➀ Broadcom demonstrated that its Tomahawk 5 TH5-Bailly CPO (co-packaged optics) solution offers higher reliability compared to traditional pluggable FR4 optics, with a 2.5M-hour MTBF;

➁ Lab stress tests showed significantly lower link failure rates and zero link flaps in the first 1 million device hours for CPO, alongside reduced power consumption;

➂ The results position CPO as a viable solution for AI clusters, potentially accelerating adoption in high-density networking environments.

➀ Broadcom showcased its Tomahawk 6 102.4T switch chips and Tomahawk Ultra at OCP Summit 2025, highlighting significant size differences due to SerDes speeds (112G vs. 224G);

➁ The 112G SerDes variant features a larger package for additional I/O lanes, while the 224G version’s higher-speed SerDes enables a compact design;

➂ The Tomahawk Ultra 51.2T chip, distinct from Tomahawk 6 in architecture and package size, underscores Broadcom’s push for next-gen Ethernet scalability in AI/HPC clusters.

➀ Broadcom has started shipping its Tomahawk 6-Davisson 102.4T switch featuring integrated co-packaged optics (CPO), achieving a 70% reduction in power consumption;

➁ The switch employs TSMC's COUPE photonic engines and 3nm Condor SerDes, with field-replaceable laser modules for improved maintenance;

➂ This innovation addresses high-power networking demands in AI clusters, enabling higher port density and energy-efficient 1.6TbE connectivity.

➀ OpenAI collaborates with Broadcom to develop its first custom AI chip ("XPU"), targeting mass production by 2026;

➁ The move aims to reduce reliance on NVIDIA's GPUs and strengthen control over AI infrastructure;

➂ The chip design combines Broadcom's semiconductor expertise with OpenAI's AI optimization requirements.

➀ Broadcom launched Jericho4, a 3nm Ethernet fabric router designed to interconnect over 1 million XPUs across distributed AI data centers, addressing power and space constraints;

➁ It offers 3.2T HyperPort technology, MACsec encryption, and lossless RoCE over 100km+, enabling high-bandwidth, secure, and efficient multi-data-center AI infrastructure;

➂ Jericho4 complies with Ultra Ethernet Consortium (UEC) standards, ensuring interoperability with ecosystem components like NICs and switches.

➀ Broadcom Jericho4 51.2Tbps AI router chip is now shipping, targeting large AI clusters with 3.2Tbps HyperPorts and 3nm process technology;

➁ Features 100Km+ RoCE deployment for long-distance AI cluster communication and 200G PAM4 signaling;

➂ Part of Broadcom's expanded switch/routing portfolio, including Tomahawk 6 and Tomahawk Ultra, for high-tier AI network infrastructure.

➀ The HPE ProLiant MicroServer Gen11 features an upgraded design with four 3.5" SATA bays and tool-less drive installation;

➀ It includes two PCIe slots for expansion, enhanced cooling with a single large fan, and Broadcom 1GbE quad-port NICs;

➀ HPE retains iLO remote management and introduces a DC power retention clip for improved stability.

➀ Broadcom launched the Tomahawk Ultra, a 51.2Tbps Ethernet switch chip optimized for small packet sizes and ultra-low latency (250ns), targeting HPC and AI clusters;

➁ The chip challenges NVIDIA's NVLink and InfiniBand by offering 'Scale-Up Ethernet' (SUE) with features like in-network collectives, compressed headers, and lossless fabric technology;

➂ Pin-compatible with Tomahawk 5 and shipping alongside Tomahawk 6, it emphasizes scalability for AI workloads while maintaining Ethernet compliance.

➀ Q1 global fabless semiconductor revenue rose 6% quarter-over-quarter to $77.4 billion, driven by strong market demand;

➁ NVIDIA led with $42.3 billion (up 12% QoQ), followed by Qualcomm ($9.47 billion, down 6%) and Broadcom ($8.2 billion, up 2%);

➂ AMD, MediaTek, and others like Marvell and Realtek rounded out the top ten, reflecting varied performance across the semiconductor sector.

➀ Broadcom unveiled its Tomahawk 6 series switches at Computex 2025, featuring 102.4T bandwidth and 64-port 1.6TbE configurations.

➁ Delta and Wistron/Wiwynn showcased prototype switches, with Delta's 3U design focusing on airflow cooling and Wistron/Wiwynn's 4U model emphasizing enhanced thermal management.

➂ Limited photography was permitted due to embargo constraints, but more Tomahawk 6-based hardware is expected to emerge following the official announcement.

➀ Broadcom launched the Tomahawk 6 series switch ASICs, supporting 102.4Tbps bandwidth and up to 64 ports of 1.6TbE, marking the transition to terabit Ethernet;

➁ The chips enable both scale-up configurations (e.g., connecting HBM4 memory across 512 XPUs) and scale-out architectures for AI clusters, reducing network tiers and latency;

➂ Features like Global Load Balancing 2.0, Cognitive Routing, and enhanced telemetry optimize data flow and cost efficiency, positioning Broadcom as a key player amid competition with NVIDIA in AI infrastructure.

➀ Broadcom's Tomahawk 6 switch series delivers 102.4 Tbps switching capacity, doubling current Ethernet switch bandwidth, targeting AI networks with 100G/200G SerDes and co-packaged optics (CPO);

➁ It supports AI cluster interconnects for over one million XPUs, featuring Cognitive Routing 2.0 for dynamic load balancing and adaptive flow control tailored for AI workloads like mixture-of-expert models;

➂ Compliant with Ultra Ethernet Consortium standards, the switch enables scalable, low-latency networks for AI training/inference and integrates with open ecosystems like OCP's SUE Framework for flexible cluster configurations.

① Apple and Broadcom are collaborating on the "Baltra" data center processor, expected to launch by 2027, featuring proprietary AI accelerators and HBM memory for Apple Intelligence services.

② The processor may scale up to 8x the CPU/GPU cores of the M3 Ultra, though AI workloads typically require specialized tensor units, suggesting a hybrid CPU-NPU design.

③ Apple’s roadmap includes M5 updates for 2025 iPads/MacBooks, M6 (Komodo), M7 (Borneo), and a high-end Sotra chip, alongside custom processors for wearables like AR glasses.

➀ NXP CEO Kurt Sievers will resign on October 28, succeeded by current President Rafael Sotomayor, who has 20 years of experience at NXP and previously held executive roles at Broadcom, Intel, and Motorola;

➁ Sievers, CEO since 2020, led strategic initiatives in automotive and IoT, including the acquisition of Marvell’s Wi-Fi business in 2019;

➂ NXP reported a 9% Q1 revenue decline to $2.84 billion, citing market uncertainty and tariff challenges while maintaining "cautious optimism."

➀ This article lists the top ten semiconductor companies in 2024, based on data from Gartner.

➁ The top five companies are Nvidia, Samsung, Intel, Hynix, and Qualcomm.

➂ The full list includes other notable companies such as Broadcom, Micron, AMD, and MediaTek.

➀ Broadcom has introduced two new 200G/lane DSP PHY designs, Sian3 and Sian2M, to address the growing demand for AI and machine learning cluster connectivity.

➁ Sian3, built on 3nm technology, offers the lowest power consumption for SMF-based 800G and 1.6T modules, improving on its predecessor by reducing power usage by over 20%.

➂ Sian2M is optimized for short-reach MMF links in AI clusters, featuring an integrated VCSEL driver and leveraging Broadcom's 200G VCSEL technology for enhanced performance and efficiency.

➀ TSMC has proposed a joint venture with US chip designers like NVIDIA and AMD to operate Intel's semiconductor fabs, with TSMC holding no more than a 50% stake.

➁ The talks are in early stages, following President Trump's push to revitalize the US semiconductor industry.

➂ TSMC would run the operations of Intel Foundry, but wouldn't own more than 50%.

➀ TSMC and Intel are reportedly still working on a potential joint venture to run Intel's manufacturing capacity;

➁ The joint venture would involve leading American fabless chip designers such as AMD, Broadcom, Nvidia, and Qualcomm;

➂ The initiative is in response to a request from President Trump's administration to strengthen Intel and ensure continued American control.

➀ TSMC is in talks with Intel, Nvidia, Broadcom, Qualcomm, and AMD about a joint venture to acquire Intel Foundry;

➁ TSMC plans to take a stake of no more than 50% in Intel Foundry to maintain US ownership, as required by the White House;

➂ The deal requires approval from the White House.

➀ Broadcom's deal with ByteDance for a custom AI accelerator chip has been canceled.

➁ The cancellation is due to tightened US semiconductor restrictions on China.

➂ Broadcom could face losses of $2B to $3B from the TikTok ban.