🕷️
SetupAI
🏠Stalecollected in 16m

Broadcom Q2 2026 Earnings: AI Chip Revenue Surges 143%

PostLinkedIn
🏠Read original on IT之家
#semiconductor#data-center#financial-resultsbroadcom-ai-semiconductorsbroadcomai-accelerator

💡Broadcom's 143% AI revenue growth is a key bellwether for the hardware infrastructure supporting the AI boom.

⚡ 30-Second TL;DR

What Changed

AI semiconductor revenue hit $10.8 billion, up 143% YoY.

Why It Matters

Broadcom's data confirms the sustained, massive demand for custom AI silicon and high-speed networking infrastructure, signaling continued capital expenditure growth in hyperscale AI data centers.

What To Do Next

Monitor Broadcom's networking roadmap for new Ethernet-based AI fabric solutions to optimize your large-scale model training clusters.

Who should care:Enterprise & Security Teams

Key Points

  • •AI semiconductor revenue hit $10.8 billion, up 143% YoY.
  • •Total revenue reached $221.87 billion, a 47.88% increase.
  • •Semiconductor solutions division grew 79% to $15.09 billion.
  • •Projected Q3 AI semiconductor revenue to exceed $16 billion.

🧠 Deep Insight

Web-grounded analysis with 25 cited sources.

🔑 Enhanced Key Takeaways

  • •Broadcom's significant AI chip revenue growth is primarily driven by its custom Application-Specific Integrated Circuits (ASICs), which are tailored for major hyperscale customers.
  • •The company's key customers for these custom AI accelerators include tech giants like Google (for its Tensor Processing Units or TPUs), Meta (for its MTIA chips), OpenAI, and Anthropic.
  • •Broadcom holds an estimated 70% market share in the custom AI accelerator design market, a dominant position built on deep, multi-year engineering relationships with these hyperscalers.
  • •Beyond custom chips, Broadcom's AI networking portfolio, featuring products like the Tomahawk 6 switches and Jericho 4 routers, is critical for interconnecting massive AI clusters, supporting both scale-up and scale-out architectures.
  • •The company's AI business is backed by a substantial order backlog of approximately $73 billion, and management projects annual AI chip revenue to exceed $100 billion by 2027.
📊 Competitor Analysis▸ Show
CompanyAI Chip FocusNetworking SolutionsKey Partnerships/Customers
BroadcomCustom AI ASICs (XPUs) for hyperscalers (e.g., Google TPUs, Meta MTIA, OpenAI, Anthropic)Tomahawk 6 (102.4 Tbps Ethernet switch), Jericho 4 (Ethernet fabric router for distributed AI)Google, Meta, OpenAI, Anthropic
Marvell TechnologyCustom AI ASICs (e.g., Amazon Trainium, Microsoft Maia)Semiconductors for data infrastructure, networking, storage, and connectivityAmazon, Microsoft, Google (newer TPU exposure)
NVIDIAGeneral-purpose GPUs (e.g., H100, GB200), Spectrum-X Ethernet platformInfiniBand (proprietary), Spectrum-X EthernetBroad range of AI developers and enterprises

🛠️ Technical Deep Dive

  • Custom AI Accelerators (XPUs): Broadcom designs bespoke AI accelerators for hyperscalers, often leveraging advanced packaging technologies like 3.5D XDSiP, which uses face-to-face 3D stacking via TSMC's SoIC process combined with 2.5D CoWoS integration. This platform enables packages exceeding 6,000 mm² of silicon with up to 12 HBM stacks.
  • Tomahawk 6 Ethernet Switch: This is the world's first 102.4 Tbps Ethernet switch, built on 3nm technology. It supports 512 ports of 200GbE or 1,024 ports of 100GbE, enabling flat, low-latency AI/ML clusters. It features advanced adaptive routing, dynamic load balancing, and congestion management for high network utilization and low job completion times (JCT). It is designed for both scale-up (up to 512 XPUs) and scale-out (over 1 million XPUs) AI networks and is compliant with the Ultra Ethernet Consortium (UEC) specifications.
  • Jericho 4 Ethernet Fabric Router: Purpose-built for distributed AI infrastructure, Jericho 4 can interconnect over one million XPUs across multiple data centers, extending AI-scale Ethernet fabrics beyond individual facilities. It features 3.2 Tbps HyperPorts that consolidate four 800GE links, deep buffering based on High-Bandwidth Memory (HBM) for lossless data transmission over distances up to 100 kilometers, and line-rate MACsec for security. It is also manufactured using TSMC's 3nm technology.
  • Jericho 3-AI: An earlier AI-focused networking chip (launched in 2023) offering 28.8 Tbps throughput, supporting up to 32,000 GPUs, and utilizing a scheduled fabric with a leaf-spine configuration and credit-based signaling to minimize congestion and tail latency. It also employs Perfect Load Balancing (PLB) by dividing Ethernet frames into equal-sized cells.

🔮 Future ImplicationsAI analysis grounded in cited sources

Broadcom's dominant position in custom AI ASICs and networking will likely solidify its role as a foundational technology provider for hyperscale AI infrastructure.
Deep, multi-year engineering relationships with major hyperscalers and a comprehensive networking portfolio create high barriers to entry and strong customer lock-in.
The integration of VMware will continue to provide stable, high-margin software revenue, diversifying Broadcom's business model and funding further AI investments.
VMware's subscription-based software offers predictable revenue streams, balancing the potentially cyclical nature of hardware sales and enabling strategic AI growth.
Broadcom's focus on open Ethernet standards for AI networking will accelerate its adoption over proprietary solutions like InfiniBand.
The multi-vendor Ethernet ecosystem, lower costs, and simplified operations offered by Broadcom's solutions are increasingly attractive to hyperscalers building massive AI clusters.

⏳ Timeline

1991
Broadcom Corporation founded by Henry Samueli and Henry Nicholas.
1998
Broadcom goes public on the NASDAQ exchange.
2016
Avago Technologies acquires Broadcom Corporation and rebrands as Broadcom Inc.
2023-11
Broadcom completes its $69 billion acquisition of VMware.
2025-08
Broadcom begins shipping the Jericho 4 Ethernet fabric router for distributed AI infrastructure.
2026-03
Broadcom's Tomahawk 6 switch chip, the industry's first 102.4 Tbps Ethernet part, enters volume production.
🏠Read original article on IT之家
📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

Same topic

Explore #semiconductor

Same product

More on broadcom-ai-semiconductors

Same source

Latest from IT之家

TVB Rebrands to Wireless Group, Shifts to AI-Driven Media

TVB Rebrands to Wireless Group, Shifts to AI-Driven Media

IT之家•Jul 24

AI-curated news aggregator. All content rights belong to original publishers.
Original source: IT之家 ↗