ANET Price History, 6M

ANET Multi-model Signals, 6M

ANET is up 43% YTD, and more than 80% from its March low. This is a fantastic performance for any stock, but shy of the parabolic runs seen in the Semis and Memories world. At the same time, it has held firm when other AI trades recently cratered, making it an important differentiator in one’s portfolio.

Looking at our multi-model signals, the ML model is currently risk-off. The ML model skipped the late-May rally, but correctly dodged the late-June and mid-July declines. It turned flat more recently after the earnings, which so far hasn’t been proven incorrect.

However, what keeps the MS position long is that ANET is in a firm uptrend (Trend model green) and our Regime model signals a favourable risk-on environment. The neural signal also likes the stock, although it doesn’t vote in the actual long-vs-flat decision. (This is a deliberate choice, but may change in the future.)

What could take ANET out of our portfolio? The most likely source is if the neural model also turns risk-off, making the overall consensus 2/4 (currently 3/4). The MS position would stay long, but the consensus would fail to meet our Systematic Portfolio requirement of 3+ model agreements.

Arista designs and sells high-speed Ethernet switching and routing systems for data centres, AI back-end fabrics, campus and WAN. Revenue is generated by selling switching and routing systems and then attaching software subscriptions and post-contract support.

The hardware is built on merchant silicon bought from Broadcom rather than designed in-house, and assembled by contract manufacturers.

The differentiation lies in EOS, a single operating system image that runs unmodified across the entire product line, and in CloudVision, the management and telemetry layer sold on a subscription basis.

You can therefore think of Arista’s moat as software wrapped around commodity hardware.

Arista benefits from secular-growth demand that is overlaid on a very cyclical customer capex budget.

Position in the AI Value Chain

Arista sits between merchant switch silicon above it and the hyperscalers below it.

Upstream, Broadcom (AVGO) supplies the Tomahawk and Jericho chips that set both the performance ceiling and the cost floor, with Marvell (MRVL) and Credo (CRDO) supplying the retimers and interconnect chips (parts that clean and boost data signals so they survive the trip across a rack).

Optics and fibre come from Coherent, Ciena (CIEN) and Corning (GLW). The boxes are built by contract manufacturers such as Jabil and Celestica.

Downstream, the customer list is short but rich: Microsoft (MSFT) and Meta (META) were the two 10%-plus end customers in 2025, at 26% and 16% of revenue, respectively. There are also Oracle (ORCL), Alphabet (GOOGL) and the neoclouds like CoreWeave (CRWV).

The closest listed rivals are Cisco (CSCO), NVIDIA (NVDA) through its Spectrum-X Ethernet line, and Juniper inside HPE.

Arista enjoyed its first-ever $3 billion quarter, rising by 38% YoY. Multiyear purchase commitments rose to approximately $9.7 billion at the end of Q2 from $3.6 billion a year ago, i.e. a near-tripling YoY. Etherlink AI fabric customer count now exceeds 100 cumulative customers, compared with the single-digit count back in 2024.

The entire quarter has therefore been positioned to signal that Arista is in a supply-constrained position vs customer demand. Management believes the industry-wide supply problem will persist until 2028, meaning it is a two-year problem.

The FY2026 revenue guidance was raised for the third time this year to approximately $12.6 billion (40% growth). Interestingly, management refused to provide an exact breakdown of where the raised revenue would come from, only pointing to the direction of the core data centre front end plus enterprise and routing adjacencies.

The gross margin improvement QoQ was mostly driven by tariff refunds. Because of the rising memory and silicon costs, gross margins would be stuck in the 62% and 64% range this year. The existing backlog means that the price pass-through (i.e. raising product pricing) is only available next year.

I am not concerned about FCF as it seemed to be a cashflow timing issue for Q1 vs Q2.

Arista’s EPS grew by 35% in the quarter, on pace with the revenue growth.

However, Arista’s SBC is growing slightly faster than the revenue, which is a real dilution cost for shareholders. It signals that human capital in this space is also expensive.

Arista sits on a tonne of cash - at $13.3 billion in the quarter. But they didn’t repurchase any shares in the quarter, leaving more than half of their $1.5 billion authorisation from May 2025 available.

I wonder if management implicitly acknowledges that its share price is not “cheap” enough to justify, especially as we had a broader AI momentum run at the beginning of Q2.

Here are the key financial metrics from the past four quarters. I designed the table this way so we can focus solely on the key top-line, profit, and cash flow metrics.

Here are Wall Street's forward estimates.

Overall, Arista is expected to grow by 38% in the next twelve months. Its EPS is expected to grow at a similar pace.

It has just posted its first $3 billion quarter and raised full-year guidance for the third time this year, so the business itself is solid. The question is the price investors pay.

At 43.6x forward earnings, the stock assumes several more years of hyperscaler acceleration. The two very large end customers, Microsoft and Meta, have to deliver it.

Base case

AI orders keep converting through 2027 and the $6.9bn deferred revenue balance unwinds into sales. The forward multiple fades from 43.6x toward 40x as growth normalises from 40% YoY to the high 20s.

Holding 40x on $4.80 FY2027 earnings gives about $192, roughly flat against $191.52 today. This is a net result of earnings growth meeting multiples compression.

Bull case

The one or two new 10%-plus customers management flagged on the earnings call actually land, and 1.6T Etherlink moves from trial to production on schedule in 2027. FY2027 earnings beat to roughly $5.40 and the multiple holds 45x, giving about $243, +27% against $191.52 today. This means earnings growth almost entirely passes through to the stock price.

Bear case

The 2019 pattern repeats, slowing ANET down next year. Revenue grew 12.1% YoY in 2019 and then fell 3.9% YoY in 2020 when one cloud customer changed its procurement plan. A repeat takes FY2027 earnings to roughly $4.20 and compresses the multiple to 28x, giving about $118, -39% from the same starting point.

Arista’s valuation today is near its 3-year high point. We have seen mean reversion at these levels before. However, Arista is also growing at the fastest pace in the past 3 years, with the market willing to bid higher for AI names.

Arista has very small D&A and a lot of cash, making EV/ EBIT a reasonable multiple for price targets (e.g. compared with P/E).

If we rein it in below the +1 sigma of its past 3-year valuation mean, which is still a slight premium for its growth, 40x NTM EV/EBIT is a reasonable target. That would suggest a price of $235, which is 22% above the latest price.

The DCF method is a bit more pessimistic on ANET in terms of valuation. To get to today’s price, I need to assume an above-target bullish case of 20-30% growth by 2032 (and a slowdown to a 3% terminal growth thereafter), with the EBIT margin holding at 50%. However, I have assumed a 5% equity premium over the 4.65% risk free rate of today, which does keep the cost of capital high (11%) even assuming a 20% debt mix in the long run.

No one actually trades on DCF models.

But all of the above is to say that the market is already bullishly positioned for Arista’s strong growth. This means that the supply constraint that management repeatedly highlights must persist. They also need to, of course, execute well enough to retain the large customers, who must not prematurely scale back their expected CapEx commitments.

Investors, at today’s valuation, are paying for a sustained AI buildout / CapEx cycle, that does not peak until at least the end of 2027. Arista is back to hypergrowth (nearly 40%), and investors need to see it sustain its execution and market position during this AI pick-and-shovel era.

Investors also need to see Arista onboarding new large customers to reduce the dependency on Microsoft and Meta. My concern is that Meta, after its recent AI ROIC scrutiny, could actually put a brake on its buildout, not least given its lack of a long-term strategy for the AI race.

Since COVID, ANET has seen three separate rallies, separated by the 2022 bear market and the 2025 Liberation Day crash.

R2 (2022-2024) ended at the 125% extension of R1 (2020-2021), in percentage terms.

If today’s R3 run replicates the rally of R1, ANET can target $220, which it almost did post-earnings. However, if R3 replicates the strength of R2, it could aim $300. As you have seen from the above valuation analysis, that is a pretty stretched target. It likely needs several more quarters for ANET to continue its growth acceleration and a friendly macro environment with lower long-end yields.

Nonetheless, the overall trend remains very positive for ANET, so there is no reason to fade the stock on the downside at this point.

Arista is a strong business that is performing handsomely. I would prefer a cheaper entry from the technicals and valuation perspective, but it’s definitely not a stock to short. Therefore, I am happily following the Systematic Portfolio inclusion, keeping an eye on the risks in case the tide turns.