Arista Networks Inc (ANET) 2023 Q1 法說會逐字稿

Welcome to the First Quarter 2023 Arista Networks Financial Results Earnings Conference Call.

(Operator Instructions)

As a reminder, this conference is being recorded and will be available for replay from the Investor Relations section at the Arista website following this call. Ms. Liz Stine, Arista's Director of Investor Relations. You may begin.

Thank you, operator. Good afternoon, everyone, and thank you for joining us.

With me on today's call are Jayshree Ullal, Arista Networks' President and Chief Executive Officer; and Ita Brennan, Arista's Chief Financial Officer. This afternoon, Arista Networks issued a press release announcing the results for its fiscal first quarter ending March 31, 2023. If you would like a copy of the release, you can access it online at 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 second quarter of the 2023 fiscal year, longer-term financial outlook for 2023 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 discussed 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.

Thank you, Liz, and happy Monday, everyone, and a happy month of May.

We delivered revenues of $1.35 billion for the quarter with a non-GAAP earnings per share of $1.43. Services and software support renewals contributed approximately 13.5% of the revenue. Our non-GAAP gross margins of 60.3% was influenced by supply chain overheads, and cloud tightening concentration. We expect our gross margins to improve every quarter throughout this year. International contribution registered at 17.5% with the Americas strong at 82.5% for the quarter. While we will shift to reporting our vertical segments on an annual basis, I would like to share some overall trends we've seen.

We do expect Cloud Titans will moderate compared to our 2022 triple-digit growth, while enterprise is likely to be more steady state. It is evident that our lead times are improving, our visibility to customer forecast, therefore, are now beyond 6 months, or now below 6 months, I should say, and they're shrinking. Despite macro uncertainty, we endorsed consensus of 26% annual growth to approximately $5.5 billion revenue in 2023.

On the product side, we made many exciting Q1 announcements. At OFC 2023, we introduced our vision for linear drive optics for intra and inter data center connectivity at 800 gig and beyond. This was a highlight for both Arista and the optical industry at large, delivering the promise of low power and improved price performance for demanding AI workloads. Speaking of AI, demand base to avoid idle state an expensive and large AI process at clusters requires that specialized AI network. Key characteristics include wire rate and loss less delivery of large and synchronized burst of data at 400- to 800-gig speeds. Today, the combination of RDMA mix, RDMA stands for remote direct memory access, and ROCE, which is our RDMA over converged Ethernet, along with the switches allows Ethernet to become that predictable transport network. Ethernet, of course, brings familiarity, great economics, massive installed base, standards with industry-wide interoperability and many merchant silicon options.

This is supporting compute and data-intensive workloads based on generative AI inference and large language model training applications. Arista's cloud customers are resonating with our AI and switching strategy for platforms. Presently, we are in the midst of trials, leading to production deployments this year in 2023. We expect AI networking to become meaningful throughout the years and through the decade ahead. In Q1 2023, Arista also formalized our new entry into the wide area network with our WAN routing system. Our enterprise class wiring platform is based on carrier and cloud-neutral transit with CloudVision Pathfinder service. Not surprisingly, we support Arista's EOS operating system stack, delivering that operational model for Network as a Service and Wide Area as a Service.

We are targeting mission-critical enterprises where high volume and encrypted traffic matter in a modern WAN. Arista has partnered with Equinix to develop and deploy the WAN routing system and WAN routing will be included as part of our network adjacency category. In the non-cloud category, we have registered a solid number of million-dollar customers, which is a direct result of our momentum in the enterprise and campus throughout the past year. Let me illustrate with a few customer wins. The first use case is an international government grid. The customer's objective was to detect illegal activities such as money laundering, terrorism scans and other criminal behavior in real time by collecting and analyzing data. Arista's data-driven AI clusters optimizes network assurance for mission-critical AI and ML workloads using advanced features like microbursts and Fan In congestion management, ultra-deep packet buffer memory with latency analyzer, provides real-time telemetry, visibility, automation and dynamic controls for their AI and ML data centers, all based on open standard ethernet.

Our second win continues on the international team and highlights our ever-growing strength in the education vertical, where Arista's proposal for edge campus platforms ranging from power over Ethernet switches, wireless access points and automation was a decision factor. We leverage CloudVision Q cognitive unified edge, coupled with Arista validated designs as an automation framework across multiple distributed locations bringing unmatched flow visibility.

The next win is in the U.S. financial sector. This customer had grown organically and inorganically through acquisitions and was looking to modernize their entire infrastructure. Moving their data closer to the cloud to enable a hybrid cloud architecture. This design included multiple greenfield data centers hosted in Equinix requiring Active/Active 400-gigabit Ethernet spine securely encrypted data center interconnect and internet connectivity at each site. For a smooth upgrade in their campus environment without disruption to their end users, Arista's SSU, or smart systems upgrade feature played a prominent role. We also help them build a digital twin of their environment, modeling their designs for automation.

The next customer highlights healthcare as a critical win for network monitoring and security analysis tools at their remote data center facilities. This holistic view of port mirroring session for traffic analysis from all the remote data centers with a superior approach. Arista centralized DMF, DANZ monitoring fabric, was better than disparate and expensive tools at each remote location. Our final customer win was looking for real-time in-house video streaming and editing capabilities. Video would be stored on their storage arrays, which could be connected at 100 gigabit ethernet and then accessed and rendered by the clients, be their PCs or MAX with 25-gigabit Ethernet.

Arista's core strength in the media vertical comes from its deep buffer virtual output queuing architecture with our R3 platforms. The simplicity, scalability and flexibility aligns -- shows this elegant design and highlights our strength in the media and entertainment vertical.

So as you can see, this is a recurring theme in all our customer wins, where Arista is deploying innovative solutions based on a consistent architecture, allowing each and every customer to modernize their network with the power of our platform.

And with that, I'd like to hand to Ita, our CFO, for financial metrics.

Thanks, Jayshree, and good afternoon.

We will now present our Q1 results and our guidance for Q2 '23 is based on non-GAAP, excludes all noncash, stock-based compensation impacts, certain acquisition-related charges and other nonrecurring items. A full reconciliation of our selected GAAP to non-GAAP results is provided in our earnings release. Total revenues in Q1 were $1.351 billion, up 54% year-over-year and well above the upper end of our guidance, $1.275 billion to $1.325 billion. We continue to experience improvements in component supply in the quarter, supporting more consistent levels of manufacturing output and some improvements in lead time. Services and subscription software contributed approximately 13.5% of revenue for the first quarter, down to 15.8% in Q4. This largely reflected accelerated growth in product revenues, while services and software continue to grow on a more consistent basis.

International revenues for the quarter came in at $236 million or 17.5% of total revenue, down from 23.5% last quarter. This quarter-over-quarter reduction largely reflected an unusually high contribution from our EMEA and region customers in the fourth quarter. Overall, we continue to see outsized growth in the U.S., largely due to ongoing domestic strength from our Cloud Titans customers.

Overall gross margin in Q1 was 60.3% in line with our guidance of approximately 60%. We continue to recognize incremental supply chain costs in the period, combined with a healthy cloud mix. Operating expenses for the quarter were $257.5 million or 19.1% of revenue, up from last quarter at $235.3 million. R&D spending came in at $164.8 million or 12.2% of revenue, up from $153.2 million last quarter. This primarily reflected increased headcount and new product introduction costs in the period.

Sales and marketing expense was $75.9 million or 5.6% of revenue compared to $67.4 million last quarter, with increased headcount costs and higher variable compensation expenses. Our G&A cost came in at $16.8 million or 1.2% of revenue, consistent with last quarter. Our operating income for the quarter was $556.8 million or 41.2% of revenue. Other income and expense for the quarter was a favorable $17.7 million, and our effective tax rate was 21.2%. This resulted net income for the quarter of $452.5 million or 33.5% of revenue. Our diluted share number was 315.6 million shares, resulting in a diluted earnings per share number for the quarter of $1.43, up 70% from the prior year.

Now turning to the balance sheet. Cash, cash equivalents and investments ended the quarter at approximately $3.33 billion. In the quarter, we repurchased $82.3 million of our common stock at an average price of $111.9 per share. We've now repurchased $825.5 million worth or 7.8 million shares at an average price of $106 per share under our current $1 billion Board authorization. This leaves $174.5 million available to 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 flow for the first quarter.

We generated approximately $275 million of cash from operations in the period, reflecting strong earnings performance, partially offset by ongoing investments for working capital. DSOs came in at 57 days, down from 67 days in Q4 reflecting a strong collections quarter with good linearity of billings. Inventory turns were 1.3x, down from 1.6x last quarter. Inventory increased to $1.7 billion in the quarter, up from $1.3 billion in the prior period, reflecting the receipt of components from our purchase commitments and a slight increase in switch-related finished goods. Our purchase commitments at the end of the quarter were $2.9 billion, down from $3.7 billion at the end of Q4. We expect this number to continue to decline in future quarters as some component lead times improve, and we work to optimize our supply position.

Our total deferred revenue balance was $1.092 billion, up from $1.04 billion in Q4. The majority of the deferred revenue balance was 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 was flat to last quarter. Accounts payable days were 55 days, up from 43 days in Q4, reflecting the timing of inventory receipts and payments. Capital expenditures for the quarter were $5.6 million.

Now turning to our outlook for the second quarter and beyond. As we move through 2023, we expect to resolve the final (inaudible) in the supply chain, allowing for more consistent manufacturing output and improving lead times to our customers. We do, however, expect these reduced lead times alter the results in reduced visibility with customers no longer needing to make purchase decisions so far in advance of deployment.

In addition, we expect some moderation in customer spending, especially with our Cloud Titans customers following a year of accelerated demand in 2022. All of that being said, we believe customer engagements and current deployment across the business support the current consensus revenue growth rate in 2023 of approximately 26%. In terms of the quarterly trends, you should expect moderating year-over-year growth as the year progresses, with more difficult prior year comps. On the gross margin front, beginning in Q2, we expect to see some steady improvement as we consume pure broker (inaudible) and have the opportunity to optimize manufacturing output while maintaining a healthy contribution from our cloud customers.

Now turning to spending and investment. We continue to monitor the overall macro environment carefully. We'll prioritize our investments as we move through the year. This will include a focus on targeted hires in R&D and go to market as the team sees the opportunity to acquire talent. On the cash front, I will continue to focus on supply chain and working capital optimization. We should expect some continued growth in inventory on a quarter-by-quarter basis as we receive components from our purchase commitments. With all of this in the backdrop, our guidance for the second quarter, which is based on non-GAAP results and excludes any noncash stock-based compensation impacts and other nonrecurring items is as follows: Revenues of approximately $1.35 billion to $1.4 billion, gross margin of approximately 61%, operating margin at approximately 40%. Our effective tax rate is expected to be approximately 21.5%, with diluted shares of approximately 317 million shares.

I will now turn the call back to Liz. Liz?

Thank you, Ita. We will now move to the Q&A portion of the Arista earnings call.

(Operator Instructions)

Thank you for your understanding. Operator, take it away.

(Operator Instructions)

Your first question comes from the line of Aaron Rakers with Wells Fargo.

I'm just curious kind of the commentary around the hyperscale cloud as component lead times shrink, how would you characterize, if at all, the visibility in that vertical? And specifically, how maybe that's evolved or changed relative to, let's say, the commentary or the thoughts a quarter ago?

Yes, Aaron. I'll kick it off and maybe Anshul can help me. As you know, historically, visibility and the cloud sites, if you take out, if you subtract the last 2 years, which were largely supply chain related, was typically 2 quarters, right? And for a period of time last year and the year before, we were starting to get 4 quarters of visibility. As our lead times are improving, our visibility is also shrinking especially with that segment because they can make decisions closer to our lead times. So I would say our visibility has reduced from last year to this year by 2 quarters and in roughly 6 months.

Your next question comes from the line of Antoine Chkaiban with New Street Research.

So at the CMD, I think you provided an AI intensive network TAM of $2 billion, $3 billion in the next few years. And during last earnings, Broadcom said that our Ethernet switch ships deployed in AI was well over $200 million in '22 and the forecast that this could grow to well over $800 million in '23. So I imagine that, that would correspond to $4 billion, $5 billion in revenues. This is, therefore, already well above the time that you estimated, Am I missing anything? Or did the TAM expand considerably more than you were anticipating at the CMD?

Sure. This is Anshul. As you know there's a lot of talk about AI, and it's a very exciting topic in many ways. First, you have to separate out numbers at Broadcom (inaudible) versus where our customers will deploy systems, right? There's an offset of when they ship chips versus when they can ship systems and often by quarter, sometimes it could be a longer the year, right, given the lead times that are going on in the market.

Second, I think AI is still in infancy. I don't think we know really how big it will be. It's clearly on a very good trajectory. It will keep on growing. And there is a great opportunity for us for sure and we're doing very well with some of our top customers as Jayshree talked about in the primary script as well.

Yes. And just to add to what Anshul said, our forecast of $2 billion to $3 billion is more in the 2025 arena. Market analysts are already showing larger numbers than that, 2025 to 2027 arena. I think market analysts are already projecting, it's double that. And certainly, Broadcom is enthusiastically looking at their chip deployment. But again, as Anshul alluded, by the time Broadcom has a chip, a chip gets built into the system by us and then the system gets deployed by our cloud customers, it can be 1 to 2 years.

Your next question comes from the line of Samik Chatterjee with JPMorgan.

If you don't mind, can we just dive -- dig a bit deeper into the inventory number, I think getting a few questions on that. Obviously, a material step up in the inventory in the quarter particularly, as you mentioned, lead times for your products are now 6 months, do we sort of conclude that your lead times will take a step function down, if you have this much inventory at this point? And if your visibility into demand is starting to come in a bit, why sort of maybe help me think about why the inventory continues to sort of move up higher from here rather than sort of start to come down in line with your visibility into demand.

Samik, this is Ita. When you think about the purchase commitments that we have and that we made some time ago, right, we're going to have to continue to work through those as we go through the rest of this year particularly on some key components there were long lead times. We had to place commitment for kind of this year, well, earlier last year. in order to secure those -- that supply. So those components will continue to come in inventory and obviously will go out of inventory as well as we build products, et cetera. So we have a healthy deployment pipeline in front of us on the system side, but we are still going to gather components based on when those purchase commitments were made and the timing of those purchase commitments.

So if you look at the total inventory plus purchase commitments, it came down in excess of $400 million this quarter. So we'll continue to kind of work that down over time. But you will see the shift from purchase commitments into inventory for us and then of signals, we'll sell that inventory.

And Samik, just to add to, we're not managing the business at just-in-time inventory. As Ita said, we have 72-week lead time still on many of our components even with the supply chain improvements. So we have to plan ahead. And if we want to get products to our customers in 6 to 12 months, assume a position on the inventory and especially do so on common components where we feel confident that there is demand, and we will continue to fulfill that demand this year and next year.

Your next question comes from the line of David Vogt with UBS.

Great. I just want to follow up basically on the question on AI inventories and sort of revenue growth expectations for the second half. So if I'm hearing you correctly, it sounds like you think given tough comps and sort of spending patterns from the Titan Group is going to come down -- bring revenue growth down to about 15% in the second half on a year-over-year basis.

And yet, as you just mentioned, inventory still need to come down. So how do we square that with sort of the optimism in the marketplace that looks like you took your TAM up from about $40 billion to about $50 billion for data center and campus in the deck. I'm just trying to square the deceleration that you're talking about versus sort of the expanded TAM that you're also kind of highlighting in the deck.

David, I think, first of all, the TAM we took up was for 2027, not Q3, Q4 2023, just to be clear. And I think we continue to feel very optimistic about our long-term demand in enterprise, cloud and AI. So we shouldn't confuse the comps and difficulty of comparing Q3 2022 with Q3 2023 with our long-term demand and TAM, both are balance statements. But as you know, the cloud is a volatile market and the Titans will spend a lot one year and then spend a little less the other year as they're digesting it and deploying it. So if you look across multiple years, we're going to have the strong demand and do well.

Yes. And David, this is very consistent with what we talked about last quarter, right? I mean we're -- because of the comps and the pattern on the comps, we're going to grow quarter-by-quarter, growing each quarter consecutively. But you will see that deceleration just because of how last year's kind of revenue trended as well, right? So if there's anything new year. In fact, we probably took up the overall number a little bit to get to 26%.

Right. We said 25% in November, and now we're saying 26%.

Your next question comes from the line of Meta Marshall with Morgan Stanley.

Maybe just varying in kind of on the cloud titan vertical. You mentioned kind of reduced visibility, but just wanted to clarify, Had you seen any changes in orders or any push outs kind of within the quarter of orders or -- and within kind of your near-term guidance of orders that you thought were going to take place that are maybe getting pushed out?

Yes, Meta, I'll let Anshul answer the question, but I would say it's sort of a give and a take. Some things are getting pushed out and some are getting pulled, the silver lining is clearly AI. That's not getting pushed out, but some of the deployments of cloud regions are getting pushed out. Anshul, you want to add to that?

If I can add some more color to key areas that we've been tracking. I know we're only going to talk about AI. This AI market is backbone, which is what started the 400-gig cycle in the first place. Those deployments are progressing as expected as well. So that part is in steady state. And obviously, AI is growing compared to what we knew before.

Your next question comes from the line of Sebastien Naji with William Blair.

Just given this discussion around generative AI, maybe can you frame for us the advantages of Ethernet for building out these AI network fabrics and any metrics you might have that highlight these advantages versus something like InfiniBand?

Sure. I think I said this before, but I think the #1 advantage of Ethernet is the fact that you're building a standard space, multi-vendor, highly interoperable network, where everything from troubleshooting to familiarity when you're connecting to the GPU process is very well known. So from a best-in-grade horizontal approach, Ethernet can win every time and (inaudible) technology generally struggle. Having said that, the vertical approach that InfiniBand adopted for high-performance computing can be applied to GPU clusters as well. So I think it all depends on the customers' clusters and how large they are and the larger they become, the more it favors Ethernet.

Your next question comes from the line of Tal Liani with Bank of America.

I want to ask about non-cloud Titans, the other part. Last year, it grew about 14.5% and it was supposed to grow -- by your guidance, kind of supposed to grow much faster this year? What happened this quarter? And again, you don't provide exact numbers even qualitatively, what happens this quarter with non-cloud Titans, how is demand shaping up when it comes to orders, I'm trying to neutralize the supply chain issue.

Yes. No, good question. Enterprise demand is pretty strong and steady. In fact, I would go as far as saying the customer activity has been just as strong as last year. And some of these macro things we hear about, we are experiencing less of it perhaps because we are a small fish in big ocean, right? So that said, obviously, our revenue has a high component to sell tightened concentration in Q1. So the demand doesn't translate into direct revenue contribution in a specific quarter. But I think you will see a number far greater than the 15% through the year.

Your next question comes from the line of Michael Ng with Goldman Sachs.

It was encouraging to hear about the endorsement of consensus at 26% year-over-year growth. I was just wondering if you could talk about what you're assuming as it relates to AI production deployments because you did talk about that trial that was underway. And any other areas of optionality that you would call out perhaps the DIY to branded switches within web scale Cloud Titans. Any updates there would be helpful.

Sure, Michael. As we said, the 7800 is Arista's flagship AI platform. And we expect the better part of last year, maybe even the year before, Anshul, and you could correct me, doing a tremendous amount of simulation on how we work with GPU clusters and different types of stuff, network interface cards, the performance, the (inaudible) list, dealing the diversity traffic, the latency, the transaction. And we believe that this will be a critical year in seeing those trials come into production. So we do expect AI to be meaningful this year as opposed to not material the last couple of years. And we believe the 7800 will be the flagship product for that.

And if I can just answer your other indirect question. Your question was how are we doing at these (inaudible) boxes.

We ship blue boxes.

We said this before. I think we are maintaining status quo. We're doing very well with our customers. They're not afraid, and we don't believe the market is shifting back to white box at all. In fact, the core development efforts are even more intense than before. But I think a lot of these things we've achieved status quo. I think that's where the market stays for now.

Your next question comes from the line of Alex Henderson with Needham.

I've got a question here, I want to split into 2 pieces. The first one is as you're looking at market share in the AI arena, does the networking piece, gain share within the AI wallet budgets? And then second, I know that you've had a very significant share advantage in high speed. You've gained significant share from your competitors for every year that I can remember. And I guess the question is, will AI drive an acceleration in your share given your dominant experience so far in delivering it. So share within the CapEx wallet and then share within the AI market, too many questions.

Alex, let me go back to your high-speed acceleration question first, and then you talk about the market share in AI because we're still kind of grappling with what is the market for AI right now. In terms of high speed, I think we've now got some killer use cases for 400- and 800-gig with AI. So you will see our strength -- going from strength to strength with 100, 200 in some cases and now 400 and 800 with AI being backfill application driving our high-speed acceleration. We feel more confident in it now. Otherwise, you could argue what is the use case for 400- and 800-gig. So that makes us very positive.

Specific to AI wallet shift, quite honestly, the greatest component of AI today is the processers, that is 80%, maybe 90% of the spend and the applications, obviously, that go with that. So if they're vertically integrated, we may not see as much of it. But if customers choose the horizontal best of breed, absolutely get our share of wallet there.

Your next question comes from the line of Amit Daryanani with Evercore.

I guess I just want to go back to this reduced visibility that you're seeing with Cloud Titans. I guess, is it your sense that you just had extra long visibility at fourth quarter and now going back to 2 quarters, which is normal. Or do you think there's a risk that it actually ends up an outright pause at some point given these companies did have a really big spending cycle with you in the last 4 or 5 quarters already. So I'm just wondering, like is this a return to normalcy? Or do you think there's risk that we end up in a pause with 1 or both of them the way we've been in 2019.

This is Anshul. Our customers have been waiting for 2.5 years for this moment. So they can return back to normalcy. Supply chain is recovering. These customers follow component lead times very closely as well to a great extent. We're coming back to where we used to be pre-COVID levels, nothing different. Nothing more than that.

Your next question comes from the line of Fahad Najam, an independent analyst.

Jayshree, I wanted to ask you a question on Broadcom's recent introduction of the Jericho 3 AI chip, and it kind of reminds me of the time when they first introduced the Jericho 2 and 2C and you were the earliest adopters of that technology and that led to your significant gains in the leaf-spine architecture. So is the Jericho 3 a similar upgrade cycle? And should we think about the same advantages you guys are enjoying in this forthcoming cycle as you did in the previous cycle? Anything you can tell us in terms of the comparison.

So a good way to look at this market and the introduction of J3, J3 AI, 400 gig is not going to end quickly and suddenly get replaced with Jericho 3, 400-gig will go on for some time, customers will take time to make changes. And so especially when they don't need more bandwidth just yet. At the same time, you'll see a quick adoption of 800-gig technology and the complementary chip that was also announced with Tomahawk 5. Between Tomahawk 5 and Jericho3 AI, the AI teams (inaudible) these are quickly as the market can get them out there.

And you may have read some white papers that were also published along with the announcement, which showed that Jericho3 AI is scaled to very large clusters, we can scale to 4,000 GPUs quickly at 800-gig, and the cluster performs at 10% better throughput than InfiniBand. So that is why there's such needs for this technology out there, and everyone thanks us to get it. But just remember, the chips have just been announced. It takes time to get the chips, build systems, better customers, to work with the trials and then go to the volume production.

It's a very exciting multiyear journey, and we really value our partnership with Broadcom. But what you're seeing here is 100 gig for mainstream enterprises, 400 gig for the cloud and 800 gig and beyond for AI use cases.

Your next question comes from the line of Matt Niknam with Deutsche Bank.

Just to go back to the macro discussion. I'm just wondering, were there any regions, customer verticals where you maybe saw some greater-than-usual slowness or lengthening sales cycles, particularly later in the quarter.

Yes, Matt, I'd say that usually, we see a very strong activity in the month of March. But in Q1, we did see some seasonality in certain regions, especially internationally. And I don't know how -- 1 quarter, doesn't have (inaudible), but we're definitely watching this.

Okay. And has that changed at all in early April?

Too early to say. No. You mean has it changed in the sense it's improved in April. Is that your question?

That's right.

Yes, April is good so far.

Your next question comes from the line of Michael Genovese with Rosenblatt.

Just one question for me. So basically, when -- it's about timing on AI. And when do we think switching will inflect? And I guess maybe the actual question is, what's your outlook on 2024 cloud spending? I mean we've talked a lot about the next 6 months, but I went about 2024, how we be thinking about that right now?

Next month (inaudible) 2024 spending because we don't know. We don't have the visibility.

A little early for that yet, Mike.

But what about on the timing of switches? I mean, as you go through all of these GPUs or processing units now, training, all of these things, what do you think the timing for switching deployments will inflect positively?

You're asking specifically AI or Cloud Titans spending?

Well, specific -- I mean, AI is clearly happening now, but 80% to 90% of the spending is stuff that you don't do. When do you think that there will be a significant uptick in that percentage of switching deployments in AI cluster data centers?

Okay. Well, as I said, I think last year was the year of trials. This year, we'll see some production and it will certainly accelerate in '24 and '25 specific to AI. But again, that's a small spend relative to our larger cloud spend, where we'd like to see more visibility on how the cloud regions are getting built out, et cetera.

Your next question comes from the line of James Fish with Piper Sandler.

Nice quarter and (inaudible) as well, of course. Purchase commitment, I wanted to circle back there as well. it's moving down as you guys anticipated. But obviously, you guys are a much larger business than you were pre-pandemic. So I guess, how are you guys thinking about the level of normalcy of purchase commitments as we kind of work through this. And obviously, Ita, you talked about that we'll see sequential impacts to cash flow still on the inventory as we kind of convert that purchase commitment to inventory. Is that something that should reverse then in early 2024? And how should we kind of think about free cash flow conversions for this year then?

Yes. Look, I think if I have my way, the purchase commitment number will come down significantly over the next, I don't know, 12 to 18 month end because we don't need it once we start to see some of these component lead times come in. So we need to -- obviously to manage that. Some have long lead times that we do want to receive, and you will see that grow in inventory, some, we'll look to reposition, if we can. But obviously, there's a keen focus on kind of managing that number now. But the net of it is, I think you grow inventory through the year, it will consume some cash, and then it will flip in 2024, where we'll actually start to kind of generate more cash as we start to bring that inventory number down.

Do we ever go back to kind of where we were before? I think probably not. I mean we probably will carry a little bit more inventory and more buffers, having just gone through what we went through the last couple of years, but it should certainly get -- come down from where it is today.

And any thoughts on the free cash flow conversion for the year?

Yes. I think for this year, inventory is a consumer of cash. So it's probably -- it's hard to know exactly what that looks like, but I think every quarter, we'll increment that inventory balance as we go through the year, that will concern some cash. But I mean the P&L is highly cash positive with the guidance that we've put out. So I think we'll still be generating a healthy amount of cash but we will build inventory balance.

Your next question comes from the line of Ittai Kidron with Oppenheimer.

Ita, I wanted to dig into the comments on gross margin where you expect them to improve through the year? 2 questions there. Number one, what -- now the supply chain is getting better. What is it in the supply chain that's still expensive that's hurting you on the gross margin side? And how does that get mitigated? And second, the improvement that you anticipate, is that just a reflection of the mix, meaning cloud perhaps moderating to your point? Or is most of the improvement driven again by supply chain -- better pricing on the supply chain side.

Yes. I mean I think in Q1, we were still consuming broker parts and other parts that we had purchased prior, right? Because I mean, obviously, you have to prepare for the quarter, that should get better in Q2 and then even more so as we go through the rest of the year, where we'll stop consuming those legacy, if you like, broker parts that you have in the pipeline. So that will definitely help.

The other thing that's important is now that we don't have the soft start of the (inaudible) and stuff, we can focus on manufacturer, we can focus on driving manufacturing, driving efficiencies there, et cetera. So we should see some improvements come out of that as we go through the year. I'm not assuming a whole lot of a change in the mix of the business, maybe a little bit more, but not a lot, just because we have a deployment pipeline for cloud, right? That is the discussion that we're having about cloud is more when do they need to place the orders for the next deployment. Now that there isn't deployments in front of us right now, right? So I think that's an important distinction, right? So I think we are pretty happy with how the cloud business is kind of flat out through the rest of the year. And the question becomes when do they place new orders (inaudible) shorter lead times and when can we see what that closing orders will look like.

Your next question comes from the line of Ben Bollin with Cleveland Research.

Ita, I guess, more specific for you. Could you share any thoughts around OpEx with respect to R&D and sales and marketing? And how those have evolved through the year, given the visibility and supply and what you're seeing out there?

Yes. I think we talked about a little bit in the script, but when we are continuing to hire, and we are continuing to make some investments, and we'll continue to do that, certainly within the envelope as we think about the business for the year today, we will continue to do that. It probably doesn't grow quite as fast as the top line, we'll see. But we are continuing to increment that quarter-over-quarter, and you'll see us continue to grow those investments. But again, it's headcount around R&D and go-to-market and really looking for good talent and places where we win an opportunity, right? But we will continue to invest, just given the envelope of business that we have (inaudible).

Your next question comes from the line of Tim Long with Barclays.

Can I ask a 2-parter on the adjacencies. First, just curious on the campus side. Obviously, you guys have been growing nicely there. How do you think that tends to be more macro sensitive? So how do you view being able to grow the same level in that sector when there becomes more natural headwinds like we're starting to see.

And then on the routing side, I'm curious now that you're out with the full platform. Jayshree, if you -- or Anshul, if you could just give us a little view of kind of what feedback has been and what you think kind of the revenue path would look like for increased routing business?

Tim, I think on the campus side, the TAM is somewhere between $10 billion and $15 billion. So I don't want to make macro an excuse because I think we have some fantastic demand because of our fantastic products. That said, obviously, the way we will see the campus demand manifest as we may see longer decision cycles as the macro continues. But the actual interest in our products is just very solid, very good because we're still operating -- we're still waiting for our first $1 billion here in the next few years. So I think it's looking good. On the (inaudible), Anshul, you want to add some comments. It's been a very exciting launch, a lot of demand, still to early, right?

There's a lot of disruption happening in the market, even at the edge, right? When you talk about campuses, when you go and talk about what's inside the building, but there's an edge that connects to the outside, too. So, we have taken a interest from our customers in this area. It is a bit too early. We just announced this offering. So too early to jump up and down and say it's lots of membership. We'll watch it how it grows. But having the integrated solution is important to go after the broader enterprise market.

Our existing customers -- they see the natural affinity to our CloudVision and EOS tax. So that would be the natural thought.

Your next question comes from the line of Erik Suppiger with JMP Securities.

Can you speak at least qualitatively about how much backlog you're using up at this point? And how long do you expect to continue working down backlog orders on a quarterly basis?

Yes. I'd love to talk about backlog, Erik. I don't think we're going to do that. Look, I think it all comes back to lead times, right? The lead times is the driver of all else, right? We (inaudible) customers had to place orders to those lead times. As we kind of improve customer (inaudible) to shorter lead time, that will slowly kind of navigate our way back to a more normal visibility window and lead time will be (inaudible). I think we're taking the first steps of that now, right? But that's going to take time to get there. But that's what will happen as lead times will naturally bring visibility back to something more normal. But that's going to take some time. We're in the very beginning of that.

Is that a multiyear process?

I don't know when we're back to normal. It will take time to get there.

Yes. We've pretty much said it will take us the back half of 2023. So I think normal is really next year.

Your next question comes from the line of Simon Leopold with Raymond James.

Just a quick one. If we could get a metric, the RPO value. And in terms of my question, it does seem as if you've had a number of announcements around some software capabilities around your campus and enterprise-focused products. And it feels like you've sort of built up a critical mass. I'm wondering how you look at that strategy in terms of its maturity and its readiness in terms of the software behind the campus portfolio, if there's anything still missing?

Yes. Thanks, Simon. I think most of our software is not stand-alone software. It's bundled either in CloudVision or EOS. There's very few stand-alone pieces, of course, as the NDR and the DANZ monitoring fabric. Specific to the campus, I would say they tend to look at it more as a solution where they bring in wired, wireless, they want an automation framework with CloudVision. They want some security capabilities. In order to say, we announced some of the missing gaps we filled. Historically, we've partnered with other vendors for Network Access Control and Arista introduced its first one. I'll talk about it more next quarter. But I think some of the gaps we are now starting to fill ourselves. But in the campus, it always tends to be a combination of platforms and software never stand-alone software alone.

Yes. The RSA announcement was actually sort of where my question was coming from. It felt like that was in my mind, maybe one of the final gaps, and that's what I'm really trying to understand here.

Yes. I was going to save this for next quarter, but I'll just give you the condensed version is the Arista Guardian for Network Identity AGNI means fire in Indian language, Sanskrit. So we'll scrap some of the ice with that fire.

Sounds good. And just the RPO value, do you have that handy?

Yes. I think it's up about between $50 million to $60 million quarter-over-quarter. It did pick up a little bit on the software and services side, but not in that range. Still small numbers, a long way to go yes.

Your next question comes from the line of Sami Badri with Credit Suisse.

Two number questions. One is, could you just give us an idea on campus revenues in 1Q '23, and that's going to trend in 2023?

And then for Cloud Titans, is that expected to grow, grow double digits? Any kind of reference that you could make for what cloud titan is going to do for the balance of 2023.

Yes. I don't think we're going to try to prevent the vertical split. I think what we said is we think our 2 biggest customers will be at least 10% customers for the year, but it's not all we've really said on that front.

And we feel good about that here. They're good partners, and despite all the volatility, we'll have a good year with them.

And on campus, I think I gave you a number for $750 million by 2025. I'll stick with that, and we'll give that to you towards the end of the year when we really achieve something.

Operator, we have time for one last question, please.

Your final question today comes from the line of George Notter with Jefferies.

I guess just following on, on some of the questions about visibility and customers. Any evidence of customers building inventory of your products whether that's just stand-alone inventory or whether it's inventory that's maybe installed in the network, but not being fully utilized. I guess what I'm really asking is any thoughts about excess inventory rather than just kind of the normal buffers that are out there.

I think you asked the question well and you almost answered it. It's just normal buffers. There's nothing excess in inventory we're seeing. Nothing abnormal.

(inaudible) that's so constrained. George. I mean.

(inaudible) should they have some. We'd like to know how they got it.

This concludes the Arista Networks First Quarter 2023 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.

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