Arista's deferred revenue is $6.87 billion and growing 69% a year. That number is doing a lot of the heavy lifting for the bull case.

Arista sells Ethernet switches and networking software for data centers, with Microsoft and Meta as its two largest customers, together representing 26% and 16% of 2025 revenue respectively. The pitch is that as hyperscalers build AI clusters, they need networking fabric that can move data between GPUs without becoming the bottleneck, and Arista's Etherlink platform has become one of the standard choices for that job, competing against proprietary alternatives like InfiniBand.

Q2 2026 was genuinely a record: $3.04 billion in revenue, the first time crossing $3 billion in a quarter, up 37.7% year over year and beating the $2.8 billion guidance. Non-GAAP EPS came in at $1.02, up 39.7%, and this marks six straight quarters of EPS beats. Operating margin sat around 49.9% on a non-GAAP basis, which is an extraordinary margin for a hardware-adjacent business. Management raised full-year 2026 revenue guidance to $12.6 billion, the third raise this year, implying 40% annual growth.

The number that stands out most to me is deferred revenue, which reached $6.87 billion, up 69% year over year. Deferred revenue represents money customers have already committed or paid that hasn't been recognized as revenue yet, so growth there effectively de-risks a meaningful chunk of future quarters before a single new switch ships. Combined with $9.7 billion in purchase commitments securing supply through 2027, the forward visibility here is real, not just a promise on an earnings call.

The honest risk is the same concentration story that shows up across this whole AI networking category. Fewer than five customers make up Arista's Cloud and AI Titans category, its largest and most strategically important segment. Management has said they expect one, maybe two more 10%-plus customers this year, widely speculated to be Google and Oracle, which would help diversify things somewhat, but Microsoft and Meta remaining "at or near" the 10% threshold means the core dependency doesn't go away. The stock trades around 54-66x forward earnings depending on the exact multiple used, and at least one fair value model puts intrinsic value around $190, implying downside from current levels if growth decelerates even modestly from here.

[Arista](https://www.stoxcraft.com/stocks/anet) has genuinely excellent unit economics and real forward visibility through deferred revenue and purchase commitments. The multiple still assumes that visibility keeps converting into growth at this pace for years. Anyone comparing this to the other AI networking names like Credo or Coherent?