Sign in

You're signed outSign in or to get full access.

Arista Networks - Earnings Call - Q4 2024

February 18, 2025

Executive Summary

  • Strong Q4 finish: revenue $1.93B (+6.6% q/q, +25.3% y/y), non-GAAP EPS $0.65, GAAP EPS $0.62; non-GAAP operating margin 47.0%. Management emphasized execution, with “over 95% year-over-year growth in operating cash flow for the quarter.”
  • Results were above company guidance: revenue topped the prior $1.85–$1.90B guide, non-GAAP gross margin slightly above the 63–64% guide, and operating margin well above ~44%.
  • FY25 outlook tightened upward: revenue growth now ~17% (≈$8.2B) vs prior 15–17%; FY25 non-GAAP GM reiterated at 60–62% and OM at 43–44%. Q1’25 guide: revenue $1.93–$1.97B, ~63% non-GAAP GM, ~44% OM.
  • AI remains the key narrative: Meta deployed Arista’s 7700R4 Distributed Etherlink Switch; Arista reiterated its 2025 targets of ~$750M AI back-end and ~$1.5B total AI networking revenue (front-end + back-end).

What Went Well and What Went Wrong

  • What Went Well

    • Top-line and profitability beat internal targets: Q4 revenue $1.93B, non-GAAP GM 64.2% (slightly above guide), non-GAAP OM 47.0% vs ~44% guided. CFO: “exceeding our guidance on all key metrics.”
    • Cash generation and balance sheet: “over 95% y/y growth in operating cash flow” in Q4; cash, equivalents and marketable securities ≈$8.3B; share repurchases continued ($123.8M in Q4).
    • AI/customer momentum: Meta AI cluster deployment; reiterated confidence in $8.2B FY25 revenue (+17%) and ~$1.5B AI revenue aspiration. “We remain optimistic about achieving our AI revenue goal of $1.5 billion in 2025.”

  • What Went Wrong

    • Gross margin compression y/y: GAAP GM 63.8% (vs 64.9% in Q4’23) as mix shifted toward large cloud/AI customers; management cited mix and absorbing some China tariffs.
    • International softness q/q: International declined to 16% of revenue (from 17.6% in Q3) due to higher relative domestic mix.
    • AI timing variability and deferred dynamics: one of five AI customers “stalled” awaiting GPUs/funding; product deferred revenue rose with more trials/acceptance clauses, increasing recognition variability into 2025.
View the full earnings report

Transcript

Operator (participant)

Welcome to the fourth quarter 2024 Arista Networks Financial Results Earnings Conference Call. During the call, all participants will be in a listen-only mode. After the presentation, we will conduct a question-and-answer session. Instructions will be provided at that time. If you need to reach an operator at any time during the conference, please press the star key followed by zero. As a reminder, this conference is being recorded and will be available for replay from the Investor Relations section on the Arista website following this call. Mr. Rudolph Araujo, Arista's Head of Investor Advocacy, you may begin.

Rudolph Araujo (Head of Investor Relations)

Thank you, Regina. Good afternoon, everyone, and thank you for joining us. With me on today's call are Jayshree Ullal, Arista Networks' Chairperson and Chief Executive Officer, and Chantelle Breithaupt, Arista's Chief Financial Officer. This afternoon, Arista Networks issued a press release announcing the results for its fiscal fourth quarter ending December 31st, 2024. If you want a copy of the release, you can access it online on our website.

During the course of this conference call, Arista Networks management will make forward-looking statements, including those relating to our financial outlook for the first quarter of the 2025 fiscal year, longer-term business model, and financial outlooks for 2025 and beyond, our total addressable market, and strategy for addressing these market opportunities, including AI, customer demand trends, supply chain constraints, component costs, manufacturing output, inventory management, and inflationary pressures on our business, lead times, product innovation, working capital optimization, and the benefits of acquisitions, which are subject to the risks and uncertainties that we discuss in detail in our documents filed with the SEC, specifically in our most recent Form 10-Q and Form 10-K, and which could cause actual results to differ materially from those anticipated by these statements. These forward-looking statements apply as of today, and you should not rely on them as representing our views in the future.

We undertake no obligation to update these statements after this call. Also, please note that certain financial measures we use on this call are expressed on a non-GAAP basis and have been adjusted to exclude certain charges. We have provided reconciliations of these non-GAAP financial measures to GAAP financial measures in our earnings press release. With that, I will turn the call over to Jayshree.

Jayshree Ullal (CEO)

Thank you, everyone, for joining us this afternoon for our Fourth Quarter 2024 Earnings Call. First, I'd like to warmly welcome our new IR leadership duo of Rudolph Araujo, our director of IR Advocacy, supported by Rod Hall, that many of you may know as our leader for IR Strategy. Special thanks to Liz Stine for her tenure both as Systems Engineer and IR lead at Arista, well, I think you'll all agree that 2024 has been a memorable and defining year for Arista. We started with an initial guidance of 10%-12% annual revenue growth. With the momentum of Generative AI, we have achieved well beyond that at almost 20% growth, achieving a record revenue of $7 billion, coupled with a non-GAAP operating margin of 47.5%. Before I dwell on that more, let me get back to Q4 2024 specifics.

We delivered revenues of $1.93 billion for the quarter, with a non-GAAP earnings per share of $0.65, adjusted for the recent 4:1 stock split. Our non-GAAP gross margins of 64.2% were influenced by efficient supply chain and manufacturing, as well as a good mix of enterprise and software in the quarter. International contribution for the quarter registered at 16%, with the Americas super strong at 84%. Now shifting to annual sector revenue for 2024, our Cloud and AI Titans contributed significantly at approximately 48%, keeping in mind that Oracle is a new member of this category. Enterprise and financials were strong at approximately 35%, while the providers, which now includes Apple, were at 17% approximately. Both Microsoft and Meta are greater than 10% concentration customers at approximately 20% and 14.6%, respectively. As you know, we cherish our privileged partnerships with both of them very much.

It has spanned over 14 years as we collaborate deeply with joint engineering and innovative AI and cloud products. In terms of annual 2024 product lines, our core cloud AI and data center products are built off a highly differentiated extensible OS stack and are successfully deployed across 10, 25, 100, 200, 400, and 800 gigabit Ethernet speeds. It delivers power efficiency, high availability, automation, and agility as the data centers demand insatiable bandwidth capacity and network speeds for both front-end and back-end storage, compute, and AI zones. This core product line drove approximately 65% of our revenue. We continue to gain market share in the highest performance of the switching category of 100, 200, and 400 gig ports to attain the number one position at greater than 40% market share, according to industry analysts in ports.

We have increased our 400 gig customer base to approximately 1,000 customers last year in 2024. We expect 800 gigabit Ethernet to emerge as an AI back-end cluster in 2025. We remain optimistic about achieving our AI revenue goal of $1.5 billion in AI centers, which includes the $750 million in AI back-end clusters in 2025. Our network adjacencies market, comprised of routing, replacing routers, and the cognitive AI-driven campus, is going well. Our investments in cognitive wired, wireless, Zero Touch Provisioning, and network identity, as well as sensors for threat mitigation, are being received extremely well by our campus customers. Our recent modern stacking introduction of SWAG, Switch Aggregation Group, is a fitting example of our compelling innovation for open and efficient networking, conserving IP addresses without proprietary methods.

The post-pandemic campus is very different, and our customers are seeking alternatives to legacy incumbents with deep Zero Trust Security, high availability, and observability embedded in the network across our software stack with CloudVision management. We are committed to the $750 million goal in 2025 and much more ahead. We have successfully also deployed in routing, edge, and peering use cases. Just in 2024 alone, we introduced six EOS software releases with greater than 600 new features across our core and adjacent offerings. The campus and routing adjacencies together contribute approximately 18% of revenue. Our third category is network software and services based on subscription models such as Arista A-Care, CloudVision, DMF observability, and advanced security sensors for network detection and response. We added over 350 CloudVision customers, translating to literally one new customer a day.

CloudVision is pivotal to building our network as a service and deploying Arista Validated Designs in the enterprise. Arista's subscription-based network services and software contributed approximately 17% of total revenue. Note that perpetual licenses do not count here and go into the core or adjacent sections. While the 2024 headline has clearly been about generative AI, Arista continues to diversify its business globally with multiple use cases and verticals. We are viewed as the modern network innovator of choice for client to campus to cloud and AI networking, ideally positioned with our differentiated foundation. We celebrated two milestones in 2024: our 10th anniversary of going public at the New York Stock Exchange and our 20th anniversary of founding. In the past decade, we have exceeded 10,000 customers with a cumulative of 100 million ports of installed base, as Arista drives the epicenter of mission-critical network transactions.

Arista 2.0 strategy is resonating exceptionally well with our customers. Customers are not only looking to connect, but unify and consolidate their data across silos for optimal networking outcomes. Our modern networking platforms are foundational for transformation from incongruent silos to centers of data, and it places us in a very unique position as the best-of-breed innovator for data-driven networking. These centers of data, as we call it, can reside in the campus as a campus center or data centers or WAN centers or AI centers, regardless of their location. Networking for AI is also gaining traction as we move into 2025, building some of the world's greatest Arista AI centers at production scale. These are constructed with both back-end clusters and front-end networks.

And as I've shared with you often, the fidelity of the AI traffic differs greatly from cloud workloads in terms of diversity, duration, and size of flow. Just one slow flow can slow the entire job completion time for a training workload. Therefore, Arista AI centers seamlessly connect to the front end of compute, storage, WAN, and classic cloud networks with our back-end Arista Etherlink portfolio. This AI-accelerated networking portfolio consists of three families and over 20 Etherlink switches, not just one point switch. Our AI for networking strategy is also doing well, and it's about curating the data for higher-level network functions. We instrument our customers' networks with our Publish-Subscribe state foundation with our software called Network Data Lake to deliver proactive, predictive, and prescriptive platforms that have superior AI ops with A-Care support and product functions.

We are pleased to surpass for the first time the $1 billion revenue mark in 2024 for the software and subscription service category. In 2024, we conducted three very large customer events in London, New York, and Santa Clara, California. Our differentiated strategy and superior products are resonating deeply as we touched over 1,000 strategic customers and partners in these exclusive events. Simply put, we outpace the industry in quality and support with the highest Net Promoter Score of 87, which translates to 93% of customer respondent satisfaction. Of course, we do that with the lowest security and vulnerabilities and steadfast network innovation. In summary, 2024 has been a pivotal turning point for Arista. It has been a key breakaway year as we continue to aim for $10 billion annual revenue with a CAGR of double digits that we set way back in November 2022 Analyst Day.

While I do appreciate the exuberant support from our analyst community on our momentum, I would encourage you to pay attention to our stated guidance. We live in a dynamic world of changes, most of which have resulted in positive outcomes for Arista. We reiterate at the upper range of our 2025 guidance of our double-digit growth at 17%, now aiming for approximately $8.2 billion in 2025 in revenue. The Arista leadership team has driven outstanding progress across multiple dimensions. In 2024, we are at approximately 4,465 employees rooted in engineering and customer investments. I'm incredibly proud of how we've executed and navigated the year based on our core principles and culture. While customers are struggling with change fatigue from our legacy incumbents, Arista is redefining the future of data-driven networking intimately with our strategic customers.

With that, I'd like to turn it over to Chantelle, who has transitioned to become our CFO and Chief Financial Officer in record time, less than a year. Over to you, Chantelle, and welcome again and happy one-year anniversary.

Chantelle Breithaupt (CFO)

Thank you, Jayshree, and congratulations on a great 2024. My first full year as CFO has been more than I could have hoped for, and I am excited about Arista's journey ahead. Now on to the numbers. As a reminder, this analysis of our Q4, our full year 2024 results, and our guidance for Q1 2025 is based on non-GAAP and excludes all non-cash stock-based compensation impacts, certain acquisition-related charges, and other non-recurring items. In addition, all share-related numbers are provided on a post-split basis to reflect the 4:1 stock split in December 2024. A full reconciliation of our selected GAAP to non-GAAP results is provided in our earnings release. Total revenues in Q4 were $1.93 billion, up 25.3% year-over-year, and above the upper end of our guidance of $1.85-$1.9 billion.

For fiscal year 2024, we are pleased to have delivered 19.5% in revenue growth driven by achievements in all three of our product sectors. Services and subscription software contributed approximately 18.3% of revenue in the fourth quarter, up from 17.6% in Q3. International revenues for the quarter came in at $311.1 million, or 16% of total revenue, down from 17.6% last quarter. This quarter-over-quarter decrease was driven by the relative increased mix of domestic revenue from our large global customers. The overall gross margin in Q4 was 64.2%, slightly above the guidance of 63%-64%, and down from 65.4% in the prior year. As a recap for the year, we delivered a gross margin result of 64.6% compared with 62.6% for the prior year. This increase is largely due to a combination of improved supply chain and inventory management.

Operating expenses for the quarter were $332.4 million, or 17.2% of revenue, up from the last quarter at $279.9 million. R&D spending came in at $226.1 million, or 11.7% of revenue, up from 9.8% last quarter. This matches the expectations discussed in our Q3 earnings call regarding the timing of engineering costs and other costs associated with the development of our next-gen products moving from Q3 to Q4. This finishes the year of R&D at 11.2% of revenue, demonstrating a continued focus on product innovation. Sales and marketing expense was $86.3 million, or 4.5% of revenue, up from $83.4 million last quarter. This was driven by continued investment in both headcount and channel programs. Our G&A costs came in at $19.9 million, or 1% of revenue, up from $19.1 million last quarter, reflecting continued investment in scaling the company.

Our operating income for the quarter was $907.1 million, or 47% of revenue. This strong Q4 finish contributed to an operating income result for fiscal year 2024 of $3.3 billion, or 47.5% of revenue. Congratulations to the Arista team on this impressive achievement. Other income and expense for the quarter was a favorable $89.3 million, and our effective tax rate was 16.7%. This lower-than-normal quarterly tax rate reflected the release of tax reserves due to the expiration of the statute of limitations and favorable changes in state taxes. This resulted in net income for the quarter of $830.1 million, or 43% of revenue. Our diluted share number was 1.283 billion shares, resulting in a diluted earnings per share for the quarter of $0.65, up 25% from the prior year. For FY 2024, we are pleased to have delivered a diluted earnings per share of $2.27, a 31.2% increase year-over-year.

Now turning to the balance sheet, cash, cash equivalents, and marketable securities ended the quarter at approximately $8.3 billion. In the quarter, we repurchased $123.8 million of our common stock at an average price of $94.80 per share. Within fiscal 2024, we repurchased $423.6 million of our common stock at an average price of $77.13 per share. Of the $1.2 billion repurchase program approved in May 2024, $921 million remains available for repurchase in future quarters. The actual timing and amount of future repurchases will be dependent on market and business conditions, stock price, and other factors. Now turning to operating cash performance for the fourth quarter, we generated approximately $1 billion of cash from operations in the period, reflecting strong earnings performance combined with an increase in deferred revenue offset by an increase in income tax payments.

DSOs came in at 54 days, down from 57 days in Q3, reflecting the timing of shipments and a strong collections performance by the team. Inventory turns were 1.4 times, up from 1.3 last quarter. Inventory increased marginally to $1.83 billion, reflecting diligent inventory management across raw and finished goods. Our purchase commitments at the end of the quarter were $3.1 billion, up from $2.4 billion at the end of Q3. As mentioned in prior quarters, this expected activity represents purchases for chips related to new products and AI deployments. We will continue to rationalize our overall purchase commitment number. However, we expect to maintain a healthy position related to key components and continue to have some variability in this number to meet customer demand and improve lead times in future quarters. Our total deferred revenue balance was $2.79 billion, up from $2.51 billion in the prior quarter.

The majority of the deferred revenue balance is services related and directly linked to the timing and term of service contracts, which can vary on a quarter-by-quarter basis. Our product deferred revenue balance increased by approximately $150 million over the last quarter. Fiscal 2024 was a year of new product introductions, new customers, and expanded use cases. These trends have resulted in increased customer trials and contracts with customer-specific acceptance clauses that have and will continue to have increased the variability and magnitude of our product deferred revenue balances. We expect this to continue into fiscal 2025. Accounts payable days were 51 days, up from 42 days in Q3, reflecting the timing of inventory receipts and payments. Capital expenditures for the quarter were $12.5 million.

In October, we began our initial construction work to build expanded facilities in Santa Clara, and we expect to incur approximately $100 million in CapEx during fiscal 2025 for this project. Now turning to our outlook for the first quarter of 2025 and the remainder of the fiscal 2025 year, we continue to gain confidence in our view for fiscal year 2025 and now place our revenue growth outlook at approximately 17%, or $8.2 billion. This is up from our initial FY 2025 guidance of 15%-17%. This reflects our combined outlook for cloud, AI enterprise, and cloud specialty providers, along with the recognition of the volatility that we have seen in the market since the beginning of the year.

For gross margin, we reiterate the range of the fiscal year of 60%-62%, with Q1 2025 expected to be above the range due to the anticipated mix of business in the quarter. Similar to others in the industry, we will continue to monitor the fluid tariff situation and be thoughtful for both the short and long-term outcomes to both our company and our customers. In terms of spending, we expect to invest in innovation, sales, and scaling the company, resulting in a continued operating margin outlook of 43%-44% in 2025. On the cash front, we will continue to work to optimize our working capital investments, with some expected variability in inventory due to the timing of component receipts on purchase commitments.

Our structural tax rate is expected to remain at 21.5%, back to the usual historical rate, up from the unusually low one-time rate of 16.7% experienced last quarter, Q4 FY 2024. With all this as a backdrop, our guidance for the first quarter is as follows: revenues of approximately $1.93-$1.97 billion, a slightly stronger seasonality in Q1 than prior year trends and outcome of the timing of our customers' priorities, gross margin of approximately 63%, and operating margin at approximately 44%. Our effective tax rate is expected to be approximately 21.5%, with approximately 1.285 billion diluted shares. In summary, we at Arista are enthusiastic about 2025 and the general networking outlook ahead. We have an impressive portfolio and are ready to solve our customers' needs across all the data centers.

Combined with our Arista team spirit, we are ready to realize our fair share of the $70 billion market TAM. With that, I too would like to welcome Rudy and Rod to the Arista IR team. Back over to you, Rudy, for Q&A.

Rudolph Araujo (Head of Investor Relations)

Thank you, Chantelle. We will now move to the Q&A portion of the Arista earnings call. To allow for greater participation, I'd like to request that everyone please limit themselves to a single question. Thank you for your understanding. Operator, take it away.

Operator (participant)

We will now begin the Q&A portion of the Arista earnings call. In order to ask a question during this time, simply press star, then the number one on your telephone keypad. If you'd like to withdraw your question, press star and the number one again. We ask that you pick up your handset before asking questions in order to ensure optimal sound quality. Our first question will come from the line of Michael Ng at Goldman Sachs. Please go ahead.

Michael Ng (Analyst)

Hi, good afternoon. Thank you for the question. I was just wondering if you could talk about the timing of how the year might look like and how do you expect Arista switches in the AI backend to be rolled out into production? Are the sale of these switches tied to the deployment of next-generation NVIDIA chips or hyperscale custom ASICs on the compute side, and is that a gating factor that you're watching out for? Thank you.

Jayshree Ullal (CEO)

Yeah, thank you, Michael. First, I want to say Arista is still committed to four out of our five AI clusters that I mentioned in prior calls. The fifth one is a little bit stalled. It is not a cloud titan. They are awaiting new GPUs and some funding too, I think. So I hope they'll come back next year, but for this year, we won't talk about them. But the remaining four, let me spend some to give some color. Three out of the four customers are expected to this year roll out a cumulative of 100,000 GPUs. So we're going to do very well with three of them on the backend. And you can imagine they're all pretty much one major NVIDIA class of GPU. They will be waiting for the next generation of GPUs, but independent of that, we'll be rolling out fairly large numbers.

On the fourth one, we are migrating right now from InfiniBand to proving that Ethernet survivable solution. So we're still, you know, they've historically been InfiniBand. And so we're still in pilots, and we expect to go into production next year. So doing very well in four out of four. The fifth one is stalled, and three out of the four are expected to be 100,000 GPUs this year.

Operator (participant)

Our next question comes from the line of Amit Daryanani with Evercore. Please go ahead.

Amit Daryanani (Analyst)

Yep, good afternoon. Thanks so much for taking my question. You know, I guess, Jayshree, there's always this concern around the impact of white-box vendors to your revenue growth. And clearly, over the last decade, I don't think it's been an impediment for the company. But can you maybe just share your perspective that when it comes to AI networks, especially the backend networks, you know, how do you see the mix evolving, white-box versus OEM solutions? And maybe just help us understand the differentiators that helped Arista be successful on the front end. Do they extend to the backend networks as well, or is there something different we should be aware about? Thank you.

Jayshree Ullal (CEO)

Yes, sure, Amit. Gosh, this seems like a déjà vu question, but thank you for asking it. I'm sure it's on a lot of minds. We've been going through this for at least 10 years on the cloud, and the first thing I just want to say is this TAM is so huge and so large. We will always coexist with white boxes and operating systems that are non-EOS, much like Apple coexists on their iPhone with other phones of different types. When you look at the backend of an AI cluster, there are typically two components: the AI leaf and the AI spine. The AI leaf connects to the GPUs and therefore is the first, if you will, point of connection, and the AI spine aggregates all of these AI leaves. Almost in all the backend examples we've seen, the AI spine is generally 100% Arista-branded EOS.

You've got to do an awful lot of routing, scale, features, capabilities that are very rich that would be difficult to do in any other environment. The AI leaf can vary. So, for example, let's take the example of the five customers I've mentioned a lot. Three out of the five are all EOS in the leaf and spine. Two out of the five are kind of hybrid. Some of them have some form of SONiC or FBOSS. And as you know, we co-develop with them and coexist in a number of use cases where it's a real hybrid combination of EOS and an open OS. So, for the most part, I'd just like to say that white box and Arista will coexist and will provide different strokes for different folks. Now, in terms of differentiators, a lot of our deployments right now is 400 gig and 800 gig.

And we see a tremendous amount of differentiation, not only, like I explained to you, in scale and routing features, but cluster load balancing, AI visibility and analytics in real time, per-span queuing, congestion control, visibility, and most importantly, Smart System Upgrade, because you sure don't want your GPUs to come down because you don't have the right software to accelerate so that the network provides the ideal foundation that if a GPU is in trouble, we can automatically give it a different connection and an alternate connection. So, tremendous amount of differentiation there, and even more valuable in a GPU which costs typically five times as much as a CPU.

Operator (participant)

Our next question comes from the line of Tim Long at Barclays. Please go ahead.

Tim Long (Analyst)

Thank you. I wanted to touch on the cloud titan numbers a few parts there. Obviously, one of them, Meta, looks like based on the numbers you gave, if I heard it right, is down year-over-year, if you could touch on that. And then the other, you know, if we do the math for the other Cloud Titans, looks like it went up a lot. I don't know. I think Oracle was kind of in that already. Was there anything else going on other than the Oracle shift with the rest of the Cloud Titans where it looked extremely strong in 2024? Thank you.

Jayshree Ullal (CEO)

Yeah. So, speaking specifically to Meta, we're obviously in a number of use cases in Meta. Keep in mind that our 2024 Meta numbers are influenced by more of their 2023 CapEx, and that was Meta's year of efficiency where their CapEx was down 15%-20%. So you're probably seeing some correlation between their CapEx being down and our revenue numbers being slightly lower in 2024. In general, I would just say all our Cloud Titans are performing well in demand, and we shouldn't confuse that with timing of our shipments. And I fully expect Microsoft and Meta to be greater than 10% customers in a strong manner in 2025 as well. Specific to the others we added in, they're not 10% customers, but they're doing very well, and we're happy with their cloud and AI use cases.

Operator (participant)

Our next question comes from the line of Ben Reitzes with Melius. Please go ahead.

Ben Reitzes (Managing Director and Head of Technology Research)

Yeah, darn Tim Long took my question. So I'm going to ask about gross margins.

Jayshree Ullal (CEO)

Okay.

Ben Reitzes (Managing Director and Head of Technology Research)

Yeah, I'll turn that guy. So about gross margins, so obviously they go between to 61% at the midpoint after being much higher in the first quarter. I would think that implies significant cloud titan mix, though, going throughout the rest of the year. Do you mind just giving some color on what is pushing down the gross margin a little more? And does that mean that Cloud Titans do accelerate throughout the year because the gross margin gets pushed down in your guidance? Thank you.

Chantelle Breithaupt (CFO)

Yeah, for sure. Sorry.

Jayshree Ullal (CEO)

Go ahead.

Chantelle Breithaupt (CFO)

Yes, sorry, Jayshree. No, absolutely. I think you got it right, Ben, from that perspective. As we entered Q4 last year, talking about 25 guidance, and as we enter into this quarter, it is and still remains a mix. You know, Jayshree kind of gave some thoughts onto the timing to the first question. So it is mixed-driven. There were some questions last quarter if it was price-driven. This is just a mixed-driven conversation. I would say we have absorbed a little bit of the specific tariffs on China in that number, so we are absorbing that on behalf of our customers. But otherwise, it's mixed-driven, and we'll continue to update as we do the quarterly guidance through the year.

Jayshree Ullal (CEO)

John, you had a question. Sorry, I was just going to add that John's done a fantastic job on the planning for China ahead of time.

So while they're absorbing the costs, most of it is related to the mix, and some of it is related to the China tariffs. Would you say that?

John McCool (SVP and Special Advisor)

That's right. Absolutely. We've been working on China mitigation for some time and happy to report we've made good progress.

Operator (participant)

Our next question comes from the line of Meta Marshall at Morgan Stanley. Please go ahead.

Meta Marshall (Analyst)

Great. Thanks. Maybe another topical question from investors just over the past month has been DeepSeek. And just as you think about kind of this one-to-one ratio you've talked about on backend versus frontend, how you kind of see that changing as we've seen some of kind of the changes to kind of thoughts around training investments. Thanks.

Jayshree Ullal (CEO)

Yeah, thank you, Meta. Well, DeepSeek certainly dips many stocks, but I actually see this as a positive because I think you're now going to see a new class of CPUs, GPUs, AI accelerators, and where you can have substantial efficiency gains that go beyond training. So that could be some sort of inference or mixture of experts or reasoning, which lowers the token count and therefore the cost. So what I like about all these different options is Arista can scale up network for all kinds of XPUs and accelerators. And I think the eye-opening thing here for all of our experts who are building all these engineering models is there are many different types, and training isn't the only one.

So I think this is a nice evolution of how AI will not just be a backend training only limited to five customers type phenomenon, but will become more and more distributed across a range of CPUs and GPUs.

Meta Marshall (Analyst)

Great. Thanks.

Jayshree Ullal (CEO)

Thank you, Meta.

Operator (participant)

Our next question comes from the line of Aaron Rakers at Wells Fargo. Please go ahead.

Aaron Rakers (Analyst)

Yeah, thanks for taking the question. Jayshree, I'm curious, just kind of thinking about some of the questions we've gotten recently, is when you see announcements like Stargate, and obviously Stargate has the involvement of one of your newer cloud titan customers, how do you conceptualize the opportunity set for Arista vis-à-vis both backend and frontend networking in deployments like that? And then do you have any thoughts on just the broader context of what you're seeing on sovereign AI opportunities in your business? Thank you.

Jayshree Ullal (CEO)

Yes, thank you, Aaron. Stargate and sovereign AI are not quite related, so let me take the first one, Stargate first. You know, if you look at how we have classically approached GPUs and collective libraries, you know, we've largely looked at it as two separate building blocks. There's the vendor who provides the GPU, and then there's us who provides the scale-out networking. But when you look at Stargate and projects like this, I think you'll start to see more of a vertical rack integration where the processor, the scale-up, the scale-out, and all of the software to provide a single point of control and visibility starts to come more and more together.

This is not a 2025 phenomenon, but definitely in 2026 and 2027, you're going to see a new class of AI accelerators and a new class of training and inference, which is extremely different than the current more pluggable Lego type of version. So we're very optimistic about it. Andy Bechtolsheim is personally involved in the design of a number of these next-generation projects. And the need for this type of, shall we say, pushing Moore's law of improvements in density and performance that we saw in the 2000s is coming back. And you can boost more and more performance per XPU, which means you have to boost the network scale from 800 gig to 1.6T. There are other things to consider, like liquid cooling and co-packaged copper and optics.

So, lots going on there that Arista's in the middle of, best of breed hardware that John McCool's team is working on, as well as Andy's team is.

Operator (participant)

Our next question comes from the line of Atif Malik at Citi. Please go ahead.

Atif Malik (Analyst)

Hi, thank you for taking my question. Jayshree, I appreciate you calling out to pay attention to the guidance. Now, you are reiterating $750 million AI backend sales this year despite the stalled order, the fifth customer. Can you talk about where is the upside coming from this year? Is it broad-based or one or two customers? And also, if you can talk to the $70 billion TAM number for 2028, how much is AI?

Jayshree Ullal (CEO)

Okay, so I'll take your second question first, Atif. On the $70 billion TAM in 2028, I would roughly say a third is AI, a third is data center and cloud, and a third is campus and enterprise, and obviously absorbed into that is routing and security and observability. I'm not calling them out separately for the purpose of this discussion, so roughly $20 billion to $25 billion on each to get to that $70 billion, so coming back to your original question, which was, help me out again.

Atif Malik (Analyst)

The $750 million in backend sales?

Jayshree Ullal (CEO)

Yeah, yeah. So as I said, we're well on our way, and three customers deploying a cumulative of 100,000 GPUs is going to help us with that number this year. And as we increase our guidance to $8.2 billion, I think we're going to see momentum both in AI, cloud, and enterprises. I'm not ready to break it down and tell you which where. I think we'll know that much better in the second half. But Chantelle and I feel confident that we can definitely do the $8.2 billion that we historically don't call out so early in the year. So having visibility of that helps.

Atif Malik (Analyst)

Thank you.

Jayshree Ullal (CEO)

Thank you.

Operator (participant)

Our next question comes from the line of Samik Chatterjee with JP Morgan. Please go ahead.

Samik Chatterjee (Analyst)

Hi. Thanks for taking the question. Jayshree, maybe I can sort of bring up one more topic that's come up a lot in the last few days, which is the value of the EOS software layer to the backend of the network, and particularly in that discussion in terms of relative competition to a white box player. How do you sort of emphasize the value of EOS to your customers? Can you sort of outline some of the sort of what are the key drivers we should keep in mind? And again, in that competitive landscape between white box and Arista. Thank you.

Jayshree Ullal (CEO)

Yeah, sure, Samik. First of all, when you're buying these expensive GPUs that cost $25,000, they're like diamonds, right? You're not going to string a diamond on a piece of thread. So first thing I want to say is you need a mission-critical network, whether you want to call it white box, blue box, EOS, or some other software. You've got to have mission-critical functions, analytics, visibility, high availability, etc. As I mentioned, and I want to reiterate, they're also typically a leaf spine network. And I have yet to see an AI spine deployment that is not EOS-based. I'm not saying it can't happen or won't happen, but in all five major installations, the benefit of our EOS features for high availability, for routing, for VXLAN, for telemetry, our customers really see that.

The 7800 is the flagship AI spine product that we have been deploying last year, this year, and in the future. Coming soon, of course, is also the product we jointly engineered with Meta, which is the distributed Ethernet switch. That is also an example of a product that provides that kind of leaf spine combination, both with FBOSS and EOS options in it. In my view, it's difficult to imagine a highly resilient system without Arista EOS in AI or non-AI use cases. On the leaf, you can cut corners. You can go with smaller buffers. You may have a smaller installation. I can imagine that some people will want to experiment and do experiment in smaller configurations with non-EOS. Again, to do that, you have to have a fairly large staff to build the operations for it. That's also a critical element.

Unless you're a large cloud titan customer, you're less likely to take that chance because you don't have the staff. All in all, EOS is alive and well in AI and cloud use cases, except in certain specific use cases where the customer may have their own operation staff to do so.

Samik Chatterjee (Analyst)

Sure. Thank you. Thanks.

Operator (participant)

Our next question comes from the line of Ben Bollin with Cleveland Research. Please go ahead.

Ben Bollin (Analyst)

Good afternoon, everyone. Thanks for taking the question. Jayshree, I'm interested in your thoughts on your enterprise strategy within G2000 and how that may be evolving as it looks like refresh opportunities are intensifying. Thank you.

Jayshree Ullal (CEO)

Yeah, no, listen, we're always looking at three major threads: our classic cloud business, our AI, and the enterprise led by Ashwin and Chris is a very significant area of investment for us. From a product point of view, we have a natural trickle-down effect from our high-end data center products to the cloud. And so whether it's the enterprise data center or the campus, I've never seen our portfolio be as strong as it is today. So a lot of our challenges and our execution is really in the go-to-market, right? And that just takes time. As you know, we've been slowly but steadily investing there. And our customer count, the number of projects we get invited to, especially as you pointed out in the Global 2000, has never been stronger.

One area I'd like to see more strength, and Chris Schmidt and the team are working on it, as you can tell from our numbers, is international. We're bringing in some new leadership there and hope to see some significant contributions in the next year or so.

Operator (participant)

Our next question comes from the line of Ryan Koontz with Needham & Company. Please go ahead.

Ryan Koontz (Analyst)

Great. Thanks. Jayshree, can you comment? There's been a lot of chatter lately about co-packaged optics. Can you maybe speak about its place in your roadmap and how investors should think about that, the effect on your TAM and your opportunities to sell?

Jayshree Ullal (CEO)

First of all, Andy has reminded me that co-packaged optics is not a new idea. It's been around 10-20 years. So the fundamental reason, let's go through why co-packaged optics has had a relatively weak adoption so far is because of field failures, and most of it is still in proof of concept today. So going back to networking, the most important attribute of a network switch is reliability and troubleshooting. And once you solder a co-packaged optics on a PCB, you lose some of that flexibility, and you don't get the serviceability and manufacturing. That's been the problem. Now, a number of alternatives are emerging, and we're a big fan of co-packaged copper, as well as pluggable optics that can complement this, like linear drive or LPO, as we call it.

Now, we also see that if co-packaged optics improves some of the metrics it has right now, for example, it has a higher channel count than the industry standard of eight-channel pluggable optics, but we can do higher channel pluggable optics as well. So some of these things improve. We can see that both CPC and CPO will be important technologies at 224 gig or even 448 gig. But so far, our customers have preferred a Lego approach that they can mix and match pluggable switches and pluggable optics and haven't committed to soldering them on the PCB. And we feel that will change only if CPO gets better and more reliable. And I think CPC can be a nice alternative to that.

Ryan Koontz (Analyst)

Appreciate your thoughts. Thank you.

Operator (participant)

Our next question comes from the line of Simon Leopold with Raymond James. Please go ahead.

Simon Leopold (Analyst)

Thank you very much for taking the question. I was hoping you could maybe double-click on the cognitive adjacencies. It's been a meaningful part of revenue. I think you said 18%. If you could offer a little bit more color about how the elements of that are trending and your expectations for how that part of the business is growing in your 2025 expectations? Thank you.

Jayshree Ullal (CEO)

Yeah, no, as you can imagine, that's an important both the routing and the campus, and we've already signed up to $750 million, so we clearly, out of our $8.2 billion, expect that to be over $1 billion this year, right? Now, in the routing use case, which is particularly enterprise and service provider related, as I've often said, it's difficult to measure it in isolation, so we're very strict about the definition being it has to be a combination of the software running with a dedicated routing hardware. So for example, if their hardware is shared across switching and routing, we don't count it there, so I think sometimes we shortchange the numbers a little bit, and more of it goes in the core, but I just want you to be aware that's a very strategic piece.

You add that to the fact that the SD-WAN market is now evolving. It's not just about how do you do encryption and tunnels and migrate from MPLS, but you really need a routed backbone. So the combination of SD-WAN in the edge and a routed backbone really falls into the sweet spot for Arista, both in the enterprise and service providers. I don't need to tell you about our campus initiatives. We are very, very keen there. We see a parting of those seas, if you will, where there's a lot of fatigue with subscription models on the campus from one competitor and another set of competitors who are trying to do a merger or acquisition. So Arista is the only pure-play campus innovator who can provide that best of breed.

And we're particularly getting traction there where our data center customers are already familiar with us, and they're using a data center spine, and they can extend that same universal spine to wired and wireless leafs. So using CloudVision as a management domain, we're seeing much, much more traction, be it for automation, Zero Trust Security, or observability with our campus products. So I think both of those are meaningful, and we expect them to exceed $1 billion. Chantelle, you want to say a few words?

Chantelle Breithaupt (CFO)

Yeah, the only other things I'd add, Jayshree, to your points, which are completely showing the intent. The other things for campus are a couple fold. First, John, who's here with us, has spent a lot of time getting us better lead times coming into this year. So we have a great lead times conversation. We're very excited about that, and the customers seem to be pretty excited as well. We also have a curated preferred partner program, particularly international, to your point earlier, Jayshree, on growing the international revenue. And we're excited because we've seen some campus-first wins where not only is the DC bringing us in, but we're actually bringing the DC through a campus win. So just add a couple more points to the enthusiasm for 2025.

Jayshree Ullal (CEO)

Thank you. Great reminders.

Operator (participant)

Our next question will come from the line of Tal Liani at Bank of America. Please go ahead.

Tai Liani (Analyst)

Hi, guys. Two questions. One on routing and one on enterprise. Enterprise grew 16% this year. What drives it? Is it just regular growth of data centers, or do you start to see enterprises investing because of AI and applications for AI? The second thing is about routing. Routing used to be a small opportunity when it was just a license. Can you elaborate on routing? What is your differentiation versus the others? Others are bundling it with optics. How do you sell it? And then how big is the opportunity? Not in terms of numbers, but is it now a hardware with software, or is it just software licensed like it used to be? Thanks.

Chantelle Breithaupt (CFO)

Jayshree, I can start with enterprise.

Jayshree Ullal (CEO)

Sure.

Chantelle Breithaupt (CFO)

And then you want to take routing. Yeah, so I think for enterprise, there are a few ways I would describe kind of our growth vectors in 2024 and continuing into 2025 and into the following years. One is coverage. You've seen we've invested in sales and marketing headcount in 2023, 2024, and going into 2025. 2024, we had double-digit increase in sales and marketing headcount. So we're just getting more coverage. We're also using that preferred partner program I mentioned to get into the enterprise. We do have international kind of campaigns that we're working on. We do have a new logo focus. And so I think that all those land and expand motions are the growth vectors. I would say for the AI perspective, speaking with the customers, it's great to move from kind of a theory to more specific conversation.

And you're seeing that in the banks and some of the higher-tier Global 2000 and Fortune 500 companies. And so they're moving from theory to actual use cases they're speaking to. And the way they describe it is it takes a bit of time. They're working mostly with cloud service providers at the beginning, kind of doing some training, and then they're deciding whether they bring that on-prem and inference. So they're making those decisions. So I think those are early days, but we're having really great conversations for the AI part of that. Jayshree, did you want to cover routing?

Jayshree Ullal (CEO)

Yeah. So, talking on routing, routing has always been a critical part of our offerings for the cloud and data center, where, as you rightly described, it was more of a software enhancement. But as we are getting more meaningful and important to the service providers, as well as to the large enterprises, I was just with a very large bank in New York last week. It was snowing there, so it was super cold. So I stayed indoors most of the time. And the use case there is not data center. It's not campus. It's all a WAN routed fabric. It's pretty amazing. And then they're looking for us to not just build that core routing as a hub, but also take it into the spoke. And you mentioned features.

I think what happened with Arista is we were largely building features for the cloud and data center, which is 20% of the features. But today, our routing portfolio is much more complete: VXLANSec, TunnelSec, MACsec encryption, MPLS, segment routing, OSPF, BGP. We got it all. So we're no more apologizing for what we don't have in routing. And we obviously have the best software stack in terms of quality and support in the industry. So while we are selling a lot of software SKUs, we are now finding ourselves in a lot of dedicated hardware SKUs, particularly with the 7280 platform. That has been a real workhorse for us and very successful.

Tai Liani (Analyst)

Great. Thank you.

Jayshree Ullal (CEO)

Thank you, Tal.

Operator (participant)

Our next question comes from the line of Antoine Chkaiban with New Street Research. Please go ahead.

Antoine Chkaiban (Analyst)

Hi. Thank you for taking my question. I'd love to get your latest perspective on what you're hearing from service providers in AI. One of your competitors mentioned that they were seeing AI driving demand for service providers because they're building out their network in anticipation of an increase in traffic driven by AI. So just wondering if you could comment on that as well. Thank you.

Jayshree Ullal (CEO)

Antoine, do you mean the classic service providers or generally the Neo Clouds?

Antoine Chkaiban (Analyst)

The classic service providers.

Jayshree Ullal (CEO)

Oh, okay. So we haven't seen a huge uptick there yet. I think maybe some experimental. But we have, to answer a question you didn't ask, we have seen much more activity. And somebody earlier that I didn't answer, the sovereign cloud, the Neo Clouds, we are seeing a new class of Tier 2 specialty cloud providers emerge that want to provide AI as a service and want to be differentiated there. And there's a whole lot of funding, grant money, real money going in there. So service providers too early to call. But Neo Clouds and specialty providers, yeah, we're seeing lots of examples of that.

Antoine Chkaiban (Analyst)

Thank you.

Jayshree Ullal (CEO)

Thank you.

Operator (participant)

Our next question comes from the line of Matt Niknam with Deutsche Bank. Please go ahead.

Matthew Niknam (Analyst)

Hey, thanks so much for taking the question. Maybe for Chantelle, I mean, you're sitting now on $8 billion worth of cash and equivalents on the balance sheet. So maybe just an update on how you'd prioritize uses of cash heading into 2025. Thank you.

Chantelle Breithaupt (CFO)

Yeah, no, it's great. Thank you for the question. We're very pleased with the performance in our capital allocation strategy has not changed coming into FY 2025. Just to kind of reiterate and remind, and I appreciate the opportunity to do so. First of all, in the sense of investing that cash where we can still get a very reasonable and respectable return continues to be a priority. Repurchasing, you saw that we did through 2024 and willing to do what we can through 2025. Organic investment. So you saw that in the sense we're still looking to scale the company in R&D, sales, back office. And probably the one that's the least on the scale is sizable inorganic activity. So I would focus on the first four, and that's how we'll remain in 2025.

Operator (participant)

Our next question comes from the line of David Vogt with UBS. Please go ahead.

David Vogt (Analyst)

Great. Thanks, guys, for taking my question. So, Jayshree, I have a question about sort of the evolution of speed deployments at some of your M and M customers. Obviously, you mentioned 400G and 800G have been obviously a principal driver. How are you thinking about how that plays out in 2025 and beyond? There's some wins out there. I know different parts of the network at 1.6, but trying to get a sense for how you think about 400 into 800 into ultimately 1.6, not just in 2025, but in 2026 and beyond. Thanks.

Jayshree Ullal (CEO)

Yeah. No, I think that's a very good question. The speed transitions because of AI are certainly getting faster. It used to take when we went from 200 gig, for example, at Meta or 100 gig in some of our Cloud Titans to 400 gig, that speed transition typically took three to four, maybe even five years, right? In AI, we see that cycle being almost every two years. So I would say 2024 was the year of real 400 gig. 2025 and 2026, I would say, is more 800 gig. And I really see 1.6T coming into the picture because we don't have chips yet. Maybe in, what do you say, John? Late 2026?

John McCool (SVP and Special Advisor)

Yeah.

Jayshree Ullal (CEO)

Real production maybe in 2027.

John McCool (SVP and Special Advisor)

That's right.

Jayshree Ullal (CEO)

So there's a lot of talk and hype on it, just like I remember talk and hype on 400 gig five years ago. But I think realistically, you're going to see a long runway for 400 and 800 gig. Now, as you get into 1.6T, part of the reason I think it's going to be measured and thoughtful is many of our customers are still awaiting their own AI accelerators or NVIDIA GPUs, which with liquid cooling, that would actually push that kind of bandwidth. So new GPUs will require new bandwidth, and that's going to push it out a year or two.

David Vogt (Analyst)

Great. Thank you very much.

Jayshree Ullal (CEO)

Thank you, David.

Operator (participant)

Our next question comes from the line of Karl Ackerman with BNP Paribas. Please go ahead.

Karl Ackerman (Analyst)

Yes, thank you. Could you discuss the outlook for services relative to your outlook for March in the full year? I ask because services grew 40% year over year, and your deferred revenue balance is up another $250 million or so sequentially and nearly $2.8 billion. So the outlook for that would be very helpful. Thank you.

Chantelle Breithaupt (CFO)

Yeah. We don't usually guide the piece parts of that. So all I would say is that the thing to keep in mind with services is a little bit of timing in the sense of catching up to product, especially kind of post-COVID. So I would kind of take the trending you see over the last few years, and that's probably your best guide looking forward. We don't guide the piece parts. Thank you.

Jayshree Ullal (CEO)

Thanks, Karl.

Operator (participant)

Our next question will come from the line of Sebastien Naji with William Blair. Please go ahead.

Sebastien Naji (Analyst)

Yeah, thanks for taking the question. And you talked about this a little bit, but there's been a lot of discussion over the last few months between the general-purpose GPU clusters from NVIDIA and then the custom ASIC solutions from some of your popular customers. I guess just in your view, over the longer term, does Arista's opportunity differ across these two chip types, and is there one approach that would maybe pull in more Arista versus the other?

Jayshree Ullal (CEO)

Yeah, no, absolutely. I think I've always said this. You guys often spoke about NVIDIA as a competitor, and I don't see it that way. I see that. Thank you, NVIDIA. Thank you, Jensen, for the GPUs because that gives us an opportunity to connect to them. And that's been a predominant market for us. As we move forward, we see not only that we connect to them, that we can connect to AMD GPUs and built-in in-house AI accelerators. So a lot of them are in active development or in early stages. NVIDIA is the dominant market shareholder with probably 80%-90%. But if you ask me to guess what it would look like two, three years from now, I think it could be 50/50. So Arista could be the scale-out network for all types of accelerators. We'll be GPU agnostic.

I think there'll be less opportunity to bundle by specific vendors and more opportunity for customers to choose best of breed.

Rudolph Araujo (Head of Investor Relations)

This concludes Arista Networks' fourth quarter 2024 earnings call. We have posted a presentation that provides additional information on our results, which you can access on the Investor section of our website. Thank you for joining us today and for your interest in Arista.

Operator (participant)

Thank you for joining us, ladies and gentlemen. This concludes today's call, and you may now disconnect.

View Q4 2024 Earnings Summary