J P M O R G A N North America Equity Research 12 December 2025 Communications Infrastructure 2026 Outlook: Crucial Year for Hyperscale AI Capacity Delivery as AI Infrastructure Needs Evolve; Towers Poised for Growth, but EchoStar Uncertainty Remains The data center industry went into overdrive in 2025, with incremental capacity builds doubling from last year, and we look for even more supply to be brought online in 2026. This year, AI-focused companies have announced a wave of deals totaling tens of billion ...

Accessible version Technology - Asia-Pacific Semis/hardware materials: raise EMC/ TUC/MPI POs on Google ASIC benefits Price Objective Change Rising TPU demand favors supply chain in different ways Considering likely rising demand for TPU into 2026/27, we (see report) upgrade GUC and WT Micro to Buy. On the semiconductor/hardware materials side, we also flag EMC, TUC, and MPI as the key beneficiaries within Google's in-house ASIC supply chain, and we reiterate Buy on these three names due to further expanded ...

Core Insights - Meta is in discussions with Alphabet to purchase billions of dollars' worth of TPU chips, potentially shifting the AI infrastructure landscape away from Nvidia's GPUs [2][3] - Google's TPUs are custom silicon optimized for machine-learning workloads, which could challenge Nvidia's long-standing dominance in AI hardware [4][6] - If Google successfully commercializes its TPUs, it could lead to a significant shift in the AI hardware ecosystem, affecting margins and competitive advantages [6][12] Company Implications - Nvidia's GPUs have been the standard for AI model training, but Google's TPUs may offer a cheaper and more efficient alternative for specific workloads [4][11] - Meta's capital expenditure plans suggest a substantial investment in inferencing-chip capacity, indicating a potential shift in demand from Nvidia to Google [6][12] - The success of Google's TPUs could lead to a reevaluation of Nvidia's growth assumptions and market position [12][13] Industry Dynamics - The introduction of TPUs could lower the total compute costs for AI workloads, prompting major technology companies to reconsider their reliance on Nvidia hardware [10][12] - If TPUs gain significant market share, it could redefine the competitive landscape in the AI industry, potentially positioning Google as a leader [12][13] - The shift from GPUs to TPUs for AI inferencing could lead to broader implications for cloud platforms and AI startups, affecting pricing and cost structures [12][13]

Core Insights - The market is becoming increasingly aware of the potential of Google's developments, particularly in relation to its cloud services and Tensor Processing Units (TPUs) [1][2] - Analysts are questioning how competitors like NVIDIA will respond to Google's advancements, especially after NVIDIA's significant investment in OpenAI [3][4] - The competition in the AI and cloud computing space is intensifying, with companies like Alphabet, Amazon, and Alibaba aiming for vertical integration in their offerings [12][15] Company Developments - Alphabet has been developing its TPUs for over ten years and has started to market them more aggressively, particularly to high-frequency trading firms [2][4] - The efficiency of Google's Gemini 3 model is highlighted as a competitive advantage, showcasing the effectiveness of its technology stack [4][11] - Alphabet's strategy includes not only hardware development but also software integration, aiming to provide a comprehensive ecosystem for AI applications [10][11] Industry Dynamics - The competition among major players like NVIDIA, Google, and Amazon is expected to drive innovation and efficiency in AI infrastructure [7][8] - The market is witnessing a shift towards energy efficiency as a critical factor for success, with companies focusing on optimizing their energy use [16][17] - Analysts are observing a divergence in stock performance among tech companies, indicating a need for investors to be discerning in their evaluations [18][21] Market Sentiment - Despite recent fluctuations in stock prices, there is a belief that the long-term outlook for AI CapEx remains positive, driven by competition and innovation [9][22] - The current market environment is characterized by a rotation into value-focused sectors, reflecting investor caution towards tech stocks [21][24] - The emotional pulse of the market suggests a reset in valuations, with potential opportunities for investors to identify undervalued stocks [25]

Core Insights - Alphabet's stock reached all-time highs following the release of its latest AI model, Gemini 3.0, which has climbed to the top of third-party AI rankings and was trained entirely on Google's custom Tensor Processing Units (TPUs) [1][2] Group 1: TPU Development and Impact - TPUs, previously viewed as a niche tool, have demonstrated their capability to train and serve advanced AI systems at a global scale, indicating a significant shift in Google's technical execution [2] - The re-engagement of co-founder Sergey Brin and the return of Nom Shazir, who were instrumental in establishing Google's AI advantage, coincides with this development [3] Group 2: Competitive Landscape - Other companies, including Anthropic, Apple, OpenAI, and Meta, are also adopting or testing TPU infrastructure, with Anthropic committing to utilize up to a million TPUs over the next few years, valued at tens of billions of dollars [3] - While GPU demand is not expected to decline immediately, the rise of TPUs suggests a potential reduction in Nvidia's dominance in the long term, positioning Google favorably in the competitive landscape with its integrated stack of chips and ecosystem [4]

Group 1 - Alphabet Inc. is expected to double in value over the next three years, driven by its strategic partnerships and investments in AI technology [1] - Google has entered into a significant agreement with AI startup Anthropic PBC, involving the supply of up to 1 million specialized tensor processing units (TPUs) [1][2] - The deal is valued at tens of billions of dollars and is set for TPU deployment in 2026, which will enhance Anthropic's computing capacity significantly [2] Group 2 - The partnership allows Anthropic to access advanced chip infrastructure, reducing its reliance on Nvidia's GPUs, which are currently scarce and expensive [2] - Google has previously invested approximately $3 billion in Anthropic, including $2 billion in 2023 and an additional $1 billion earlier this year [2] - Alphabet Inc. operates through various segments including Google Services, Google Cloud, and Other Bets across multiple regions globally [3]

Summary of Conference Call Notes on Alphabet Inc. and GCP Company and Industry Overview - **Company**: Alphabet Inc. (GOOGL) - **Industry**: Internet and Cloud Computing - **Market Cap**: $3,095,474 million as of October 23, 2025 - **Stock Rating**: Overweight - **Price Target**: $270.00 Key Points and Arguments Anthropic and Google Cloud Partnership - Anthropic has announced a significant expansion with Google Cloud, including access to up to 1 million TPUs, which is expected to enhance the capacity for training and serving Claude models [1][3] - This deal is projected to contribute approximately $9 billion to $13 billion annually to Google Cloud revenue in 2027, with a potential upside of 100-900 basis points to 2026 revenue [1][3][7] - The agreement is valued in the tens of billions of dollars and is expected to provide over a gigawatt of capacity by 2026 [1][3] Growth Projections - Google Cloud revenue is forecasted to grow by 35% in 2026, significantly above market expectations [3] - Anthropic's expected top-line CAGR is around 150% from 2025 to 2027, indicating strong growth potential [1] Competitive Landscape - Despite the partnership with Google Cloud, Anthropic continues to rely on AWS as its primary cloud service provider for training, raising questions about AWS's competitive position [2][8] - The TPU v7 Ironwood chip, designed for inference, is set to ramp in 2026, which may shift workloads on GCP towards inference tasks [2] Financial Implications - The estimated spending by Anthropic with GCP over six years could range from $50 billion to $80 billion, aligning with the scale of the deal [3] - The pricing model for the TPUs, including potential discounts and capacity phasing, remains uncertain, which could impact revenue projections [7] Risks and Considerations - There are concerns regarding AWS's ability to compete effectively for Anthropic's incremental capacity, which may depend on performance and compute efficiency [8] - The overall market dynamics and competition in the AI cloud space could influence future growth and revenue for both GCP and AWS [8] Analyst Insights - Analysts express confidence in Alphabet's long-term growth driven by AI innovations across its platforms, including Search and YouTube, which are expected to enhance revenue and EBITDA growth [21][24] - The consensus rating distribution shows 84% of analysts rating Alphabet as Overweight, indicating strong market confidence [26] Additional Important Information - The report highlights the importance of ongoing partnerships and technological advancements in maintaining competitive advantages in the cloud computing sector [2][24] - The anticipated ramp-up of new TPU models and their efficiency improvements could play a crucial role in driving future revenue growth for Google Cloud [9][13] This summary encapsulates the critical insights from the conference call regarding Alphabet Inc.'s strategic moves in the cloud computing space, particularly through its partnership with Anthropic, and the implications for future growth and competition in the industry.

Investment Rating - The report maintains an Overweight (OW) rating on Broadcom (AVGO) and Marvell (MRVL) [35][49]. Core Insights - The high-end custom ASIC chip market is projected to reach $30 billion by CY25, growing at a 30% CAGR, driven by the aggressive adoption of AI technologies [5][11]. - Broadcom and Marvell are positioned as market leaders, holding 55-60% and 15% market shares respectively, and are expected to dominate the ASIC market due to their strong design capabilities and extensive customer pipelines [8][11]. - The demand for custom ASICs is increasing as cloud/hyperscale companies seek differentiation, better performance, and lower costs compared to off-the-shelf solutions [9][11]. Summary by Sections Market Opportunity - The high-end ASIC market is estimated to be a $30 billion opportunity with a 30% CAGR, primarily fueled by the demand for AI compute accelerators [5][11]. - Custom AI ASICs are expected to account for 40% of the AI XPU/GPU market this year, indicating a significant shift towards custom solutions [19]. Company Performance - Broadcom is on track to generate approximately $19.5 billion in AI revenues this fiscal year, with expectations to exceed $31 billion next fiscal year, reflecting a 60% year-over-year growth [5][9]. - Marvell is ramping production of its first two AI ASIC programs and is projected to achieve $4 billion in AI revenue this year, up from $1.8 billion last year [10][11]. Design and Technology - Both Broadcom and Marvell possess advanced design capabilities, including the ability to handle complex chip designs with over 100 billion transistors, which is a key differentiator in the market [6][20]. - The emergence of custom ASICs is driving demand for EDA software and critical IP, benefiting companies like Synopsys, Cadence, and ARM [7][11]. Customer Engagement - Broadcom and Marvell have secured significant design wins with major cloud/hyperscale customers, ensuring a strong pipeline for future growth [23][32]. - The report highlights the collaborative nature of ASIC development, where cloud titans partner with semiconductor companies to leverage their design expertise and manufacturing capabilities [8][20].