only question that actually matters: how long can Nvidia keep taxing hyperscalers at 70%+ margins before they revolt? Because the real threat to Nvidia isn't a competing GPU. It's Alphabet Inc‘s (NASDAQ:GOOG) (NASDAQ:GOOGL) Google's TPU — and what it represents: the moment hyperscalers stop outsourcing the most profitable part of AI.Track NVDA stock here.TPUs Aren't Trying To Bea ...

Core Insights - Berkshire Hathaway added only one new stock to its portfolio in Q3, which is a major tech company [1] - Warren Buffett will step down as CEO at the end of the year, but will remain as chairman, ensuring continuity in investment strategy [2] - Buffett's investment strategy focuses on companies with steady growth and reliable profits, typically avoiding technology firms [3] Berkshire Hathaway's Portfolio - The "Magnificent Seven" includes major tech companies like Apple, Amazon, Alphabet, Nvidia, Microsoft, Meta Platforms, and Tesla [4] - Berkshire has invested in three of the Magnificent Seven stocks since 2016, which now represent 23.4% of its $309 billion equity portfolio [5] Amazon - Amazon constitutes 0.8% of Berkshire's portfolio and is involved in e-commerce, cloud computing, streaming, and digital advertising [6] - AWS generated $33 billion in revenue in Q3, a 20% year-over-year increase, marking its fastest growth since Q4 2022 [8] - Berkshire first invested in Amazon in 2019 and is expected to benefit from Amazon's growth driven by AI [9] Alphabet - Alphabet makes up 1.6% of Berkshire's portfolio, with a nearly $5 billion stake acquired in Q3 [11] - The company has integrated AI into its search engine, which has led to a 34% revenue growth for Google Cloud in Q3 [14] - Alphabet's P/E ratio is 27.2, making it the second-cheapest stock in the Magnificent Seven, aligning with Buffett's value investing approach [14] Apple - Apple is the largest holding in Berkshire's portfolio, accounting for 21% [15] - Berkshire invested approximately $38 billion in Apple from 2016 to 2023, with the value of its stake peaking at over $170 billion [15] - Apple's devices are designed for the AI era, and the company has 2.35 billion active devices globally, positioning it as a potential leader in AI software distribution [16][18]

J P M O R G A N North America Equity Research 13 November 2025 Hardware & Networking Key Takeaways from Expert Discussion on Scaling Up AI Datacenters We hosted a conference call with Sri Kanajan, Data Scientist at Scale AI and former Senior Data Scientist at Meta, to discuss the scaling up of AI Datacenters. Please see our takeaways from our last discussion with Mr. Kanajan on Hyperscale AI Datacenter Architecture on June 2025 here, which focused on algorithmic efficiencies shifting compute intensity towar ...

Core Insights - Focusing on technology giants with durable competitive advantages is a strategic investment approach in the current market environment [1] - Amazon's recent $38 billion partnership with OpenAI positions it prominently in the AI cloud sector [1] Cloud Infrastructure Leadership - AWS holds a 29% share of the global cloud infrastructure market, surpassing Azure's 20% and Google Cloud's 13% [4] - In Q3 2025, AWS generated $33 billion in revenue, reflecting a 20.2% year-over-year growth, the fastest since 2022 [6] - AWS's operating income increased by 9.6% year-over-year to $11.4 billion, resulting in an operating margin of over 34% [6] - AWS has an annualized run rate of $132 billion and a backlog of $200 billion, indicating strong revenue visibility [7] AI Capacity - Amazon has added over 3.8 gigawatts of data center power capacity in the past year, with plans to double this by 2027 [9] - The company expects to add at least one additional gigawatt of power capacity in Q4 2025 [9] Custom Silicon Advantage - Amazon's custom chips, Trainium and Inferentia, provide superior price-performance compared to other AI chips [11] - Trainium2 has become a multibillion-dollar business, with a 150% quarter-over-quarter revenue growth in Q3 [11] - The company plans to expand its AI compute cluster to 1 million Trainium2 chips by the end of 2025 [11] - Trainium2 is positioned as 30% to 40% better in price-performance than many GPU options, with Trainium3 expected to deliver 40% better performance than Trainium2 [12] Complete AI Stack - AWS offers platform services like SageMaker and Bedrock, enabling clients to build and deploy custom AI models [14] - The introduction of open-source capabilities like Strands and infrastructure building blocks like AgentCore supports the development of agentic AI [15] Competitive Position - AWS's rapid capacity expansion, custom silicon development, and focus on AI platform services indicate its growing momentum in the AI cloud race [16] - Despite competitors like Microsoft and Alphabet growing faster, AWS is well-positioned to be a significant player in the AI cloud boom [16]

Core Insights - AI datacenters are entering a new phase focused on inference rather than training, which is expected to reshape the competitive landscape and spending patterns in the industry [1][2][8] Shift from Training to Inference - The focus is shifting from training large models to optimizing inference processes, with techniques like distillation and quantization making inference cheaper and more efficient [2][3] - By 2027, inference is expected to dominate incremental compute spending, with a notable shift already occurring in 2025-2026 [3] Beneficiaries of the Shift - Broadcom is highlighted as a key beneficiary due to its custom ASICs that support inference for major companies like Google, Amazon, and Meta [4] - Marvell Technology is also positioned to benefit as inference workloads increasingly rely on Ethernet and PCIe, moving away from expensive training-oriented technologies [5] Hardware and Networking Trends - Celestica is well-positioned as the industry moves towards standardized, cost-effective inference hardware, allowing operators to source from multiple vendors [6] - Arista Networks continues to support high-performance training networks, but the shift towards Ethernet in inference may create new opportunities for networking companies [6] Power Efficiency and Deployment - Inference is significantly less power-hungry than training, often requiring 5-10 times less power, making it easier to deploy in datacenters with limited grid capacity [7] - The trend towards making AI cheaper, faster, and easier to run is expected to drive spending towards companies like Broadcom and Marvell [8]

Core Insights - Amazon's aggressive capital expenditure plans for AWS indicate strong confidence in the growth of the cloud computing market, positioning the stock as an attractive investment opportunity despite short-term margin pressures [1] Financial Performance - The company reported third-quarter operating income of $17.4 billion, showcasing robust profitability while scaling infrastructure investments [2] - AWS generated $27.5 billion in revenues during the quarter, reflecting a 19% year-over-year growth, with an operating margin of 38% [2] - Management guided fourth-quarter revenues between $181.5 billion and $188.5 billion, with expected operating income ranging from $16 billion to $20 billion, indicating sustained profitability despite high capital expenditures [3] Capital Expenditure Plans - Amazon projected capital expenditures exceeding $75 billion for 2024, primarily for AWS infrastructure to support AI workloads and expand data center capacity [3][10] - Increased capital spending addresses the rising demand for AI computing resources, with AWS leveraging custom silicon offerings like Trainium and Inferentia chips to maintain competitive advantages [5] Competitive Landscape - AWS announced significant partnerships with leading AI companies and expanded its generative AI services through Amazon Bedrock, enhancing its machine learning capabilities and launching new data center regions [4] - Competitors like Microsoft and Google are also ramping up capital expenditures, with Microsoft planning nearly $80 billion for fiscal 2025 and Google around $75 billion for 2024, focusing on AI and cloud infrastructure [7] Share Price and Valuation - Amazon shares have returned 13.5% year-to-date, compared to 16.5% for the Zacks Internet – Commerce industry and 8.8% for the Zacks Retail-Wholesale sector [8] - The Zacks Consensus Estimate for Amazon's 2025 earnings is $7.15 per share, reflecting a 29.29% increase from the previous year [12] - From a valuation perspective, Amazon appears overvalued with a forward 12-month price/earnings ratio of 32.9X, higher than the industry's 25.68X [14]

Core Insights - Broadcom has secured a significant $10 billion chip deal with a new customer, which has generated considerable interest among investors [1] - Speculation suggests that the mystery customer could be Anthropic, a competitor to OpenAI, which aligns with Broadcom's strategic interests in the AI sector [2][4] Company Overview - Broadcom operates across more than two dozen business segments, with a notable revenue stream from custom application-specific integrated circuits (ASICs) driven by rising AI demand [3] - The company has established relationships with major hyperscalers, including Meta Platforms, Alphabet, and ByteDance, indicating its strong position in the AI hardware market [3] Potential Customer Analysis - Anthropic has gained traction in the AI landscape since the launch of ChatGPT, receiving substantial investments from Amazon and Alphabet [4][6] - The company has recently partnered with Google Cloud Platform and plans to utilize Alphabet's custom tensor processing units (TPUs), showcasing its diversified approach to chip platforms [7][8] Market Dynamics - Broadcom's existing supply relationship with Alphabet positions it favorably to support Anthropic's growing compute needs as AI workloads become more complex [9] - The overall investment in AI infrastructure is projected to reach $7 trillion through the decade, presenting a significant growth opportunity for Broadcom's custom silicon and networking solutions [14] Investment Perspective - Despite a recent contraction in Broadcom's stock valuation, the company remains well-positioned to attract new hyperscaler customers and deepen existing partnerships in the AI sector [13][15] - Broadcom is viewed as a compelling buy-and-hold opportunity as the demand for AI infrastructure continues to accelerate [15]

Core Insights - Nvidia is the leading provider of artificial intelligence chips, achieving a market cap of $4.5 trillion, with Google as a significant client purchasing GPUs to meet AI compute demands [1][2] - Google is not only a buyer but also a developer of AI chips, recently announcing its most powerful chip, Ironwood, which is designed for heavy AI workloads and is over four times faster than its predecessor [2][3] - Google’s TPUs provide a competitive edge in the cloud market, with a notable increase in cloud revenue by 34% year-over-year to $15.15 billion, driven by strong demand for AI infrastructure [8][9] Company Developments - Google has developed its seventh generation of Tensor Processing Units (TPUs), which are application-specific integrated circuits crucial for AI tasks [3][4] - The company has been proactive in securing large contracts, including a significant deal with Anthropic valued in the tens of billions, expected to bring over a gigawatt of AI compute capacity online by 2026 [12][13] - Google’s cloud segment is experiencing substantial growth, with a backlog of $155 billion and a forecasted increase in capital expenditures to $93 billion for the year [8][21] Competitive Landscape - Google is ahead of competitors like Amazon and Microsoft in deploying custom AI chips at scale, with analysts noting that Google is the only major player to have deployed TPUs in large volumes [5][4] - While Nvidia remains a dominant player in AI chips, analysts suggest that growing familiarity with Google’s TPUs could drive further growth in Google Cloud [22][23] - The demand for TPUs is so high that analysts recommend Google consider selling these systems externally to customers, indicating a closing gap between TPUs and Nvidia’s offerings [23]