Arista Networks' latest quarter shows AI demand lifting a networking vendor well before the most contested part of the AI fabric has opened up. According to The Next Platform, the company reported September-quarter product sales of $3.19 billion, up 66.9 percent year on year. Services revenue grew 17.9 percent to $467.6 million.
Total revenue came to $3.66 billion, up 58.5 percent, with sequential growth of 20.5 percent. Operating income reached $1.64 billion and net income $1.56 billion, equal to 42.7 percent of revenue. The source describes that as the highest profit level the company has recorded. Arista ended the quarter with $14.36 billion in cash and equivalents, $11.4 billion in customer purchase commitments and $6.84 billion in deferred revenues.
Current AI revenue is coming from front-end networks, which connect clusters to the outside world and storage, and from R series switches with routing functions used for scale-across links between datacenters and regions. Power limits in a given area are one reason operators spread clusters across sites. CEO Jayshree Ullal said Arista expects roughly $3.5 billion in AI front-end and back-end networking sales in 2026, doubling to about $7 billion in 2027.
Scale-up networking, the fabric inside a rack or row, is not yet a meaningful revenue source. Ullal said Arista has worked with AMD, Arm, Meta Platforms, Microsoft and Qualcomm on rackscale switching. She expects ESUN-based trials to begin in 2027, starting with racks of 128 or 144 accelerators and reaching as many as 1,000 across a row. Production rollouts are not expected until 2028.
The source notes that Arista has not committed to UALink switching, though third-party UALink silicon exists from Astera Labs, Marvell, Upscale AI and possibly Panmnesia. It also says Arista has no play in networking within accelerator packages. The broader argument is that open Ethernet-based approaches will have to challenge Nvidia's NVSwitch and NVLink, and that this transition will take time.
Why it matters: Arista's results suggest that front-end and scale-across networks are currently carrying AI-related spending, while the scale-up market remains a 2027 to 2028 question. Operators and investors should treat the timing as the company's stated expectation, not a settled outcome.
TwinStack automated analysis, based on reporting by The Next Platform. Check the original for full details.