I paused while looking at a single line regarding deferred revenue in an Arista Networks (ANET) disclosure. It had increased by 691 TP3T in just one year, accumulating to $6.8 billion. Looking solely at the order flow, it couldn't be better, but my conclusion is a Hold, with a 12-month baseline target price of $208. I will explain why the order volume and the target price have diverged, even opening up the client list to clarify.
The stock prices in this article are based on the US market close on September 15. The Korean Won amounts were calculated using a standard exchange rate of 1,345.6 KRW per dollar on the same day.
Table of Contents
- Arista Networks (ANET) 800G Order: How Far Have We Come in Numbers?
- Q2 2026 results and the third guidance upgrade
- Arista (ANET) Customer Concentration: Two locations account for 421 TP3 T of sales.
- Arista Networks (ANET) Valuation — Is a forward P/E of 37x expensive?
- Arista (ANET) side-by-side with competitors
- Arista Networks (ANET) Target Price Scenarios: $208, $258, $140
- Arista (ANET) Risks and Four Bear Triggers
- What I particularly look at
- Arista Networks (ANET) Investment Opinion Conclusion
- Frequently Asked Questions
- References
Arista Networks (ANET) 800G Order: How Far Have We Come in Numbers?
I will start with three lines of conclusion. First, Arista Networks (ANET)’s order-winning power is confirmed by the numbers. Deferred revenue at the end of the second quarter of 2026 was $6.8659 billion (approximately 9.2388 trillion KRW), an increase of 69.01 TP3T from a year ago.
Second, the customer base generating those orders is actually narrowing. As of 2025, Microsoft accounted for 261 TP3T of revenue and Meta for 161 TP3T, bringing the combined total to 421 TP3T. This is an increase from 351 TP3T a year ago.
Third, the 12-month baseline target price is $208 (approximately 279,885 KRW), and the investment rating is Hold (Medium confidence). This value was calculated based on an appropriate forward P/E of 33 times.
Let me start by summarizing what kind of company Arista is in a single sentence. It is a company that manufactures high-speed Ethernet switches connecting thousands of GPUs (Graphics Processing Units, semiconductors used for AI learning and inference) inside AI data centers. I view this business by likening it to highway tollgates. It is a structure where more tollgates are needed as the volume of vehicles (data) increases; however, the problem is that there are effectively only two construction companies building these highways. I will discuss this further below.
The company does not separately disclose how many orders have been received. So, regarding the balance sheet, Deferred salesI will look at that instead. Deferred revenue is the amount that could not be recorded as revenue because money has already been received from customers but equipment or services have not yet been fully delivered. Simply put, you can think of it as the size of "orders received." I personally created a table of the five quarters and placed them side by side.
| Item (Unit: million dollars) | 2Q2 of 2025 | Q3 2025 | Q4 2025 | Q1 of 2026 | 2Q of 26 |
|---|---|---|---|---|---|
| sales | 2,204.8 | 2,308.3 | 2,487.8 | 2,709.0 | 3,035.7 |
| Gross profit margin | 65.2% | 64.6% | 62.9% | 61.9% | 62.9% |
| Operating Profit Margin (GAAP) | 44.7% | 42.4% | 41.5% | 42.7% | 45.4% |
| Total deferred sales | 4,061.7 | 4,686.2 | 5,372.4 | 6,198.7 | 6,865.9 |
| inventory | 2,059.1 | 2,155.7 | 2,247.1 | 2,380.0 | 2,535.3 |
The line I paid the most attention to in this table is deferred revenue. Usually, sales grow faster than deferred revenue. Since there are goods to sell, they keep clearing out their backlog of orders. However, it is the opposite for Arista. The rate at which orders are accumulating is outpacing the rate at which sales are taking place. Deferred revenue has risen to a scale equivalent to 651 TP3T of the $10.54 billion (approximately 14.1837 trillion KRW) in sales over the past 12 months.
You also need to look at the inventory side. Inventory increased by 23.11 TP3T to $2.5353 billion (approximately 3.4115 trillion KRW), while sales increased by 37.71 TP3T. This means that goods are selling faster than they are accumulating in the warehouse. I interpreted this combination as, “Demand was real, and now we are in a phase where the manufacturers are catching up.” Suppliers of parts The fact that interconnect companies like Amphenol (APH) also performed well.I see it as another side of the same picture.
Let me summarize the progress on the product side. The cumulative number of customers for the Etherlink family of AI fabric products has exceeded 100. Considering there were only four or five in early 2024, this represents a twentyfold increase in just two and a half years. Furthermore, with the addition of the new 1.6-terabit 7060X-E7, the product lineup has expanded to include rack-level AI configurations.
Q2 2026 results and the third guidance upgrade
The second-quarter results released on August 4 marked the first time in the company's history that quarterly revenue exceeded $3 billion. Revenue stood at $3.0357 billion (approximately 4.0848 trillion KRW), an increase of 37.71 TP3T from a year ago and significantly surpassing the company's guidance of $2.8 billion. Non-GAAP diluted earnings per share rose 39.71 TP3T to $1.02. Non-GAAP figures are adjusted by the company to "exclude one-time costs and focus only on core performance.".
The operating profit margin was particularly impressive. It came out at 45.41 TP3T on a GAAP basis and 49.91 TP3T on a non-GAAP basis. This means that even after deducting labor costs and R&D expenses from 100 won in revenue, there is still a profit of over 45 won. It is rare for a hardware company to achieve such a profit margin. Original company press releaseThe figures for each segment are also provided, so you might want to take a look at them as well.
The guidance released alongside it garnered even more attention. They raised their 2026 annual revenue forecast to $12.6 billion (approximately 17 trillion won), and this year Third upwardIt was. The annual growth rate is 401 TP3T. The Q3 guidance was revenue of approximately $3.3 billion, a gross profit margin of approximately 631 TP3T, an operating profit margin of 48–491 TP3T, and diluted earnings per share of $1.06–$1.08.
There is a point here where I paused for quite a while. The basis for this upward revision is not that new demand has increased, but Because the supply has been releasedThat is the case. Management has been talking about parts bottlenecks since the fall of 2025, and now that they have been resolved, they are able to sell the volume they previously couldn't sell. This is welcome news because it implies that demand existed all along; however, looking at it from the other side, it also means that the room for the next upward movement is limited to the amount of the remaining supply cleared.
The stock price reaction was exactly that much. The day after earnings, the stock price rose to $214.89, hitting an all-time high, but the closing price on September 15 was $192.84 (approximately 259,486 KRW). This is 10.31 TP3T down from the peak. You can view this as the market buying on the third upward movement and then retracing a significant portion of it.
Arista (ANET) Customer Concentration: Two locations account for 421 TP3 T of sales.
This is the point I most wanted to make in this post. You need to look at who is standing behind a good order.
| year | Microsoft | Meta | Total of two places |
|---|---|---|---|
| 2022 | 16% | 26% | 42% |
| 2023 | 18% | 21% | 39% |
| 2024 | 20% | 15% | 35% |
| 2025 | 26% | 16% | 42% |
The fact that two customers account for 421 TP3T of revenue means that if the company earns 100,000 won, 42,000 won comes out of the pockets of these two companies. Furthermore, in 2025, Microsoft alone accounted for 261 TP3T. A single customer took up more than a quarter of the total.
Even if we broaden the scope slightly, the picture remains similar. The company discloses its customers by dividing them into three groups; as of 2025, large cloud and AI customers accounted for approximately 481 TP3T, general enterprises 321 TP3T, and AI and specialized service providers 201 TP3T. The structure is such that nearly half comes from just a handful of sources. Quarterly report filed with the SECThe description regarding customer focus is included exactly as is, so you can verify it yourself.
The reason this is a problem is that these customers are all companies capable of designing their own chips and networks. They buy Arista equipment because it is faster and cheaper right now, not because they are structurally bound by the relationship. If one company reduces its volume while redesigning its next-generation data center, subsequent orders will stop, no matter how much deferred revenue has accumulated. I view this not as "revenue locked in a contract," but as "revenue maintained by relationships.".
Arista Networks (ANET) Valuation — Is a forward P/E of 37x expensive?
First, let's start with the basic figures. The market capitalization is $243.2 billion (approximately 327.25 trillion won), and it holds $13.3433 billion (approximately 17.9547 trillion won) in cash and short-term investment assets without a single penny of debt. It is a net cash company. The free cash flow for the past 12 months was $5.1553 billion (approximately 6.937 trillion won). Free cash flow is the actual money remaining after deducting capital expenditures from cash earned.
The current forward P/E ratio (how many times the stock price is compared to expected earnings over the next year) is 37.4 times. This value is based on the 2027 consensus earnings per share of $5.16 (approximately 6,943 won). The consensus refers to the average of analyst forecasts, or market expectations.
You cannot tell whether 37.4 times is expensive or cheap just by looking at the current multiple. So, I looked for two separate baselines. One is Arista's own 5-year average forward P/E. 36.9 times, the other is the competitor's median 26 timesIt was. Mix these two with your own history 65% and the competitor's 35% to find the appropriate multiplier. 33.0 timesWe set it as such. Since our growth rate and profit margin are clearly higher than our competitors, we placed more weight on our own history.
Then, the current 37.4 times is compared to the appropriate 33.0 times. 13.2% PremiumIt is not extremely overvalued, but appears to be in a "slightly expensive and reasonable range.".
However, I also calculated the theoretical fair value separately, and the results diverged quite a bit here. I used the Discounted Cash Flow (DCF) method, which calculates the company's future earnings at their current value and adds them. I set the Weighted Average Cost of Carrying Out (WACC) to 9.5–11.51 TP3T, the perpetual growth rate to 3.0–3.51 TP3T, and the forecast period to 5 years.
| DCF assumption | 값 | reason |
|---|---|---|
| Discount rate (WACC) | 9.5~11.5% | Downward revision of cost of equity based on beta 1.62 to reflect zero debt and net cash structure |
| End-of-life growth rate | 3.0~3.5% | Long-term price + economic growth level |
| forecast period | 5 years | AI Capital Investment Cycle Visibility |
| Starting free cash flow | $5.8 billion | 2026 estimate (upward revision from the last 12 months of $5.16 billion) |
| Calculated fair price range | $95–$149 | The median is about $114 |
To be honest, the answers from the two methods differ significantly. In terms of multiples, it is hovering around a fair value, but in terms of cash flow, it is clearly overvalued. The current market capitalization is 47 times the free cash flow over the past 12 months. I recommend using the multiple method for calculating the target price, but the gap indicated by DCF is Scenario weightsWe decided to reflect this in it. I will explain why we divided it that way right below.
Arista (ANET) side-by-side with competitors
| company | Preceding P/E | Growth and Profitability Profile | Financial structure |
|---|---|---|---|
| Arista (ANET) | 37.4 times | Quarterly revenue +37.71 TP3T, operating profit margin 45.41 TP3T | Net cash of $13.3 billion |
| Cisco (CSCO) | About 26 times | FY2026 Revenue of $63.3 billion, low growth, large installed base | holding borrowings |
| Broadcom (AVGO) | About 23 times | Q3 FY2026 AI Revenue 3.2x, Switch Chip Supply | Large borrowings related to the acquisition |
| CIEN | About 32 times | Benefiting from Optical Transmission-Centric Data Center Interconnects | neutrality |
After making the table, one thing kept bothering me. When I analyzed Broadcom (AVGO), I set the fair forward P/E at 25x., That Broadcom is the very company that manufactures the Ethernet chips used in Arista switches. It is a structure where the company that assembles equipment using those chips commands a transaction price 14 times higher than the company that manufactures the chips. The common explanation is that software (the EOS operating system) creates that difference, but I only agree with that explanation halfway.
Even for the same networking equipment, how the market treats it varies widely. Extreme Networks (EXTR) plummeted 201 TP3 T due to a single guidance reduction despite strong earnings. There is an enemy. I view this as a case demonstrating that while the premium is maintained as long as the label of "AI data center" is attached, the multiple collapses once the growth rate drops.
Arista Networks (ANET) Target Price Scenarios: $208, $258, $140
Let me start by disclosing the growth rate assumptions. When we multiplied the past 3-year average annual revenue growth rate of 27.11 TP3T by 0.4, the analyst consensus of 221 TP3T by 0.4, and the industry growth rate of 201 TP3T by 0.2, and added them together, the weighted growth rate is 22.6%It came out. If we apply this to the projected earnings per share of $5.16 for 2027, it comes out to about $6.30 for 2028. This is the earnings the market will be seeing 12 months from now.
| scenario | 28-year EPS | Applied Drainage | Target price | Won | Compared to the current price | weight |
|---|---|---|---|---|---|---|
| stress | $6.90 | 37.5 times | $258 | Approximately 347,165 won | +33.8% | 20% |
| basic | $6.30 | 33.0 times | $208 | Approximately 279,885 won | +7.9% | 45% |
| Weakness | $5.40 | 26.0 times | $140 | Approximately 188,384 won | -27.4% | 35% |
Let me explain why I allocated the weights this way. Looking solely at the multiples, the premium was 131 TP3T, so I could have increased the base weight. However, the median fair value indicated by the DCF was $114, which was significantly lower than the current price, and the structure remained unchanged with two clients accounting for 421 TP3T of revenue. When these two factors overlap, the downside opens up deeply when it does. Therefore, I raised the bearish weight to 351 TP3T.
The weighted average of the three scenarios yields an expected stock price of $194.2. Compared to the current price of $192.84. Expected return +0.71 TP3TIn my opinion, the Buy option falls far short of the expected return of 121 TP3T or higher, but it is above the Sell zone of -81 TP3T.
For reference, the average target price of 28 Wall Street analysts is $241.04, the median is $246.5, and the investment opinion is close to Strong Buy. Consensus aggregationIf you look at the data, most brokerage firms raised their target prices by $20 to $50 immediately after the August earnings. There are two reasons for the significant difference from my figures. One is that they are simply adding around 40 times to the 2028 earnings, and the other is that they hardly reflect customer concentration in their valuation.
Arista (ANET) Risks and Four Bear Triggers
I have summarized the factors that are causing this stock to weaken into four categories. I have written them in the order of “what happened → how it spread → where it is confirmed.”.
- Switching to in-house design for major customers → Revenue 26% hole → Disclosure of customer concentration in annual report. If Microsoft increases the proportion of its own network in next-generation AI data centers, the biggest pillar will be shaken. It is sufficient to simply check the paragraph regarding the proportion of top customers in the annual reports.
- NVIDIA's expansion into Ethernet → Intensified competition → Growth rate of Etherlink's cumulative customer base. NVIDIA's strategy of selling GPUs and networks as a bundleWe are pushing [it]. The key indicator is how much the cumulative customer base of 100 increases in the next quarter.
- Decline in gross profit margin → Reduction in premium multiple → Quarterly gross profit margin. It has already dropped from 65.21 TP3T to 62.91 TP3T, and the Q3 guidance is also at the 631 TP3T level. If it falls below 601 TP3T, the logic of the “software premium” weakens.
- Slowing growth rate of deferred sales → A leading signal of a downturn in orders → Sum of deferred sales on the quarterly balance sheet. It is currently strong at 691 TP3T, but I consider the moment this growth rate falls below the sales growth rate to be the turning point. You can add the current and non-current deferred sales in the balance sheet of the quarterly report.
On the other hand, there are reassuring aspects as well. The company has zero debt and $13.3 billion in cash. It has sufficient financial strength to withstand a sudden cooling of demand. With a profit margin in the 451 TP3T range, it can maintain profitability even if sales drop by 201 TP3T. The risk associated with this stock is not that the company goes bankrupt, but The side where the drainage is cutIt is in.
What I particularly look at
I personally created a five-quarter table and examined it for quite a while; the most interesting pattern was that the growth rate of deferred revenue consistently outpaced the growth rate of revenue. Usually, it is the opposite. I interpreted this as a signal that unsold orders are piling up, and I believe revenue visibility is significantly secured at least through the first half of 2027. That is why, even in a bearish scenario, I did not project earnings per share for 2028 to fall below $5.40.
However, I keep hesitating because the upward revision in August guidance stemmed from the resolution of supply issues rather than demand. It is indeed good news, but once the bottlenecks are cleared, the next upward revision is only possible if there is genuine new demand. I view this stock not as a "bet on AI capital investment," but rather as a "bet on Microsoft and Meta's decisions regarding their next three-year plans." Since the structure is such that the numbers shift entirely if the two clients' roadmaps change, no matter how well the company performs, I have honestly decided to be only half-confident about this aspect.
Arista Networks (ANET) Investment Opinion Conclusion
My investment opinion is Hold (Medium confidence), the 12-month basic target price is $208 (approx. 279,885 won)no see.
I reached this conclusion in three steps. First, I looked at valuation signals; the forward P/E of 37.4x was 13.21 TP3T higher than my estimated fair value of 33.0x, placing it at a level where it was "not extreme, but not cheap either." Next, I calculated the expected return and found it to be +0.71 TP3T, which was far from the +121 TP3T required for a Buy rating according to my criteria. Finally, I examined the fundamentals, which were almost flawless. With revenue growth of 381 TP3T, an operating profit margin of 451 TP3T, zero debt, and $13.3 billion in net cash, the stock offered downside support. However, judging that this was already reflected in the stock price, I decided to hold.
To summarize, I suggest looking at it this way. If you already hold the stock, following just two numbers each quarter—deferred revenue growth and gross profit margin—is sufficient. If you are looking to enter the market, I believe the first step is to calculate whether your position allows you to withstand a bearish trend of $140. The next earnings announcement is scheduled for early November.
Frequently Asked Questions
What is the target price for Arista Networks?
The baseline 12-month target price is $208 (approximately 279,885 KRW). The target is set at $258 for the bullish scenario and $140 for the bearish scenario. Since the weighted average of the three scenarios is +0.71 TP3T, the investment rating has been set to Hold.
Is it always a good thing if deferred revenue increases?
Generally, it is a good sign. It means the payment has already been received and only needs to be recorded as revenue. However, you need to distinguish whether the increase in deferred revenue is due to a high volume of orders or a backlog caused by an inability to deliver goods. Since Arista's inventory is growing more slowly than revenue, I assessed it as being closer to the former.
How dangerous is a customer concentration of 42%?
It is not at a level that would cause immediate problems, but it is a level that should be reflected in the valuation. Since both clients are companies with in-house design capabilities, they are not tied down by long-term contracts. Due to these factors, I have raised my weighting for the bearish scenario to 351 TP 3 T.
Why did Arista's stock price fall from its August high?
There were no issues with the earnings themselves. The stock hit an all-time high of $214.89 on August 4, the day after the earnings report, but the trend reversed after the market fully reflected the third guidance upgrade. There was also an impact as AI-related stocks in general were suppressed in September due to the overlapping debate regarding the slowdown in AI capital investment.
What happens to Arista if Nvidia sells Ethernet switches directly?
I don't think it will be easy to replace it immediately. Arista's strengths lie in its EOS network operating system and operational know-how rather than its hardware, and the burden of switching operating systems is significant for large clients. However, sharing market share in new deployments could be sufficient pressure. You just need to monitor the growth rate of EtherLink's cumulative customer base.
References
- Arista Networks IR — Q2 2026 Earnings Press Release
- SEC EDGAR — Arista Networks Quarterly Report (Form 10-Q)
- Full Transcript of Arista Networks Q2 2026 Earnings Conference Call
- Morningstar — Analysis of Significantly Upgraded Guidance Driven by Supply Improvements
- StockAnalysis — ANET Analyst Target Price Consensus
The draft was created by organizing publicly available earnings data and market data using AI research tools, and I personally checked every line of the figures and calculations in the tables before publication. Stock prices, exchange rates, and estimates are based on the time of writing and may change over time. Please remember that this post was written to share information, not as a recommendation to buy or sell specific stocks, and that the final judgment and consequences rest solely with the investor.