Arista targets Nvidia with 1.6T Ethernet platform for AI supersystems
Arista Networks announced its 7060XE7 Series 1.6T switches, powered by Broadcom's Tomahawk 6 silicon, which deliver 100 Tbps switching capacity. This development is seen as a significant inflection point for Ethernet in high-performance AI networking, challenging InfiniBand's dominance. The new switches emphasize rack-scale integration, liquid cooling, and linear pluggable optics for power efficiency and serviceability in large AI data centers.
Key Takeaways
- 7060XE7 Series delivers 100 Tbps switching capacity using Broadcom Tomahawk 6 silicon.
- Linear Pluggable Optics (LPO) technology reduces interconnect power consumption by approximately 60% versus standard modules.
- Specialized liquid-cooled platform (7060XE7-64PRS-RV3-L) lowers PUE overhead to 5-15% by removing internal fans.
- Two-tier scale-out architecture supports up to 4.5 times more GPUs than standard fixed-box configurations.
- Air-cooled models are scheduled for Q4 2026, with liquid-cooled systems following in Q1 2027.
Why It Matters
The shift to 1.6T Ethernet marks a critical pivot from discrete hardware to integrated 'AI supersystems' that challenge Nvidia's proprietary InfiniBand stronghold. By achieving performance parity with InfiniBand while maintaining open-standard flexibility, Arista is enabling hyperscalers and enterprises to build multi-vendor AI environments. This transition is essential for streaming operators and cloud providers scaling generative AI workloads that now approach one million GPUs, where network uptime and power efficiency directly dictate training ROI. Watch for adoption rates among Tier-2 cloud providers that lack the engineering resources to build custom transport protocols.
Additional Context
The 1.6T launch arrives as Ethernet emerges as the dominant fabric for AI back-end networking. Per Dell’Oro Group in June 2026, Ethernet switch sales in AI clusters more than doubled in Q1 2026, now accounting for approximately two-thirds of the total data center switch market. This represents a significant reversal from 2023, when Nvidia's InfiniBand held an 80% share. Analysts at Piper Sandler noted in January 2026 that Arista’s growth is increasingly tied to the 'Goldilocks zone' of AI spend, with Microsoft, Meta, and Oracle currently accounting for more than 50% of its total business. Competition in the Ethernet space is also intensifying through the Ultra Ethernet Consortium (UEC). The UEC 1.0 specification, finalized in June 2025, aimed to provide the lossless transport and multi-pathing capabilities previously unique to InfiniBand. While Nvidia continues to push its Spectrum-X Ethernet platform as a vertically integrated alternative, Arista's alignment with UEC and Broadcom allows it to capture demand from hyperscalers seeking to de-risk their supply chains and avoid single-vendor lock-in. Investing.com reported in June 2026 that Arista’s revenue grew 31% over the last twelve months, reflecting this accelerated replacement cycle for 800G and 1.6T systems. Technologically, the industry is moving toward 200G-per-lane signaling to meet the bandwidth demands of next-generation accelerators like Nvidia’s Blackwell and AMD’s Instinct series. Per Network World in January 2026, the IEEE 802.3dj standard supporting 1.6T remains on track for finalization by late 2026. This timeline aligns with Arista's projected production volume for its liquid-cooled platforms in early 2027, which are designed to sit directly within the high-density thermal environments required for frontier AI model training.
Read full article at siliconangle.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source