Astera Labs and Arista Networks Scale as AI Networking Bottlenecks Shift
Record revenues and expanding customer bases signal a broader industry transition toward Ethernet-centric AI fabrics.
Astera Labs and Arista Networks are seeing rapid growth as the expansion of AI infrastructure shifts the industry's primary bottleneck from raw compute to the networking fabric. This transition is driving a surge in demand for high-speed connectivity solutions required to link massive accelerator clusters.
Astera Labs reported record Q2 2026 revenues of $392.4 million, representing a 104% increase year-over-year. The company also projects that future Scorpio X-Series content will exceed $1,000 per XPU. Simultaneously, Arista Networks has expanded its reach in the AI sector, surpassing 100 cumulative AI fabric customers using its Etherlink switches during the second quarter of 2026. Arista expects its AI-related revenue to reach at least $3.6 billion by the end of 2026.
The Shift to Ethernet-Centric Architecture
This growth occurs as AI infrastructure evolves toward faster links and more complex rack-scale topologies. To support larger accelerator clusters, the industry is moving away from legacy setups toward Ethernet-centric AI networking. This shift necessitates the adoption of 1.6-terabit Ethernet platforms and advanced connectivity tools that can handle the massive data throughput required by next-generation models.
Why the 'Plumbing' Matters
As AI models grow in complexity, the ability to move data between GPUs and XPUs becomes as critical as the processing power of the chips themselves. Companies providing the essential "plumbing" of AI—including switches, retimers, and software-defined networking—have become critical infrastructure plays. The rapid scaling of these technologies indicates that hyperscale data centers have entered a new phase of expansion, where low-latency connectivity is the primary requirement for maintaining performance gains.
Future Outlook
Market observers are now watching for the continued adoption of high-speed standards and the further penetration of Ethernet-based fabrics into the AI space. While the current trajectory for Astera Labs and Arista Networks is steep, the long-term ceiling depends on how quickly hyperscalers can deploy these complex topologies at scale to meet the demands of increasingly large-scale AI training and inference workloads. This evolution suggests that the networking layer is no longer a supporting component but a primary driver of AI cluster efficiency.