Jayshree Ullal

Thank you, Liz, and thank you, everyone, for joining us this afternoon for our third-quarter 2024 earnings call. We delivered revenues of $1.81 billion for the quarter, with a record non-GAAP earnings per share of $2.40. Services and software support renewals contributed strongly at approximately 17.6% of revenue. Our non-GAAP gross margin of 64.6% was influenced by both pressure from cloud titan customer pricing, offset by favorable enterprise margin and supply chain hygiene. International contribution for the quarter registered an approximately 18% with the Americas very strong at 82%.
Clearly, Q3 2024 had a lot of bright spots in the quarter and are encouraged by the strength and momentum of the company. At the recent tenth anniversary in June and 2024 celebration and vision event, we covered a lot of ground on what we would have otherwise said in an Analyst Day.
So today, I'd like to briefly expand on our Arista 2.0 plans for 2025. We believe that networks are emerging at the epicenter of mission-critical transactions and our Arista 2.0 strategy is resonating well with customers. We are, we believe, the only pure-play network innovator for the next decade.
Our modern networking platforms are foundational for transformation from silos to centers of data. This can be a data center a campus center, a WAN center or an AI center. At the heart of this is our state-oriented public subscribed network data lake EOS software stack for multimodal data sets. One simply cannot learn without having access to all this data. So it is all about the data.
We provide customers the state foundation for data for AI and machine learning without which AI and ML would just be buzzwords. Arista is well positioned with the right network architecture for client to campus, data center, cloud and AI networking. Three principles guide us and differentiate us in bringing this data-driven networking.
Number one, best-in-class, highly available proactive products with resilience and hitless upgrade built in at multiple levels; two, zero-touch automation and telemetry with predictive client to cloud one-click operations with that granular visibility that relies less on number three, prescriptive insights for deeper AI for networking delivering AIOps and algorithms for security, availability and root cause analysis. Networking for AI is gaining a lot of traction as we move from trials in 2023 to more pilots in 2024, collecting to thousands of GPUs, and we expect more production in 2025 and 2026.
In all vernacular, Arista AI centers are made up of both the back-end clusters and front-end networks. AI traffic differs greatly from cloud workloads in terms of diversity, duration, and size of flow. The fidelity of AI traffic flows with the slowest flow matters and one slow flow could slow down the entire job completion time is a crucial factor in networking. Our AI centers connect seamlessly from the back end to the front end of compute, storage, WAN, and classic cloud networks. Investor emerging as a pioneer and scale-out Ethernet accelerated networking for large-scale training and AI workloads.
Our new Etherlink portfolio with wire speed 800-gig throughput and non-blocking performance, scales from single tier to session two tier networks for over 100,000 GPUs, potentially even 1 million AI accelerators with multiple tiers. Our accelerated AI networking portfolio consists of three families with over 20 switching products and not just one point switch.
At the recent OCP in mid-October 2024, we officially launched a very unique platform that distributed Etherlink7700 to build two tier networks for up to 10,000 GPU clusters. The 77R4 DES platform was developed in close collaboration with Meta. And why it may physically look like and be cable like a 2-tier leaf spine network, DES provides a single-stage forwarding with highly efficient spine fabric, eliminating the need for tuning and encouraging fast failover for large AI accelerator-based clusters.
It complements our Arista flagship 7800 AI spine for the ultimate scale with differentiated fare and fully scheduled cell spraying architecture with a virtual output curing fabric saving valuable AI processor resources and improving job completion time. I would like to now invite John McCool, our Chief Platform Officer, to describe our 2024 platform and supply chain innovations after a challenging couple of years. John, over to you.

John McCool

Thank you, Jayshree. I'm pleased to report Arista 7700 R4 distributed Ecolink switch, the 7800 R4 Spine, along with the 7060 X6 AI leaf that we announced in June have entered into production providing our customers the broadest set of 800 gigabit per second Ethernet products for their AI networks.
Together with 800 gigabit per second parallel optics, our customers are able to connect to 400 gigabit per second GPUs to each port increasing the deployment density over current switching solutions. This broad range of Ethernet platforms allows our customers to optimize density and minimize tiers to best match the requirements of their AI work. As our customers continue with AI deployments, they're also preparing their front-end networks.
New AI clusters require new high-speed connections into the existing backbone. These new clusters also increased bandwidth on the backbone to access training data, capture snapshots and deliver results generated by the cluster.
This trend is providing increased demand for 7800 R3 400-gigabit solution. While the post-pandemic supply chain has returned to predictability, lead times for advanced semiconductors remain extended from pre-pandemic levels. To assure availability of high-performance switching silicon, we've increased our purchase commitments for these key components.
In addition, we will increase our on-hand inventory to respond to the rapid deployment of new AI networks and reduce overall lead times as we move into next year. Our supply chain team continues to work closely with planning to best align receipt of these purchases with expected customer delivery. Next-generation data centers integrating AI will contend with significant increases in power consumption while looking to double network performance.
Our tightly coupled electrical and mechanical design flow allows us to make system-level design trade-offs across domains to optimize our solutions. Our experience in co-design with the leading cloud companies provides insight into the variety of switch configurations required for these tightly coupled data center environments.
Finally, our development operating software with SDK integration, device diagnostics and data analysis supports a fast time to design and production with a focus on first-time results. These attributes give us confidence that we will continue to execute on our road map in this rapidly evolving AI networking segment. Back to you, Jayshree.

Jayshree Ullal

Thank you, John. And congrats on a very high performance here for you to you and your new executives, [Alex Rose, Mike Capas, Luke] and the entire team. You guys have really done a phenomenal job. Critical to the rapid adoption of AI networking is the Ultra Ethernet consortium specification expected imminently with Arista's key contributions as a founding member. The UEC ecosystem for AI has evolved to over 97 members.
In our view, Ethernet is the only long-term viable direction for open standard space AI networking. Arista is building holistic AI centers powered by our unparalleled superiority of US and the depth of automation and visibility software provided by CloudVision.
Arista EUS delivers dynamic methods using cluster load balancing for congestion control and smart system upgrades where the traffic for AI continues to flow in the midst of an upgrade. Arista continues to work with AI accelerators of all types, and we're agnostic to nice to bring advanced EOS visibility all the way down to the host.
Shifting to 2025 goals. As we discussed in our New York Stock Exchange event in June, our TAM has expanded to $70 billion in 2028. And you know we've experienced some pretty amazing growth years with 33.8% growth in '23 and 2024 appears to be heading at least to 18%, exceeding our prior predictions of 10% to 12%.
This is quite a jump in 2024, influenced by faster AI pilots. We are now projecting an annual growth of 15% to 17% and next year, translating to approximately $8 billion in 2025 revenue with a healthy expectation of operating margin.
Within that $8 billion revenue target, we are quite confident in achieving our campus and by back-end networking targets of $750 million each in 2025 that we set way back one or two years ago.
It's important to recognize though that the back end of AI will influence the front-end AI network and its ratios. This ratio can be anywhere from 30% to 100% and sometimes, we've seen it as high as 200% of the back-end network depending on the training requirements.
Our comprehensive AI center networking number is therefore likely to be double of our back-end target of $750 million, now aiming for approximately $1.5 billion in 2025. We will continue to aim for double-digit annual growth and a three-year CAGR forecast of teams in the foreseeable future of 2024 to 2026.
More details forthcoming from none other than our Chief Financial Officer, so over to you, Chantelle.

Chantelle Breithaupt

Thank you, Jayshree. Turning now to more detail on the financials. This analysis of our Q3 results and our guidance for Q4 fiscal year '24 is based on non-GAAP. Excludes all noncash stock-based compensation impacts, intangible asset amortization and other nonrecurring items. A full reconciliation of our selected GAAP to non-GAAP results and earnings release.
Total revenues reached $1.81 billion, marking a 20% year-over-year increase. This strong performance exceeded our guidance range of $1.72 billion to $1.75 billion. Services and subscription software contributed approximately 17% of revenues in the third quarter.
International revenues for the quarter came in $330.9 million or 18.3% of total revenue down from 18.7% last quarter. This quarter-over-quarter decrease reflects an increased contribution from domestic shipments to our cloud and enterprise customers.
Overall, gross margin in Q3 was 64.6%, above the upper range of our guidance of approximately 64%, down from 65.4% last quarter and up from 63.1% in Q3 prior year. This year-over-year improvement is driven by stronger enterprise margins and supply chain discipline in the current quarter.
Operating expenses in the quarter were $279.9 million or 15.5% of revenue, down from last quarter at $319.8 million. R&D spending came in at $177.5 million or 9.8% of revenue, down from $216.7 million last quarter. An item of note is that there were additional R&D-related expenses originally expected in Q3 that are now expected to materialize in the Q4 quarter.
R&D had increased low double-digit percentage versus Q3 in the prior year. Sales and marketing expense was $83.4 million or 4.6% of revenue, down slightly from last quarter. Our G&A costs came in at $19.1 million or 1.1% similar to last quarter. Our operating income for the quarter was $890.1 million or 49.1% of revenue. This was favorably impacted by the shift of R&D-related expenses from Q3 now anticipated in Q4 of this year.
Other income and expense for the quarter was a favorable $85.3 million, and our effective tax rate was 21.1%. This resulted in net income for the quarter of $769.1 million or (inaudible) of revenue. Our diluted share number was 320.5 shares resulting in a diluted earnings per share number for the quarter of $2.40, up 31.1% from the prior year. This, too, was favorably impacted by the shift in R&D-related expenses from Q3 to Q4. Now turning to the balance sheet.
Cash, cash equivalents and investments ended the quarter at approximately $7.4 billion. In the quarter, we repurchased $65.2 million of our common stock at an average price of $318.40 per share. Of the $1.2 billion repurchase program approved in May 2024,
$1 billion remains available for repurchase of future quarters. The actual timing and amount of future repurchases will be dependent upon market and business conditions, stock price and other factors. Turning to operating (inaudible) performance for the third quarter.
We generated approximately $1.2 billion of cash from operations in the period, reflecting strong earnings performance combined with favorable working capital results. DSOs came in at 57 days, down from 66 days in Q2, reflecting a strong collections quarter combined with contributions from linearity of billing.
Inventory turns were 1.3 times, up from 1.1% last quarter. Inventory decreased to $1.8 billion in the quarter, down from $1.9 billion in the prior period, reflecting a reduction in our raw materials inventory. Our purchase commitments and inventory at the end of the quarter totaled $4.1 billion, up from $4 billion at the end of Q2. We expect this number to continue to have some variability in future quarters as a reflection of demand for our new product introductions. Our total deferred revenue balance was $2.5 billion, up from $2.1 billion in Q2.
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 increased approximately $320 million versus last quarter. Fiscal 2024 continues to be a year of new product introductions, new customers and expanded use cases.
These trends have resulted in even contracts with customer-specific acceptance clauses and has and will continue to increase the variability of magnitude of our product deferred revenue balances. Accounts payable days were 42 days, down from 46 days in Q2, reflecting the timing of inner receipt payments.
Capital expenditures for the quarter were $7 million. In October, we began our initial construction work to build expanded facilities and (inaudible) expect to incur approximately $15 million during Q4 for this protest. Now turning to the fourth quarter.
Our guidance for the fourth quarter, which is based on non-GAAP results and excludes any noncash stock-based compensation impacts, intangible asset amortization and other nonrecurring items is as follows: revenues of approximately $1.85 billion to $1.9 billion, gross margin of approximately 63% to 64%; operating margin of approximately 44%. Our effective tax rate is expected to be approximately 21.5% on with diluted shares of approximately 321 million shares on a pre-split basis.
On the cash front, while we have experienced good increases in operating cash over the last couple of quarters, we anticipate an increase in working capital requirements in Q4. This is primarily driven by increased inventory in order to respond to the rapid deployment of AI networks and to reduce overall lead times as we move into 2025, mentioned in John's prepared remarks. We will continue our spending investment in R&D, go-to-market activities and scaling the company.
Additionally, in Q4 as part of our ongoing commitment to creating long-term value for our shareholders and enhancing the accessibility of our stock, we are pleased to announce that Arista's Board of Directors has approved a [4-for-1] stock split. This decision reflects our confidence in the continued growth and prospects of the company.
It's important to note that while the stock split increases the number of shares outstanding, it does not change the intrinsic value of the company nor does it impact our financial performance or strategy. The split is designed to make our stock more accessible and attractive to a wider range of investors, particularly retail investors, which we believe will ultimately support broader ownership and improve trading dynamics.
Transitioning now to fiscal year 2025. As Jayshree mentioned, we are putting revenue growth of 15% to 17%. The expected revenue mix is forecasted to have an increased weighting of (inaudible), placing the gross margin outlook at 60% to 62% and operating margin at approximately 43% to 44%.
Our commitment remains to continue to invest in R&D, go-to-market and the scaling of the company as we forecast to reach approximately $8 billion in revenue in 2025. We reiterate our double-digit growth forecast in the foreseeable future, and a three-year revenue CAGR goal of mid-teens for fiscal year '24 through '26. We are excited by the current and future opportunities to serve our customers as the pure-play networking innovation company and to deliver strong returns to our shareholders.
I will now turn the call back to Liz. Liz?

Liz Stine

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.

Question and Answer Session

Operator

(Operator Instructions) Samik Chatterjee, JPMorgan.

Samik Chatterjee

Hey, thanks for taking my question. A strong set of results, but if I can ask one on the guidance, if you don't mind. Jayshree, you're guiding here to the $750 million of TI target that you had issued previously and you're also guiding to meet your campus revenue target.
So if I take those two into account, it does imply that the ex AI and ex campus business is only growing single digits next year. This is on the yields of coming through a double-digit year in 2024 where you comped backlog digestion in 2023. So just maybe have parse through that as to why there's a significant (inaudible) in the non-AI noncampus business implied in the numbers?
And what maybe is driving that -- in your expectations, what's driving that outlook? Thank you.

Jayshree Ullal

Thank you, Samik. As you know, our visibility only extends to roughly about six months, right? So we don't want to get ahead of ourselves on how much better we can do and that's how we started 24 either, and we were pleasantly surprised with the faster acceleration of AI pilots. So we definitely see that our large cloud customers are continuing to refresh on the cloud, but are pivoting very aggressively to -- so it wouldn't surprise me if we grow faster in AI and faster in campus in the new center markets and slower in our classic markets called that data center and cloud. And this is the best we can see right now.
It doesn't mean we couldn't do better or worse. But as far as our visibility goes, I think this represents a nice combination of all our different customer segments and all our different product sectors.

Samik Chatterjee

Thank you.

Operator

Antoine Chkaiban, New Street Research.

Antoine Chkaiban

Hi, thank you very much for taking my question of. Can you maybe provide an update on the four major AI trials that you gave in the past? How are things progressing versus your expectations as of 90 days ago. And when do you expect the move to production to happen? And what scale are we talking about?

Jayshree Ullal

Yes. No, thank you. That's a good question. Arista now believes we're actually five out of five, not four out of five. We are progressing very well in four out of the five clusters. Three of the customers are moving from trials to pilots this year, inspecting those three to become 50,000 to 100,000 GPU clusters in 2025. We're also pleased with the new Ethernet trial in 2024 with our fifth customer.
This customer was historically very, very InfiniBand driven. And we are now moving in that particular fifth customer, we are largely in a trial mode in 2024, and we hope to go to pilots and production. There is one customer who -- so three are going well. One is starting. The fifth customer is moving slower than we expected.
They may get back on their feet. In 2025, they're awaiting new GPUs, and they've got some challenges on power cooling, et cetera. So three, I would give an A. The fourth one, we're really glad we won, and we're getting started and the fifth one, I'd say, steady state, not quite as great as we would expect -- have expected them to be.

Antoine Chkaiban

Thanks for the color.

Operator

Tal Liani, Bank of America.

Tal Liani

Hi, guys. NVIDIA in the last quarter the launch of the Spectrum X, it shows that in data center switching their market share went from like 4% to 15%. Does it mean that you're seeing increased competition from NVIDIA? And is it competing with you on the same spot? Or is it more competing with white boxes? And the second question is about white boxes.
What is the outlook for white box participation in Gen AI? Is it going to be higher or lower than in front-end data centers?

Jayshree Ullal

Okay. Thanks, Tal which question do you want me to answer?

Tal Liani

Let's go with NVIDIA. Give me the gift of --

Jayshree Ullal

Somebody else may ask the question anyway. So you'll get your answer. But just to answer your question on NVIDIA. First of all, we view NVIDIA as a good partner. If we didn't have the ability to connect to their GPUs, we wouldn't have all this AI networking demand.
So thank you, NVIDIA -- thank you, Jensen, for the partnership. Now as you know, NVIDIA sells the full stack and most of the time, it's with InfiniBand, and with the (inaudible) acquisition, they do have some Ethernet capability. We personally do not run into the Ethernet capability very much. We run into it, maybe one or two customers. And so generally speaking, Arista has looked upon as the expert there.
We have a full portfolio. We have full software. And whether it's the large scale-out ethernet working customers like the Titans or even the smaller enterprises, we're seeing a lot of smaller GPU plus of enterprise, Arista is looked upon as the expert there. But that's not to say we're going to win 100%. We welcome NVIDIA as a partner on the GP and a fierce competitor, and we look to compete with them on the Ethernet switching.

Operator

Simon Leopold, Raymond James

Simon Leopold

I'll tag team with Tal. So we'll partner once again here. I do want to look at this competition or competitive landscape broadly in that what I'm trying to understand is how it may be changing with the advent of AI. So not just hearing from you about white box, but also competitors like Cisco and Juniper and Nokia. So really an update on the competitive landscape would be helpful.

Jayshree Ullal

Thank you, Simon. That's a nice broad question. So since you asked me specifically about AI as opposed to cloud, let me parse this problem into two halves, the back end and the front end, right? At the back end, we're natively connecting to GPUs. And there can be many times, we just don't see it because somebody just muddles it in the GPU in particular, NVIDIA.
And you may remember a year ago, I was saying we're outside looking in because most of the bundling is happening within (inaudible), I would expect on the back end, any share Arista get, including that $750 million is incremental. It's brand new to us. We were never there before.
So we'll take all we can get, but we are not claiming to be a market leader there. We're, in fact, claiming that there are many incumbents there with InfiniBand and smaller versions of Ethernet that Arista is looking to gain more credibility and experience and become the gold standard for the back end.
On the front end, in many ways, we are viewed as the gold standard to competitively. It's a much more complex network. You have to build a leaf-spine architecture. John alluded to this, there's a tremendous amount of scale with L2, L3, EVPN, VXLAN, visibility, telemetry, automation routing at scale, encryption at scale. And this, what I would call accelerated networking portfolio complements NVIDIA's accelerated compute portfolio.
And compared to all the peers you mentioned, we have the absolute best portfolio of 20 switches and three families and the capability and the competitive differentiation is bar none. In fact, I am specifically aware of a couple of situations where the AI applications aren't even running on some of the industry peers you talked about, and they want to swap theirs for ours.
So feeling extremely bullish with the 7,800 flagship product, the newly introduced 700 that we worked closely with Meta, the 7060, this product line running today mostly at 400 gig because a lot of the (inaudible) the ecosystem isn't there for 800. But moving forward into 800, this is why John and the team are building the supply chain to get ready for it. So competitively, I would say, we're doing extremely well in the front end.
-- and it's incremental on the back end. So -- and overall, I would classify our performance in AI coming from being a now 12 years ago to where we are today (inaudible)

Simon Leopold

Thank you.

Jayshree Ullal

Thank you, Simon.

Operator

Ben Reitzes, Melius Research.

Ben Reitzes

I wanted to ask a little bit more about the $750 million in AI for next year. Has your visibility on that improved over the last few months. I wanted to reconcile your comment around the fifth customer not going slower than expected. And it sounds like you're now in 5 on 5, but wondering if that fifth customer going slower is limiting upside or limiting your visibility there? Or has it actually improved and it's gotten more conservative over the last --

Jayshree Ullal

Somebody has to bring a conservative end, but I think we're being realistic. So I think you said it right. I think on three out of the five, we have good visibility, at least for the next 6 months, maybe even 12, John, what do you think?

John McCool

Yes.

Jayshree Ullal

On the fourth one, we are in early trials, we got improving to do. So let's see, but we're not looking for 2025 to be the bang up year on the fourth one. It's probably 2026. And on the fifth one, we're a little bit stalled, which may be why we're being careful about predicting how they'll do. They may step in nicely in the second half of '25, in which case, we'll let you know.
But if they don't, we're still feeling good about our guide for '25. Is that right, (inaudible)

Chantelle Breithaupt

I would totally agree. It's a good question, Ben. But I think out of the five the way Jayshree categorized them, I would completely agree.

Ben Reitzes

Okay. Thanks.

Operator

Karl Ackerman, BNP Paribas.

Karl Ackerman

Thank you. Jayshree, could you discuss whether the programs are engaged with on hyperscalers? Will it be deploying your new EtonLink switches and AI spine products on 800-gig ports. In other words, have these pilots or trials been on 400 gig and production beyond 800 gig. And I guess if so, what's the right way to think about the hardware mix of sales of 800 gig in '25?

Jayshree Ullal

Yes. That's a good question. I mean, just going back to again, it was always hard to tell that 100 and 400 because somebody can take their 400 and break it into breakouts of 100 -- so I would say today, if you ask John and I, a majority of the trials and pilots are on 400 because people are still waiting for the ecosystem at 800, including the NIC and the UEC and the packet spring capabilities, et cetera. So while we're in some early trials on 800, majority of 400. Majority of 2024 is 400 gig.
I expect as we go into 25, we will see a better split between 400 and 800.

Karl Ackerman

Thank you.

Jayshree Ullal

Thank you, Karl.

Operator

Ryan Koontz, Needham & Co.

Ryan Koontz

Great. Thanks for the question. I was hoping we could touch base on your campus opportunities a bit. Where are you seeing the most traction in terms of your applications? Is this primarily from your strength in core moving big bits around the campus core? Or are you seeing WiFi? Can you maybe just update us on the campus applications and verticals you're seeing the most traction in?

Jayshree Ullal

Yes. Yes. Ryan, let me try and step back and say -- tell you that our enterprise opportunity has never been stronger. As a pure-play innovator, we are getting invited more and more into enterprise deals even though sometimes we don't often have the sales coverage for it. And what I mean by that is, I think Arista is being sought for a network design that doesn't have five operating systems and different silos and not just the get code.
And there's an awful lot of competitive fatigue and add to the fact that there's an awful lot of consolidation going on and a lot of our peers in the industry are looking at other things, whether it's observability or being other products together. So our enterprise opportunity now, we don't just characterize as data center. There's data center, there's campus center, there's WAN center and of course, there's a little bit of AI in there, too. So now let me address your (inaudible) question more specifically. Clearly, one of the first place as everybody went on (inaudible) is the universal spine.
They go, oh, okay, I can have the same spine from my data center and campus. This is so cool. So that activity has already started, and a big part of our $750 million projection comes from the confidence that they've already put in a platform and a foundation to get ready for more spine. Then if Kumar (inaudible) way here, he'd say, but Jayshree, you need to measure the edge ports, which is the power Ethernet, the wired and the WiFi. And this is super important, (inaudible) or laughing.
Yes. And so he would say, you got to get that right. And so number one, we're in the spine; two, we're making stronger progress on the wired. Our weakest partly because we are data center folks, and we're still learning how to sell radio is the WiFi that we plan to fix that, and this is where the extra coverage will come in. So I would say more of our strength is coming into wired and spine.
We are doing very well in pockets of WiFi, but we need to do better. Chantelle, you want to add something? .

Chantelle Breithaupt

Just to take the second part, I think you were asking about some of the verticals in your question. I just wanted to add some of the verticals. I think where we're seeing some strength data center and campus, I would say financials, health care, media, retail, (inaudible)

Jayshree Ullal

Fed and (inaudible), that's a good one. This is historically an area we have not paid attention to the federal market we're getting very serious about, including setting up its own subsidiary. So Chantelle, you've been a huge part of pushing us there. So thank you.

Ryan Koontz

Thanks so much.

Operator

Amit Daryanani, Evercore.

Amit Daryanani

Good afternoon. Thanks for taking my question. I guess I'm hoping you could spend some time on the sizable acceleration we're seeing both on your total deferred number, but also the product deferred number is going up pretty dramatically. Jayshree, historically, when product default goes up in such a dramatic manner, you actually end up with really good revenue (inaudible) in the out years, and you're guiding for revenue that you decelerate in '25. Maybe just help me connect like what's the delta, why product different what makes the acceleration that we historically has.

Jayshree Ullal

I'm going to let Chantelle the expert answer the question, but I will say one line. Remember, in the case of those examples you're quoting, the trials were typically, I don't know, 6 to 12 months this can be multiple years and can take a lot longer to manifest. It may not all happen in 2025. Over to you, Chantelle.

Chantelle Breithaupt

I think, yes. So thank you, Jayshree. So part of it is the type of use case, the type of customers, the mix of product that goes in there. they all have bespoke time frames, Jayshree referred to, you're starting to see those lengthen. And the other thing, too, is that this is what we know now as you move through every quarter, there are deferred in and out.
So this is what we know at this time. And it's a mix of the variables that we told you before. And then as we move through '25, we'll continue to update.

Amit Daryanani

Okay. Got it.

Operator

Meta Marshall, Morgan Stanley.

Meta Marshall

Great. Thanks. Jayshree, I just wanted to get a sense of, clearly, you keep -- clearly have these four main trials and have added a fifth. But just how are you thinking about adding other either Tier 2 opportunities or sovereigns or just some of these other customers that are investing heavily in AI and how do you see those opportunities developing those for Arista?

Jayshree Ullal

This is a good question. So we're not saying these are the (inaudible). But these are the five we predict can go to 100,000 GPUs and more. That's the way to look at this. So there are largest AI titans, if you will.
And they can be in the cloud, hyperscaler Titan group, they could be in the Tier 2 as well, by the way, very rarely would they be in a classic enterprise. By the way, we do have at least 10 to 15 trials going on in the classic enterprise too, but they're much smaller GPU counts, so we don't talk about it.
So we're folding on the big five to the point that they really skew our numbers and they're very important to establish our our beachhead, our innovation and our market share in AI, but there's definitely more going on. In terms of specifically your question on Tier 2 and will there be more there will be more, but these are the five we see in that category and they're spread across both the Tier 1 Titan Cloud as well as the Tier 2.

Meta Marshall

Thank you.

Operator

Sebastien Naji, William Blair.

Sebastien Naji

Yes, good evening. Thanks for taking the question. Just specifically on the Ethernet or Etalin portfolio, could you maybe rank order or comment on what you see as the opportunity across each of the three families a single-tier lease spine and then the tolling switch as we're going into 2025 and beyond.

Jayshree Ullal

I'll take a crack at it, but John helped me out here because this is clearly a guestimate. It's probably one we should say no comment, but we'll try to give you color. On (inaudible), I would say the fixed 7060 switches in terms of units are very popular because it's a single switch. It's one our customers are familiar with. It's based on an intense partnership with Broadcom.
So we've done Tomahawk 1, 2, 3, 4, and here we are on 5, right? So I would say, volume-wise, that's the big one. Going into the other extreme, the 7800 in volume may be smaller, but in dollars, is extremely strategic, and this is where we feel competitively again, working with our partners in Broadcom with the Jericho and (inaudible) family. This is just -- what would you say, John, a real flagship, right? In dollars, that's the stealer, if you will.
And then the 7,700 is the best of both worlds that gives you all the capabilities of the 7,800 in a mini configuration up to 10,000 GPUs. It's brand new. So -- but I think it's going to -- and competitively, there's no peer for this. Nobody else does this, but us with a scheduled fabric in a single stage. We did this in a very close collaboration, John, with Meta, right?
So you guys have been working together, John, for 18 months, two years, I would say. So I think we know less about how to qualify that, but it could be very promising and it could be a fast accelerator in the next couple of years.

John McCool

Yes, 7,700 people are interested in the very large scale is attractive for the 700. Between the 7060 and the 7800, we do see people that are optimizing a mix of both of those products in the same deployment so they can get the minimum number of tiers but have the maximum amount of GPUs that fit their use case. So we do see a lot of tailoring now around the size of the deployments based on how many GPUs they want to deploy their data center.

Jayshree Ullal

Yes, that's a really good point. And then suddenly, they'll go, okay, I want to go from a four-way rates to an eight way. And then suddenly, you have to add more line cards in your 7,800 and come running to you for more supply chain.

Sebastien Naji

Sure. Great. Thank you, both.

Operator

Aaron Rakers, Wells Fargo.

Aaron Rakers

Yes, thanks for taking the question. I wanted to segue off the competitive landscape and just ask you about when I look at your 2025 outlook as well as the midterm model that you provided, it looks like you're making some assumptions of some margin declines. I'm curious of what's underlying those expectations of gross margin declines? Is it mix of customers? Do you expect multiple 10% plus customers in 2025? Just any help on what's factored into that -- those margin expectations?

Chantelle Breithaupt

I would say, absolutely, and the outlook that you referred to, it is customer mix. We're expecting John to continue the great supply chain discipline that he's been doing with his team. So it is a BIC comment only. And as for the 10% customers, I would say the one dynamic, maybe it's a bit cheeky to say it as the denominator gets bigger, that gets a bit tougher. So we'll see as we go in the out years.
But right now, we'll just keep to the the ones that we currently talk about, and we'll see how that goes from '25 and '26.

Jayshree Ullal

It's going to get harder and harder to have 10 customers. So I believe M&M will still be that in 2025, but I don't anticipate there's any others at the moment.

Liz Stine

Operator, we have time for one last question.

Operator

Atif Malik, Citigroup.

Atif Malik

Hi. Thank you for taking my question. Jayshree, some the recent conferences, you've talked about every dollar spent on back end is at least 2 times on the front end. What signs are you looking for to see the lift from AI on the front end or classic cloud from the pressure on the bandwidth.

Jayshree Ullal

Yes. Listen, I think it all depends on Atif their approach to AI. If they just want to build a back-end cluster and prove something out, they just look for the highest job training completion and intense training models. And it's a very narrow use case. But what we're starting to see more and more like I said, is for every dollar spent in the back end, you could spend 30% more, 100% more, and we've even seen a 200% more scenario, which is why $750 million will carry over to, we believe, next year, another $750 million on front-end traffic that will include AI, but it will include other things as well.
It's (inaudible) be unique to AI. So I wouldn't be surprised if that number is anywhere between 30% and 100%, so the average is 1-0%., which is 2 times back and over. So feeling pretty good about that. Don't know how to exactly count that as (inaudible), which is why I qualify it by saying increasingly, if you start having inference, training, front end, storage, (inaudible) classic cloud all come together, the AI -- the pure AI number becomes difficult to track.

Atif Malik

Thanks so much.

Liz Stine

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

Operator

Thank you for joining. Ladies and gentlemen, this concludes today's call. You may now disconnect.