AI Networking and Optics: NVLink, Ethernet, CPO — 2026-09-04
Broadcom reported record AI semiconductor revenue of $16.7 billion in Q3 fiscal 2026, driven by custom XPUs and AI networking, while projecting $21.7 billion for Q4. Meanwhile, Chinese optical transceiver leaders like Innolight and Eoptolink confirmed that 1.6T shipments are ramping up, with orders extending into 2027, signaling a robust scale-out network demand cycle.
AI Networking and Optics: NVLink, Ethernet, CPO — 2026-09-04
Top developments
Broadcom’s AI Revenue Surges to $16.7B Amidst Supply Constraints
In its Q3 fiscal 2026 earnings report released on September 3, Broadcom announced record revenue of $29.6 billion, with its AI semiconductor division contributing $16.7 billion—a 221% year-over-year increase. Management projected Q4 AI revenue to reach $21.7 billion, noting that customer demand for custom XPUs and Ethernet-based AI networking solutions currently exceeds the company’s own $115 billion target for fiscal 2027. This surge underscores the critical role of Broadcom’s Tomahawk and Jericho chips in the scale-out fabric of major AI clusters, although supply chain constraints remain the primary bottleneck rather than lack of orders.
Chinese Optics Makers Confirm 1.6T Ramp and 2027 Order Visibility
Leading Chinese optical transceiver manufacturers, including Innolight (Zhongji Innolight) and Eoptolink, have reported strong H1 2026 results driven by the mass production of 1.6T optical modules. Reports indicate that Innolight saw sequential growth in 1.6T shipments in Q2, boosting gross margins, while Eoptolink’s high-margin 1.6T products continue to ramp up. Digitimes noted on September 3 that these vendors have secured orders extending into 2027, confirming sustained demand for high-speed interconnects required for next-generation AI training clusters. This trend highlights the indispensable role of Asian optics suppliers in the global AI networking supply chain, particularly for Nvidia and hyperscaler platforms.
Optical Circuit Switches Gain Traction for Next-Gen AI Fabrics
A new wave of photonics startups is attracting significant investment to develop optical circuit switches (OCS) that promise lower power and higher density than traditional packet-switched Ethernet fabrics. Companies like iPronics are raising hundreds of millions of dollars to commercialize these technologies, which aim to reduce the energy overhead of moving data between GPUs in massive clusters. While still emerging, OCS technology represents a potential architectural shift away from the current dominance of copper and pluggable optics in scale-up domains, offering a path to more efficient "AI factories".
Local view
Chinese financial media outlets, including Sina Finance and Zhihu, have focused heavily on the H1 2026 earnings of domestic optics leaders like Innolight and Eoptolink, highlighting their role as key beneficiaries of the global AI infrastructure buildout. Analysts note that Innolight’s 1.6T silicon photonics modules now hold a significant market share, with some reports suggesting over 60% in specific segments, solidifying their position as primary suppliers to Nvidia and major cloud providers. There is also growing discussion regarding the potential impact of US regulatory pressures on these suppliers, though industry insiders describe it as a manageable "stress test" given the critical nature of their components.
Context & numbers
- Broadcom AI Revenue: $16.7 billion in Q3 FY2026, up 221% YoY; guided to $21.7 billion in Q4 FY2026.
- Optical Transceiver Market: Projected to reach $55 billion by 2030, driven by 800G and 1.6T adoption.
- 1.6T Shipments: Global 1.6T optical module shipments are expected to reach 5 million units in 2026, with Nvidia alone procuring over 300,000 units in 2025.
- Ethernet vs. InfiniBand: Ethernet continues to dominate AI networking switch sales, though InfiniBand has seen a resurgence in specific high-performance compute segments.
On the radar
- Broadcom Stock Volatility: Despite record AI revenue, Broadcom shares fell 6% following soft overall guidance, highlighting investor sensitivity to non-AI segment performance.
- 1.6T Ecosystem Maturity: The ramp of 1.6T modules is tightly coupled with the maturity of the 200G/lane SerDes ecosystem; delays in this area could impact the pace of scale-out network upgrades.
- Supply Chain Constraints: Broadcom CEO Hock Tan explicitly stated that supply chain constraints, not demand, are the limiting factor for AI chip shipments, a trend likely to persist through late 2026.
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
