PRIMER 23 March 2026 China Semiconductors on 23-Mar-2026 Future of Tech: AI Datacenter Networking Primer Qingyuan Lin, Ph.D. +852 2123 2654 qingyuan.lin@bernsteinsg.com Stacy A. Rasgon, Ph.D. +1 213 559 5917 stacy.rasgon@bernsteinsg.com Francis Ma +852 2123 2626 francis.ma@bernsteinsg.com +852 2123 2694 zheng.cui@bernsteinsg.com Arpad von Nemes +1 917 344 8461 arpad.vonnemes@bernsteinsg.com Alrick Shaw +1 917 344 8454 alrick.shaw@bernsteinsg.com See the Disclosure Appendix of this report for required disclo ...

Core Insights - The report discusses the rapid development of the supernode and Scale-up network industry, highlighting the competitive landscape involving Nvidia, Google, AMD, and Huawei, which is challenging Nvidia's dominance in the market [2][22]. Group 1: Nvidia - Nvidia maintains a leading advantage in supernode technology through NVLink and NVLink Switch, with plans to release advanced solutions like GH200 NVL72 and GB200/GB300 NVL72 between 2024 and 2025, expecting a shipment of approximately 2,800 units by 2025 [4][40]. - The NVLink architecture is designed for high bandwidth and low latency data transmission, with NVLink 5 Switch achieving a single GPU-to-GPU bandwidth of 1,800 GB/s and a total bandwidth of 130 TB/s for 72 GPUs by 2025 [5][40]. - Future developments include the introduction of NVSwitch Gen6 and Gen7, which will further enhance GPU-to-GPU communication bandwidth to 3.6 TB/s [5]. Group 2: Huawei - Huawei is working on the Lingqu protocol, which is transitioning to an open standard, although it has not yet gained widespread acceptance in the domestic industry [6]. - The Atlas 950 supernode, expected to launch in Q4 2026, will feature a total computing power of 8 EFLOPS (FP8) and a memory capacity of 1,152 TB, significantly surpassing Nvidia's offerings [7]. - Huawei's approach involves a hybrid design of copper and optical interconnects to balance complexity, reliability, and power consumption while maintaining system scalability [7]. Group 3: Google - Google has established a mature optical interconnect supernode with its TPU series, including TPU v4, TPU v5p, and TPU v7, which are set to be released between 2023 and 2025 [8]. - The TPU v7 will be utilized by Anthropic, which plans to procure nearly 1 million TPU v7 Ironwood AI chips for deployment in its data centers [8]. - Google's competitive edge lies in its unique application of optical circuit switches (OCS) in Scale-up networks, creating a significant technological barrier against competitors [9]. Group 4: AMD - AMD's UALink has emerged as an important open standard, with the first version released in January 2025 and a second version expected in 2026, gaining widespread industry support [10]. - The Helios supernode is positioned as a strong competitor to Nvidia's NVL72 series, featuring a dual-width rack design that allows for future scalability without redesigning infrastructure [10]. - The Helios rack is anticipated to become a mainstream choice in the industry, with significant advantages in power consumption compared to Nvidia's offerings [10].

Investment Rating - The report maintains a "Positive" outlook on the supernode and Scale-up network industry, highlighting its rapid development and potential as a key infrastructure for AI applications [2]. Core Insights - The supernode and Scale-up network are critical infrastructures that break through computing and communication bottlenecks, supporting trillion-level large models and high real-time applications. The report analyzes the progress and advantages of leading AI computing chip manufacturers, including NVIDIA, Google, AMD, and Huawei, in this field [4][24]. Summary by Sections 1. NVIDIA - NVIDIA's leading advantage in supernode technology is based on NVLink and NVLink Switch. The company plans to launch several mature supernode solutions, including GH200 NVL72 and GB200/GB300 NVL72, with an expected shipment of approximately 2,800 units by 2025 [5][6]. - The NVLink technology enables high bandwidth and low latency data transmission, with NVLink 5 Switch supporting a single GPU-to-GPU bandwidth of 1,800 GB/s and a total bandwidth of 130 TB/s for 72 GPUs [6][40]. - Future developments include the introduction of the Vera Rubin NVL144 and Rubin Ultra NVL576, which will increase the number of interconnected GPUs from 72 to 576 [5][6]. 2. Huawei - Huawei has introduced the Lingqu protocol, transitioning to an open standard from version 2.0, although it has not yet gained widespread acceptance in the domestic industry. The company aims to catch up with NVIDIA in supernode performance through a clustered approach [7][8]. - The Atlas 950 supernode, expected to be released in Q4 2026, will have a total computing power of 8 EFLOPS (FP8) and a memory capacity of 1,152 TB, significantly surpassing NVIDIA's offerings [7][8]. 3. Google - Google has established a mature optical interconnect supernode with its TPU series, including TPU v4, TPU v5p, and TPU v7, which have been recognized by external enterprises [9][10]. - The competitive advantage of Google's TPU supernode lies in its unique application of optical circuit switches (OCS), which creates a high barrier to entry in the optical interconnect field [9][10]. 4. AMD - AMD's UALink has become an important open standard, with the 1.0 version released in January 2025 and the 2.0 version expected in 2026. The UALink ecosystem is anticipated to see significant development by 2027, with over 100 member units supporting it [11]. - The Helios rack from AMD is positioned as a strong competitor to NVIDIA's NVL72 series, featuring a dual-width design that balances complexity, reliability, and performance [11]. 5. Investment Strategy - The report suggests a positive outlook for Google, AMD, and domestic supernode manufacturers, as well as for NVIDIA's supply chain, including PCB backplanes, high-speed copper cables, optical modules, and cooling systems [13][14]. - The market is expected to continue reassessing the value of Google, AMD, and domestic supernode sectors as competition intensifies [13].

Investment Rating - The report assigns a strong buy rating for Astera Labs, anticipating a performance that will exceed the benchmark index by more than 20% over the next six months [52]. Core Insights - Astera Labs reported a robust revenue growth of 92% year-over-year for FY2025Q4, with total revenue reaching $270.6 million. The full-year revenue for 2025 was $852.5 million, reflecting a 115% increase [3][8]. - The company is focusing on expanding its presence in the high-speed interconnect market, particularly in AI deployment and emerging technologies [5][15]. - The strategic emphasis includes significant investments in AI, PCIe Gen 7, and optical interconnect technologies, with expectations of a tenfold market opportunity growth in the next five years [5][15]. Revenue Overview - For FY2025Q4, Astera Labs achieved total revenue of $270.6 million, a 17% quarter-over-quarter increase and a 92% year-over-year increase. The full-year revenue for 2025 was $852.5 million, up 115% year-over-year [3][8]. - The Non-GAAP gross margin for Q4 was 75.7%, with a Non-GAAP operating profit of $108.8 million, resulting in an operating margin of 40.2% [3][8]. Business Segment Performance - Signal Conditioning: The Aries product line saw nearly 70% year-over-year growth, with the Aries Gen 6 being the only PCIe 6 DSP retimer solution available for mass delivery [4][9]. - Intelligent Cable Modules: The Taurus series experienced over fourfold revenue growth in 2025, driven by the mass delivery of 400G designs [4][10]. - Switching Matrix: The Scorpio P series exceeded the target of 10% of total revenue for the year, remaining the only PCIe 6 switching architecture product available for mass shipment [4][11]. - Emerging Products: The Leo CXL memory expansion product made significant progress in 2025, with collaborations with Microsoft, Intel, and SAP expected to lead to initial production in the second half of 2026 [4][12]. FY26Q1 Guidance - Astera Labs expects FY2026Q1 revenue to be between $286 million and $297 million, representing a quarter-over-quarter growth of approximately 6% to 10%. The Non-GAAP gross margin is projected to be around 74%, with earnings per share estimated at $0.53 to $0.54 [5][13].

Group 1 - Marvell Technology, Inc. is gaining attention as an AI stock, with analyst support from Stifel, which reiterated a Buy rating and set a price target of $114.00 [1] - The company has completed the acquisition of XConn Technologies, aimed at enhancing its UALink capabilities in the long term [2] - Following the acquisition, Stifel updated its revenue model, expecting initial contributions to start in fiscal Q3 2027, with an annualized run rate of $50 million by Q4 [3] Group 2 - The acquisition will lead to an estimated increase in FY27 non-GAAP operating expenses by $25 million annually and a decrease in Other Income by approximately $12 million due to reduced interest income from a $325 million cash outlay [4] - The projected diluted weighted-average shares outstanding will increase by approximately 2.7 million shares due to equity issued for the acquisition, resulting in an adjustment of FY27E NG EPS estimate from $3.41 to $3.37, while maintaining FY28E NG EPS at $4.90 [5] - FY29E revenue estimate has been adjusted from $17.58 billion to $17.73 billion, with NG EPS estimate unchanged at $6.68, indicating potential upside if the integration of XConn is successful [5]

Financial Data and Key Metrics Changes - Astera Labs reported Q4 2025 revenue of $270.6 million, representing a 17% increase from the previous quarter and a 92% increase year-over-year [6][24] - For the full year 2025, revenue reached $852.5 million, up 115% compared to the prior year [6][7] - Non-GAAP gross margin for Q4 was 75.7%, down 70 basis points from the previous quarter [25] - Non-GAAP operating margins for Q4 were 40.2%, down 150 basis points from the previous quarter [25] Business Line Data and Key Metrics Changes - Scorpio P-Series exceeded the target of 10% of revenue, becoming the only PCIe 6 fabric shipping in volume [8] - Aries portfolio grew nearly 70% year-over-year in 2025, driven by demand for custom AI accelerators [9] - Taurus revenue grew by more than 4x year-over-year in 2025, supported by new 400G designs [10] Market Data and Key Metrics Changes - The market opportunity for Astera Labs' intelligent connectivity platform is expected to expand to $25 billion over the next five years [13] - The demand for AI infrastructure is being driven by significant capital expenditures from major hyperscalers, with Google and AWS guiding nearly $400 billion in total CapEx spending for 2026 [7] Company Strategy and Development Direction - Astera Labs is strategically investing in expanding its team and capabilities, including the establishment of a new design center in Israel [14] - The company aims to deliver technology enhancements across its core portfolio while expanding into new categories, including custom connectivity solutions and optical engines [17][21] - Astera Labs is focused on building multi-generational partnerships with leading AI platform and cloud service providers [16] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in the growth potential driven by increasing AI infrastructure deployments and the complexity of high-speed interconnect architectures [13] - The company anticipates continued growth for Scorpio P-Series and Scorpio X-Series, with significant revenue opportunities expected in 2026 and beyond [8][19] - Management highlighted the importance of maintaining a strong execution mindset to capitalize on the expanding market opportunities [14][23] Other Important Information - Mike Tate will transition from CFO to a strategic advisor role, with Desmond Lynch joining as the new CFO effective March 2nd [15] - The company is engaged in discussions with multiple customers regarding the adoption of UALink and NVLink Fusion technologies [31][34] Q&A Session Summary Question: What is the visibility on UALink adoption for next-gen designs? - Management confirmed that AWS and AMD have publicly endorsed UALink, indicating strong support and readiness for deployment in 2027 [31][32] Question: What is driving the increase in operating expenses? - The increase is attributed to significant revenue opportunities presented by customers, necessitating investment in R&D and the recent acquisition of a capable team for the Israel Design Center [41][42] Question: How does Astera view the coexistence of UALink and ESUN technologies? - Management believes both technologies will coexist, with customers likely to stick with solutions that align with their existing software stacks [43][44] Question: What is the growth outlook for Scorpio? - Scorpio P-Series has exceeded 15% of sales and is expected to continue growing, with Scorpio X-Series ramping up in the second half of 2026 [47][66] Question: How will customized solutions impact average selling prices and margins? - Customized solutions are expected to have a higher attach rate, although average selling prices may vary due to partner contributions [75]

Financial Data and Key Metrics Changes - Astera Labs reported Q4 2025 revenue of $270.6 million, representing a 17% increase from the previous quarter and a 92% increase year-over-year [6][24] - For the full year 2025, revenue reached $852.5 million, up 115% compared to the prior year [6][7] - Non-GAAP gross margin for Q4 was 75.7%, down 70 basis points from the previous quarter [26] - Non-GAAP operating margins for Q4 were 40.2%, down 150 basis points from the previous quarter [26] - Cash flow from operating activities for Q4 was $95.3 million, with cash, cash equivalents, and marketable securities totaling $1.19 billion at the end of the quarter [27] Business Line Data and Key Metrics Changes - Scorpio P-Series exceeded the target of 10% of revenue, remaining the only PCIe 6 fabric shipping in volume [8] - Aries portfolio grew nearly 70% year-over-year in 2025, driven by demand for custom AI accelerators [9] - Taurus revenue grew by more than 4x year-over-year in 2025, supported by new 400-gig designs [10] - Scorpio X-Series is expected to grow revenue in the first half of 2026, with high-volume production anticipated in the second half [8] Market Data and Key Metrics Changes - The market opportunity for Astera Labs' intelligent connectivity platform is projected to expand by more than 10x over the next five years, reaching $25 billion [13] - The annual merchant scale-up switching market opportunity is expected to grow to roughly $20 billion by 2030 [19] - Strong spending from U.S. hyperscalers, with Google and AWS guiding nearly $400 billion in total CapEx spending for 2026, supports robust market trends [7] Company Strategy and Development Direction - Astera Labs aims to deliver technology enhancements to its core portfolio while expanding capabilities into custom connectivity solutions and optical engines [17][21] - The company is strategically investing in expanding its team and capabilities, including the establishment of an advanced design center in Israel [14] - Astera Labs is focused on building multi-generational partnerships with leading AI platform and cloud service providers [16] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in the growth potential driven by increasing AI infrastructure deployments and the complexity of high-speed interconnect architectures [13] - The company anticipates continued growth for Scorpio P-Series and Scorpio X-Series, with significant revenue opportunities from AI connectivity solutions [28] - Management highlighted the importance of investing in R&D to capitalize on emerging opportunities presented by customers [42][56] Other Important Information - Mike Tate will transition from CFO to a strategic advisor role, with Desmond Lynch joining as the new CFO effective March 2nd [15] - Astera Labs has established partnerships with Microsoft, Intel, and SAP to evaluate CXL memory expansion capabilities [11] Q&A Session Summary Question: Acceptance of UALink for next-gen designs - Management confirmed AWS and AMD's support for UALink, indicating strong customer trust and potential revenue opportunities [32][34] Question: Increase in operating expenses - Management explained that the increase in OpEx is driven by significant revenue opportunities and investments in R&D, including the Xscale acquisition [42][56] Question: Coexistence of UALink and ESON technologies - Management stated that both technologies will coexist, with customers likely to use solutions that align with their software stacks [45][46] Question: Scorpio sales performance - Scorpio P exceeded the 15% sales milestone and is expected to continue growing significantly in 2026 [49][66] Question: Custom solutions and average selling price - Management indicated that custom solutions will have a higher attach rate, although ASPs may vary due to partner contributions [76] Question: Growth of Taurus line - Management noted that Taurus revenue growth is driven by both market growth and Astera's ability to gain market share [78]

Financial Data and Key Metrics Changes - Astera Labs reported Q4 2025 revenue of $270.6 million, representing a 17% increase from the previous quarter and a 92% increase year-over-year [5][24] - For the full year 2025, revenue reached $852.5 million, up 115% compared to the prior year [5] - Non-GAAP gross margin for Q4 was 75.7%, down 70 basis points from the previous quarter [25] - Non-GAAP operating margins for Q4 were 40.2%, down 150 basis points from the previous quarter [25] - Cash flow from operating activities for Q4 was $95.3 million, with cash, cash equivalents, and marketable securities totaling $1.19 billion at the end of the quarter [26] Business Line Data and Key Metrics Changes - Scorpio P-Series exceeded the target of 10% of revenue, becoming the only PCIe 6 fabric shipping in volume [8] - Aries portfolio grew nearly 70% year-over-year in 2025, driven by demand for custom AI accelerators [9] - Taurus revenue grew by more than 4x year-over-year in 2025, supported by new 400-gig designs [10] - Scorpio X-Series is expected to transition to high-volume production in the second half of 2026 [8] Market Data and Key Metrics Changes - The market opportunity for Astera Labs' intelligent connectivity platform is projected to expand by more than 10x over the next five years, reaching $25 billion [12] - Major U.S. hyperscalers, including Google and AWS, are guiding nearly $400 billion in total CapEx spending for 2026, indicating robust spending trends in the AI and cloud infrastructure space [6] Company Strategy and Development Direction - Astera Labs is strategically investing in expanding its team and capabilities, including the establishment of an advanced design center in Israel [13] - The company aims to deliver technology enhancements across its core portfolio while expanding into new categories, including custom connectivity solutions and optical engines [17] - Astera Labs is focused on building multi-generational partnerships with leading AI platform and cloud service providers [16] Management's Comments on Operating Environment and Future Outlook - Management expressed confidence in the growth potential driven by increasing AI infrastructure deployments and the complexity of high-speed interconnect architectures [12] - The company anticipates significant growth in the AI connectivity space, with a focus on addressing customer needs for flexible connectivity solutions [18] - Management highlighted the importance of investing in R&D to capitalize on emerging revenue opportunities presented by customers [41] Other Important Information - Mike Tate will transition from CFO to a strategic advisor role, with Desmond Lynch joining as the new CFO effective March 2nd [15] - The company announced a significant warrant agreement with Amazon, allowing for the purchase of up to $6.5 billion in products based on performance conditions [34][36] Q&A Session Summary Question: What is the outlook for UALink and its acceptance among customers? - Management noted positive endorsements from AWS and AMD regarding UALink, indicating strong customer interest and readiness for deployment in 2027 [31][32] Question: What is driving the increase in operating expenses? - The increase is attributed to significant revenue opportunities and investments in R&D, including the recent acquisition of Xscale and the establishment of the Israel Design Center [41][42] Question: How does Astera view the coexistence of UALink and ESON technologies? - Management believes both technologies will coexist, with customers likely to stick with solutions that align with their existing software stacks [43][44] Question: What is the growth outlook for the Scorpio product line? - Scorpio P has exceeded 15% of sales in 2025, with expectations for continued growth as new customers ramp up production [47][48] Question: How does the company expect to benefit from optical technology? - Management anticipates that optical technology will significantly enhance the scale-up market opportunity, with initial deployments expected around 2028 [86]

Core Viewpoint - The SerDes technology, which has been around for decades, is gaining significant attention in the semiconductor industry due to the increasing demands of artificial intelligence, particularly in data transfer speeds between GPUs [1][3][12]. Group 1: SerDes Technology Overview - SerDes, a combination of Serializer and Deserializer, simplifies data transmission by converting multiple parallel data streams into a single high-speed stream, which is essential for efficient GPU communication [3][9]. - The latest NVIDIA AI system, GB200 NVL72, can exchange approximately 130 TB of data per second between 72 GPUs, illustrating the critical role of SerDes in handling massive data flows [3][4]. Group 2: Market Growth and Investment - The global SerDes market is projected to grow from $745.3 million in 2024 to approximately $2 billion by 2032, with a compound annual growth rate (CAGR) of 13.45% [6]. - Major tech companies like Amazon, Microsoft, Google, and Meta are expected to increase their capital expenditures significantly, with a total projected spending of $224.4 billion in 2024, rising to $315 billion in 2025, much of which will be directed towards SerDes-dependent technologies [6]. Group 3: Technical Challenges and Future Developments - The design and validation of SerDes are notoriously difficult, leading many chip companies to license verified SerDes PHY IP [11]. - The next generation of SerDes, 224G, is set to enter mass production between 2025 and 2026, which will be crucial for maintaining competitive advantages in the AI infrastructure market [11][12]. - The integration of optical communication through co-packaged optics (CPO) is gaining attention as a solution to the limitations of copper wiring at high speeds [10][11]. Group 4: Competitive Landscape - NVIDIA's NVLink is currently the standard for connecting AI accelerators, but the establishment of the UALink standard by AMD, Intel, Google, and Meta could challenge this dominance [12]. - The competition for talent in SerDes design is intensifying, as attracting skilled engineers is becoming a structural challenge for companies in the semiconductor industry [12][13].

Core Viewpoint - The SerDes technology, which has been around for decades, is gaining significant attention in the semiconductor industry due to the increasing demands of artificial intelligence, particularly in data transfer speeds between GPUs [1][3][12]. Group 1: SerDes Technology Overview - SerDes, a combination of Serializer and Deserializer, simplifies data transmission by converting multiple parallel data streams into a single high-speed stream and then back again, akin to a high-speed train carrying goods instead of multiple trucks [3][9]. - The latest AI systems, such as NVIDIA's GB200 NVL72, can exchange approximately 130 TB of data per second, equivalent to streaming 6,000 to 10,000 two-hour Netflix 4K movies in one second [3][4]. Group 2: Market Growth and Investment - The global SerDes market is projected to grow from $745.3 million in 2024 to approximately $2 billion by 2032, with a compound annual growth rate of 13.45% [6]. - Major tech companies like Amazon, Microsoft, Google, and Meta are expected to increase their capital expenditures significantly, reaching $224.4 billion in 2024 and $315 billion in 2025, with a substantial portion directed towards SerDes-dependent components [6][12]. Group 3: Technical Challenges and Developments - The current mainstream 112G SerDes technology transmits data at 112 Gbps per lane, with the next generation 224G expected to be mass-produced by 2025-2026, achieving 1.6T Ethernet [9][10]. - The limitations of copper wires at high speeds necessitate the development of Co-Packaged Optics (CPO), which integrates optical modules directly next to chips to enhance data transmission distances [10][11]. Group 4: Competitive Landscape - The race for 224G mass production is critical, with companies like Synopsys, Cadence, and Marvell currently leading in providing stable solutions [11]. - The establishment of UALink by AMD, Intel, Google, and Meta poses a potential challenge to NVIDIA's NVLink, which is currently the standard for connecting AI accelerators [12][13]. Group 5: Future Outlook - As AI models scale and the need for faster data transfer increases, the importance of SerDes technology will continue to rise, making it a crucial factor in the competition for AI infrastructure [13].