
Last 4 earnings calls · through Q2 FY2026
The bull case is that demand for AI and cloud networks exceeds what Arista can build, so growth is set by supply rather than by customer interest.
Management has raised its full-year revenue outlook at each of the last three reports and says the increases come from several product lines, not one.
Arista runs a single operating system across front-end, back-end and campus networks, which management says wins business that commodity hardware cannot serve.
High cash generation lets Arista commit to multiyear chip and memory purchases ahead of demand without borrowing or diluting shareholders.
The bear case is that component shortages rather than customer orders determine revenue, and management says the shortfall may persist for years.
Larger cloud and AI customers buy at lower prices, and price increases have not yet reached the existing backlog to offset rising memory costs.
Two cloud customers supply a large share of sales, so their buildout pace and architecture choices drive results more than any other factor.
Arista commits to components and holds inventory well before customers accept delivery, which ties up cash and risks write-downs if projects slip.
In short: AI demand outruns component supply, so guidance keeps rising while shipments run late.
Management has raised its AI goal repeatedly, most recently to at least $3.6 billion, and says scale-across routing will be about 30% of it, so the number rests on more than one use case.
Nine months ago management guided 2026 to about 20% growth; it now expects about $12.6 billion, and says the increase spreads across AI, front-end data center, campus and routing.
Management says demand exceeds supply for wafers, chips, optics and memory, and that the shortage may last until 2028, so reported growth reflects what it can procure rather than what customers order.
Deferred product revenue keeps building because AI equipment must be installed and signed off before it is billed, so a shipment and the revenue it produces can land in different years.
In short: Cash generation and operating margin are strong, while cloud mix and memory costs lower gross margin.
That cash is what funds multiyear chip and memory commitments and the inventory build ahead of AI deployments, without any meaningful debt on the balance sheet.
Management once said it was paying almost any price for memory and that its commitments were not enough; it now says memory is contracted for the year with visibility well into next year.
Cloud and AI titans pay lower prices than enterprises, and the June quarter was also helped by one-time tariff refunds that management put at 20 to 30 basis points.
Management first said it would raise prices on memory-heavy products this year; it now says existing backlog pricing means the benefit lands late in 2026 or in 2027.
In short: Arista leads high-speed Ethernet switching, while a few cloud customers supply much of revenue.
It also sells both front-end and back-end AI networks under one operating system, which management argues white box vendors and single-stack rivals like NVIDIA cannot match.
Arista helped write the Ultra Ethernet and scale-up Ethernet specifications and says its whole switch line already complies, which pushes AI back-end buying toward standard Ethernet gear it sells.
Microsoft and Meta have each supplied a large share of sales for years, so a change in their network architecture or a shift toward buying elsewhere moves the whole company.
Management accepts that commodity hardware is good enough for basic use cases and that its own lower-margin blue box line ships without the full EOS software, which limits profit on the largest racks.