Arista Networks (NYSE: ANET) announced on 7 October 2026 an open Ethernet rack-scale portfolio for scale-up and scale-out AI infrastructure. Its new Etherlink SU-144 architecture links up to 144 accelerators in a single-hop network within a rack and scales to 1,024 across racks, the company says.
Arista says the platforms were engineered in partnership with AMD, Arm, Broadcom, d-Matrix, Meta, Microsoft and Qualcomm, and use the Ethernet for Scale-up Networks (ESUN) initiative. It describes them as an open, standards-based alternative to proprietary compute islands.
Business Pill · WHAT IS INSIDE WHAT
A one-minute explainer of embedded dependency: when one supplier’s part sits inside another product, the buyer depends on both, often without seeing it. It teaches the general idea only and says nothing about any company in this story.
The key insight: Arista is selling the rack, not just the switch. Its release describes liquid-cooled racks that bundle switches, optics, cooling and rack control, and pitches Ethernet as the link between accelerators inside the rack, the job it says proprietary compute islands do today. As we read it, the partner list is the argument: chip designers and two hyperscalers lined up behind an open standard.
For scale-up, Arista describes three physical form factors: an orthogonal chassis, a single-hop single-rack design with 100 to 400 kW of liquid cooling support; a cabled backplane; and a cross-rack system that works as a modular standalone fabric across multiple racks.
For scale-out, it says the rack-scale systems support several scale-up protocols, including ESUN, Nvidia’s NVLink, NVLink Fusion and PCIe, and run either Arista EOS or an open network operating system. The racks use liquid-cooled 7060EX7 switches with 102.4 Tbps of capacity each.
Arista says next-generation optics such as XPO and open CPO/NPO raise capacity from the 1.6 Pb/s per rack of today’s OSFP-based switches to 6.5 Pb/s per rack, which it says translates to nearly a 50% reduction in overall data center footprint.
Per-rack bandwidth as Arista states it in its release of 7 October 2026: 1.6 Pb/s with today’s OSFP-based switches and 6.5 Pb/s with next-generation optics such as XPO and open CPO/NPO. Arista says this translates to nearly a 50% reduction in data center footprint. Company figures.
What Arista Says It Is For
“The network in the AI era is no longer just an independent pipe between servers. It’s now a tightly integrated rack-scale backplane that is essential to maximizing compute,” said Tyson Lamoreaux, Senior Vice President, Cloud and AI Networking at Arista.
He said the aim is to let cloud operators and enterprises scale heterogeneous compute and optimize their token economics “without proprietary lock-in.” Arista says it will demonstrate the systems at the OCP Global Summit, 12 to 15 October 2026.
The Structural Read
It hedges on the scale-out side. Arista says its scale-out racks support ESUN, Nvidia’s NVLink, NVLink Fusion or PCIe, so a buyer can keep Nvidia’s interconnect inside the rack and use Arista between racks.
Density is the economic claim. Arista says next-generation optics lift per-rack capacity from 1.6 Pb/s to 6.5 Pb/s, which it says translates to nearly a 50% reduction in data center footprint.
The release is a design announcement. In what we read there is no price, no ship date and no named customer; Arista says it will demonstrate the systems at the OCP Global Summit on 12 to 15 October.
Tyson Lamoreaux, SVP Cloud and AI Networking, Arista Networks
“The network in the AI era is no longer just an independent pipe between servers.”
Three Implications
ETHERNET MOVES INTO THE RACK Arista says Etherlink SU-144 links up to 144 accelerators in one hop per rack and 1,024 across racks.
OPEN, BUT COMPATIBLE Arista says the scale-out racks also support Nvidia’s NVLink and NVLink Fusion, alongside ESUN and PCIe.
WATCH FOR PRICE AND SHIP DATES The release we read gives neither; the next public showing is the OCP Global Summit, 12 to 15 October.
The framework’s starting point: “This entire ecosystem has been shaped by constraints, bottlenecks, chokepoints, and scarcity at every layer.”
Arista’s launch targets one of those layers: the link between accelerators inside a rack, which it describes as proprietary compute islands today. As we read it, an open Ethernet alternative is an attempt to make that layer less scarce.
What Is Not Established
We read Arista’s release in full. In the release we read, there is no pricing, no shipping date and no named customer, and the partner quotes it links were not part of the text we read. We did not contact Arista or its partners.
The bandwidth, density and footprint figures are Arista’s own, and the release carries a forward-looking statements caution.
Business Engineer Framework
The Map of AI — Where Networking Sits
The Map of AI places more than 200 companies across nine layers of the stack, from silicon to application. It shows where networking vendors sit between the chipmakers and the hyperscalers that build AI clusters.
Arista says it is taking Ethernet into the rack-scale scale-up domain, with Etherlink SU-144 linking up to 144 accelerators in one hop per rack and 1,024 across racks, built with AMD, Arm, Broadcom, d-Matrix, Meta, Microsoft and Qualcomm, and with scale-out racks that also support Nvidia’s NVLink. The release we read gives no price or ship date.
A note on sourcing. We read Arista’s release of 7 October 2026 in full. The bandwidth, density and footprint figures are the company’s own. We did not read the linked partner quotes or contact Arista or its partners. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
