AI Networking and Optics: NVLink, Ethernet, CPO — 2026-09-11
Broadcom reported a record-breaking fiscal Q3 2026, with AI semiconductor revenue tripling to $16.7 billion and raising FY2027 AI revenue guidance to $115 billion. Meanwhile, Arista Networks highlighted sustained AI demand at the Goldman Sachs conference, reinforcing the "Ethernet vs. InfiniBand" shift as Ethernet switch sales double while 1.6T transceiver shipments begin to ramp.
AI Networking and Optics: NVLink, Ethernet, CPO — 2026-09-11
Broadcom Q3 FY2026: AI Revenue Triples, Guidance Raised
In its Q3 2026 earnings call on September 9, Broadcom (AVGO) announced that AI semiconductor revenue more than tripled year-over-year to $16.7 billion. The company raised its forecast for AI chip revenue in the fiscal year ending October 2027 to approximately $115 billion, up from prior estimates, and projected a long-term AI revenue target of $230 billion by 2028. This surge underscores the intense hyperscaler demand for custom AI accelerators (XPUs) and networking ASICs, positioning Broadcom’s Tomahawk and Jericho Ethernet platforms as critical infrastructure alongside Nvidia’s offerings.
Arista Networks Signals Sustained AI Growth at Goldman Sachs
At the Goldman Sachs conference on September 9, Arista Networks executives reiterated that AI demand remains the primary driver of growth. The company emphasized its strong position in high-speed Ethernet switching for AI clusters, noting that hyperscalers are increasingly adopting Ethernet-based fabrics for scale-out networks. This commentary aligns with broader market trends where Ethernet switch sales have doubled, challenging InfiniBand’s dominance in certain AI training workloads due to cost-effectiveness and ecosystem maturity.
Nvidia Reaffirms Full-Stack Infrastructure Strategy
On September 7, reports highlighted Nvidia’s strategic pivot from being solely a chip supplier to an infrastructure platform provider. Jensen Huang described the vision as a "full-stack AI factory," integrating NVLink for scale-up, InfiniBand and Spectrum-X Ethernet for scale-out, and co-packaged optics (CPO). This approach aims to capture more value across the networking layer, with NVLink Fusion and Spectrum-X Photonics expected to drive significant revenue as customers seek integrated, optimized solutions for million-GPU clusters.
Local view
No recent local-language media coverage specific to this week's networking developments was found in the provided data.
Context & numbers
- Broadcom AI Revenue: Q3 FY2026 AI semiconductor revenue reached $16.7 billion, tripling year-over-year.
- FY2027 Guidance: Broadcom now forecasts ~$115 billion in AI chip revenue for FY2027.
- Ethernet vs. InfiniBand: Market analysis indicates Ethernet switch sales have doubled, gaining significant share in AI networking, though InfiniBand remains resilient in high-performance training segments.
- 1.6T Transceivers: Sample shipments of 1.6 Tbps transceivers have begun, with volume ramping projected for the second half of 2026.
On the radar
- Broadcom Q4 Guidance: Analysts are closely watching Q4 guidance for potential adjustments given the mixed consensus versus actual performance mentioned in recent Futurum Group analysis.
- 1.6T Ramp-Up: The industry awaits concrete shipment data for 1.6T optical transceivers in late 2026, which will serve as a key indicator of next-generation AI cluster deployment speeds.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
