AI Networking and Optics: NVLink, Ethernet, CPO — October 8, 2026
Arista Networks has launched a new rack-scale Ethernet portfolio, positioning open standards as a direct alternative to Nvidia’s proprietary NVLink for AI scale-up fabrics. Meanwhile, Nvidia is accelerating its co-packaged optics (CPO) roadmap with Lumentum lasers powering next-gen switches, and CScale raised $145M to develop optical networking chips targeting 2028 shipments.
AI Networking and Optics: NVLink, Ethernet, CPO — October 8, 2026
Top developments
Arista Enters the Scale-Up Ring with Rack-Scale Ethernet
On October 7, 2026, Arista Networks announced a new portfolio of liquid-cooled, Ethernet-based scale-up switches designed to replace proprietary interconnects like Nvidia’s NVLink within the rack. This move positions Ethernet not just for scale-out clusters but as the "backplane" for tightly coupled AI accelerators, challenging the dominance of closed ecosystems. The initiative includes integration partnerships aimed at supporting diverse accelerator landscapes beyond Nvidia GPUs.
Nvidia’s Co-Packaged Optics Move from Roadmap to Revenue
Nvidia is transitioning its co-packaged optics (CPO) technology from conceptual roadmap to commercial revenue, with Lumentum supplying the critical laser sources for its Spectrum-X and Quantum-X switch lines. This shift addresses the power and density bottlenecks of traditional pluggable optics in AI factories, where electrical switching limits are becoming acute as compute grows by 2.5x annually.
CScale Secures $145M for Optical Interconnect Chips
AI networking startup CScale raised $145 million in a funding round backed by Nvidia and Intel Capital, bringing its total funding to $188 million. The company is developing optical networking chips targeted for shipment in 2028, aiming to provide high-bandwidth, low-latency alternatives to current copper-based scale-up interconnects.
US Considers Restrictions on Chinese 3.2T Transceivers
Reports indicate the US government is considering export restrictions on future generations of Chinese optical transceivers, potentially starting with 3.2T modules. This regulatory signal impacts the global supply chain, as Chinese manufacturers like Innolight (Zhongji InnoLight) and Eoptolink dominate the high-speed module market. Analysts suggest the primary impact may be felt in 2028, with potential benefits for non-Chinese suppliers or those with diversified manufacturing bases.
Local view
Sina Finance and Sohu highlight the surge in performance of Chinese optical module leaders amid the AI boom. Zhongji InnoLight reported H1 2026 revenue of 41.778 billion CNY (+182.49% YoY) and net profit of 13.651 billion CNY (+241.70% YoY), driven by strong demand in silicon photonics and CPO. Eoptolink also saw significant growth with H1 revenue of 20.910 billion CNY (+100.34% YoY). Industry reports note that while 1.6T modules are ramping up, the market is already looking toward 3.2T/6.4T NPO/CPO solutions, with silicon photonics penetration expected to rise.
Context & numbers
- Ethernet Switch Growth: Ethernet switch sales for AI networking have doubled, with Ethernet becoming the dominant fabric as 800G ramps and 1.6T takes form.
- 1.6T Ramp: Sample shipments of 1.6Tbps transceivers have begun, with mass ramping projected for the second half of 2026.
- Broadcom Revenue: Broadcom’s AI-related revenue is growing at 221%, driven by custom ASICs and networking chips like Tomahawk.
- Nvidia Share Shift: Nvidia’s share of the Ethernet switch market rose from under 4% to 21.5% in one year, reflecting its aggressive push into open-standard networking via Spectrum-X.
On the radar
- OFC 2026 Insights: Recent workshops at OFC 2026 highlighted that optical circuit switching (OCS) is becoming a core technology for hyperscalers and Nvidia to reduce latency and power in next-gen supercomputers.
- Arista Guidance: Investors are closely watching Arista’s raised revenue guidance for fiscal 2026, which reflects continued fast growth in AI data center connectivity.
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
