Our cost per token is the lowest in the world, you can't beat it. I've said before, if you have the wrong architecture, even if it's free, it's not cheap enough. And the reason for that is because no matter what happens, you still have to build a gigawatt data center. You still have to build a gigawatt factory.And that gigawatt factory for 15 years advertised across that gigawatt factory is about $40 billion. Even when you put nothing on it, it's $40 billion in. You better make for darn sure you put the bes ...

The promise of Agentic AI is enormous. The challenge is that most enterprises are sitting on a mountain of fragmented data and legacy systems that were never designed to support AI-driven workflows.This is where Dell and NVIDIA are stepping in—with a validated, jointly engineered platform that connects data, AI and real-world customer experiences. Sponsored by @DellTech https://t.co/O8Rqvqktc6 ...

Marvell Technology Group Ltd | North America Quick Thoughts on NVDA/ MRVL Partnership NVIDIA and Marvell partner on NVLink and silicon photonics, with a $2bn investment, underscoring how severe networking bottlenecks have become and highlighting the strategic positioning of Marvell's portfolio as AI scaling drives greater reliance on advanced interconnect. What Happened: NVIDIA and Marvell announced a partnership to integrate NVIDIA's NVLink ecosystem with Marvell's XPU and scale-up networking portfolio. Ma ...

Flash | 31 Mar 2026 07:17:14 ET │ 8 pages Taiwan Semiconductors and Electronic Components & Equipment AI chip structural upgrade cycle intact CITI'S TAKE Companies Mentioned: ASE Technology Holding (3711.TW; NT$328.5; 1; 31 Mar 26; 13:30) | King Yuan Electronics Co (2449.TW; NT$261.0; 1; 31 Mar 26; 13:30) | NVIDIA Corp (NVDA.O; US$165.17; 1; 30 Mar 26; 16:00) | TSMC (2330.TW; NT$1760.0; 1; 31 Mar 26; 13:30) | Unimicron (3037.TW; NT$444.5; 1; 31 Mar 26; 13:30) | WinWay Technology (6515.TW; NT$7150.0; 1H; 31 ...

MARCH 2026 See Disclosure Appendix of this report for important disclosures and analyst certifications ARTIFICIAL INTELLIGENCE: THE AI INFRASTRUCTURE VALUE CHAIN Our fact base and framework for comparing fundamental upside and identifying winners across the AI value chain. The AI boom versus bubble debate has raged on since 2023. The range of outcomes could scarcely be wider, from advanced superintelligence to a crash of epic scale. While it is clearly difficult to pinpoint the exact trajectory, we see AI s ...

Dissecting Nvidia Blackwell - Tensor Cores, PTX Instructions, SASS, Floorsweep, Yield 剖析英伟达 Blackwell 架构——张量核心、PTX 指令、SASS、残次品利用、良率 Microbenchmarking, tcgen05, 2SM MMA, UMMA, TMA, LDGSTS, UBLKCP, Speed of Light, Distributed Shared Memory, GPC Floorsweeps, SM Yield 微基准测试、tcgen05、2SM MMA、UMMA、TMA、LDGSTS、UBLKCP、光速指标、 分布式共享内存、GPC 残次品利用、SM 良率 KIMBO CHEN AND DYLAN PATEL KIMBO CHEN 与 DYLAN PATEL APR 01, 2026 2026 年 4 月 1 日 ∙ PAID ∙ 付费内容 136 1 Share 分享 Nvidia's Datacenter Blackwell GPU (SM100) represents one of the ...

Summary of TSMC and CPO Industry Conference Call Company and Industry Overview - The conference call primarily discusses TSMC's advancements in Co-Packaged Optics (CPO) technology and its implications for the semiconductor industry, particularly in relation to NVIDIA and other key players in the market [1][2][3]. Key Points and Arguments CPO Development and Production - CPO is still in the R&D phase, with unclear mass production yield rates. Significant volume production is expected to begin with the Feynman architecture, which will incorporate 3D stacking technology [1][2]. - NVIDIA has decided to skip the Near Packaged Optics (NPO) solution, focusing on CPO to address supply bottlenecks and thermal interference issues associated with copper cables [1][7]. - TSMC is leading the front-end wafer-level processes for CPO, while SPIL is responsible for back-end packaging. CPO capacity is expected to start in 2028, with significant volume production in 2029 [1][13]. Yield and Technical Challenges - Current yield rates for CPO are low, with R&D yields sometimes reaching 50%-60%, but mass production yields remain uncertain due to differences between R&D and scale production [3][6]. - The main technical challenges involve heterogeneous integration, particularly the assembly and layout optimization of optical engines on 2.5D packaging substrates [3][4]. Market Dynamics and Future Outlook - Onto Innovation has sold out its 2026 capacity, with revenue growth expected to be between 38%-50% due to strong demand for interposers and CoWoS-related products [1][14][19]. - The G5 equipment from Onto is currently under validation with TSMC and Micron, aiming to enter the market in 2026 [1][16][29]. Competitive Landscape - Samsung is aggressively introducing Hybrid Bonding technology in HBM4, while SK Hynix and Micron are cautious due to cost considerations [1][13]. - TSMC's CPO business is seen as a new growth area, with high technical barriers and limited competition, aligning with TSMC's strategy of focusing on high-moat businesses [7][10]. Revenue and Capacity Planning - TSMC's CPO business is still in the early stages, with no clear production orders from NVIDIA for the Ruben generation, making discussions about capacity expansion premature [8][20]. - The company is facing significant pressure on capacity due to high demand, with orders extending into 2027. Strategies to increase capacity include restarting closed U.S. factories and utilizing outsourced production in Southeast Asia [22][29]. Pricing and Profitability - The company is considering price increases in response to high demand, with a gross margin target of 58%. The first quarter of 2026 is expected to see margins between 54%-56% [23][24]. Product Positioning - The G3 and G5 series products are positioned to coexist, targeting different application areas. G5 aims to enter the CoWoS detection market, complementing G3 rather than replacing it [24][26]. Additional Important Insights - The G3 Plus is still in the conceptual stage, aiming to enhance G3's speed by 30%-50% [25]. - The G5 equipment's production is currently limited to in-house manufacturing, with standardization and external production processes yet to be established [29]. - The outlook for 2027 is optimistic, with potential revenues exceeding $15 billion if G5 passes customer validation [29][30].

Summary of OCS Switch Industry Conference Call Industry Overview - The OCS (Optical Circuit Switch) technology is expanding its application from TPU cabinet connections to cross-data center Scale-across connections, with Oracle placing orders in 2026 and companies like Microsoft, Meta, and NVIDIA entering prototype testing phases [1][2] Market Dynamics - The market size is driven by the number of channels, with mainstream products expected to upgrade from 128 channels to over 300 channels by 2026. Total demand is projected to reach 70,000 to 80,000 units by 2028, with an anticipated annual growth rate potentially doubling [1][3] - Current demand from Google is over 10,000 units annually, with an upgrade path to 300+ channels by 2026. New customers like Oracle are expected to contribute an additional 1,000 to 2,000 units [3][4] Key Technologies - The main OCS technology solutions include: - **MEMS**: Most mature and widely used - **Liquid Crystal**: Developed in collaboration with Coherent and Microsoft, expected to have a longer lifespan and begin mass production in 2026 - **Optical Waveguide**: Short switching delay but currently limited to 32 channels, with 64 channels still under development [1][4] - BOM costs for all solutions are approximately $200 per channel, while the selling price is around $500 per channel [4] Supply Chain Insights - Domestic suppliers play crucial roles in the OCS supply chain, focusing on machine assembly, core materials, and optical components: - **Tengjing Technology**: Core supplier for liquid crystal solutions, expected revenue over 200 million yuan in 2026 [1][5] - **Dekeli**: Focuses on optical waveguide machine assembly, with samples sent to Google and NVIDIA [5][6] - **Lingyun Optics**: Collaborating on piezoelectric ceramic solutions [6] Challenges and Future Trends - The main supply-side bottleneck is low assembly efficiency, limiting production capacity. Current output is significantly below planned levels, but larger-scale production is expected to begin by Q3 2026 [1][6] - Key validation points for future development include: - Increased production capacity and cost reduction - Successful application of OCS technology in more scenarios, particularly in ASIC or NVIDIA's Scale-Up interconnect solutions, which could accelerate industry trends [6]

Core Insights - Chinese GPU and AI chip makers captured nearly 41% of China's AI accelerator server market in 2025, significantly reducing Nvidia's market dominance [2][5] - The shift is attributed to China's increasing caution about reliance on foreign chips, leading to a push for domestic alternatives amid U.S. export controls [3][5] Market Share Dynamics - Nvidia remains the market leader with approximately 2.2 million AI accelerator cards shipped, holding a 55% market share, but this represents a decline from its previous dominance [4][7] - Chinese vendors collectively shipped 1.65 million cards, marking a significant milestone in the market, with Huawei leading at around 812,000 cards, followed by Alibaba's T-Head with approximately 265,000 cards [5][6] Competitive Landscape - Baidu's Kunlunxin and Cambricon each shipped around 116,000 cards, ranking them jointly third among Chinese vendors [6] - Other Chinese vendors like Hygon, MetaX, and Iluvatar CoreX accounted for 5%, 4%, and 3% of total shipments, respectively [6] Government Initiatives - In 2025, the central government initiated a new wave of AI infrastructure spending, encouraging local governments to accelerate the establishment of intelligent computing centers with directives to prioritize domestic products [6]

Core Viewpoint - The Islamic Revolutionary Guard Corps (IRGC) of Iran has declared that 18 U.S. tech companies operating in the Middle East, including major firms like Nvidia, Apple, Microsoft, and Google, will be considered "legitimate targets" for attacks in retaliation for U.S. and Israeli actions against Iran [1][2]. Group 1: Threats and Targets - The IRGC has announced that attacks on the identified tech companies will commence at 8 p.m. Tehran time on April 1, warning employees to evacuate their workplaces for safety [2]. - The list of targeted companies includes Cisco, HP, Intel, Oracle, IBM, Dell, Palantir, JP Morgan, Tesla, GE, Spire Solutions, Boeing, and UAE-based AI company G42 [3]. Group 2: Context and Response - U.S. tech firms have been increasing their investments in the Middle East, particularly in AI infrastructure, due to the region's low energy costs and land availability [4]. - Intel has stated that the safety and wellbeing of its team is the top priority and is taking measures to protect its workers and facilities in the region [4].