The operator introduces the conference call for Arista Networks' First Quarter 2024 financial results, with participants in listen-only mode and a Q&A session to follow. Liz Stine, Director of Investor Relations, introduces CEO Jayshree Ullal and CFO Chantelle Breithaupt. Forward-looking statements will be made regarding financial outlook, market opportunities, and potential risks and uncertainties.

In the first quarter of 2024, Arista reported strong financial results, with revenue of $1.57 billion and non-GAAP earnings per share of $1.99. The company is positioning itself as a leader in the data-driven networking market, with a focus on AI and cloud networking. The management team, including co-founders Ken and Andy, is highly experienced and engaged in the company's initiatives. The Chief Development Officer, Hugh Holbrook, is leading major platform initiatives alongside John McCool and Alex.

The engineering team at Arista is highly regarded and the company is grateful for Anshul Sadana's 16 years of service. Anshul expresses his gratitude and announces his decision to take a break and explore new opportunities. He thanks his colleagues, leadership, and two special people - Andy Bechtolsheim and Jayshree Ullal. Jayshree expresses her appreciation for Anshul and announces that Arista will not be replacing the COO role.

Arista Networks is leveraging its executives to drive new initiatives and has seen success in all three sectors. They are focused on quality and innovation and are building a programmable network with various overlays. Ken Duda, the CTO, discusses the NetDL data lake and how it supports their network as a service strategy. Historically, network data was discarded, but Arista is now collecting and interpreting it to improve network operations.

The use of network addresses and port numbers alone does not provide enough information about user activity. Arista has developed NetDL, which collects and stores large amounts of data in real time, including network events, counters, and flow data. This data is enriched with user service and application layer data from external sources, allowing for a better understanding of network performance and user experience. NetDL is a crucial part of Arista's EOS stack and enables advanced functionality, such as fabric-wide visibility and network as a service. Arista's network as-a-Service strategy involves providing customers with tools to build and operate services, with Arista's CV-NAS handling equipment selection, deployment, and troubleshooting.

CV-NAS provides end-user self-service for customers to manage their service instance, provision new endpoints and virtual topologies, set traffic prioritization policies, and access rules. NetDL is the database foundation for NAS, which is not a separate SKU but a collection of functions in Cloud Vision. These include AVD, AGNI, UNO, and ZTN, and serve as a strategic business opportunity and guiding principle for Cloud Vision development. However, the core promise to prioritize quality remains unchanged. Ken is praised for his execution in the Arista way.

CloudVision UNO, based on the EOS Network Data Lake Foundation, provides fault detection, correction, and recovery for improved observability of the entire network. Arista's AI for networking platforms have been successful, with a recent customer highlighting a 24,000-node GPU cluster based on the 7800 AI Spine. Ethernet is proving to offer at least 10% improvement in job completion performance compared to InfiniBand, making it a critical infrastructure for both front-end and back-end AI data centers. The trend of AI networking is expected to continue throughout the year and decade.

In the paragraph, the speaker discusses the importance of seamless communication among compute nodes for AI applications and their progress in four major AI Ethernet clusters. They also mention their increasing confidence in achieving their AI target and their projection of above 10% to 12% annual growth in 2024. The CFO then reviews the financial specifics of their Q1 results and guidance for Q2, which exceeded expectations due to strength in the enterprise and cloud verticals.

In the first quarter, services and subscription software accounted for 16.9% of revenue, while international revenues were at $316 million, representing 20.1% of total revenue. This decrease from the previous quarter was due to quarterly volatility and a higher contribution from EMEA customers in the prior quarter. The growth margin for the quarter was 64.2%, higher than the guidance of 62%, and was driven by supply chain productivity gains, a stronger mix of enterprise business, and a favorable revenue mix. Operating expenses were $265 million, with R&D spending at $164.6 million, sales and marketing expense at $83.7 million, and G&A costs at $16.7 million. Income from operations was $744 million, and net income was $637.7 million, representing 40.6% of revenue.

In the first quarter, the diluted share number was 319.9 million shares resulting in a diluted earnings per share of $1.99, a 39% increase from the previous year. The company also ended the quarter with $5.45 billion in cash, cash equivalents, and investments. They completed a $1 billion stock repurchase program and authorized a new $1.2 billion program. The company generated $513.8 million in cash from operations but saw a slight increase in inventory and an increase in deferred revenue.

The majority of the deferred revenue balance is related to services and may fluctuate due to timing of service contracts. The product deferred revenue balance decreased by $25 million. The company expects significant new product introductions and customer growth in 2024, which may increase the volatility of the product deferred revenue balance. Accounts payable days decreased from the previous quarter. The company is raising its revenue guidance for fiscal year 2024 and expects a gross margin of 62% to 64%. The company will continue to monitor the market and make investments in targeted hires, R&D, and go-to-market team.

The company has raised its expectations for the full year, with a focus on customer mix. The gross margin guidance suggests a higher contribution from enterprise, but there has been unexpected activity in all three sectors (cloud and AI titans, providers, and enterprise) in the first quarter.

The company has raised its guidance earlier than expected due to feeling good about their performance in all three areas. They would have ideally liked to look at two quarters before raising guidance, but the diversified momentum and mix of momentum gave them confidence. When asked about the half-over-half growth, the CEO explained that as the company's numbers get larger, activity needs to translate to larger numbers. The company is cautiously optimistic about their increased guidance and sees it as a sign of confidence.

The operator introduces the next question from George Notter of Jefferies, who asks about the comparison between Ethernet and InfiniBand. Jayshree Ullal explains that while InfiniBand has traditionally been considered lossless, Ethernet has shown to be 10% faster in job completion time in real-world environments. She also mentions the potential for further improvements with the Ultra Ethernet Consortium. Notter clarifies if they are discussing Rocky Ethernet, to which Ullal confirms.

Jayshree Ullal discusses the improvements in RDMA over Ethernet and the potential impact of UEC. She also addresses the company's work on a UEC compatible version of Ethernet and the need for both halves to work together. In response to a question about NVIDIA, she mentions going head to head in the market and the potential impact of their systems-based approach with BlackRock.

The speaker expresses gratitude to NVIDIA and its CEO for their contribution in creating a market for AI networking. They acknowledge the presence of NVIDIA GPUs and NICs in the market and see them as a partner rather than a competitor. They are confident in their own company's ability to provide a scale-out network while working with NVIDIA's scale-up GPUs. The questioner asks about the speaker's commitment to remaining the CEO in light of recent executive transitions at Arista.

The speaker responds to a question about the reasons customers would continue to use InfiniBand instead of switching to Ethernet. She mentions that Anshul, an executive, has decided to leave the company, but they have a strong bench and she is committed to staying for a long time. The next question is about the new network computing capabilities introduced by NVIDIA, and the speaker is asked if this is something that could be introduced by the broader merchant Silicon and internet ecosystem. She responds affirmatively.

Jayshree Ullal discusses the feasibility of doing in-network compute functions and offloading the network for some of those functions. She mentions that it can be done both at the compute layer and network layer, but the timing of 800 gig optics availability and the development of a full ecosystem for it may take more than a year. There will likely be early introductions of optical and switching products, but the full ecosystem may not be available until 2025 or 2026.

The speaker discusses the impact of AI networking on Arista's business, stating that there will be a need for higher performance systems as GPUs with 800 gig become more prevalent. They do not currently take the front-end inferencing portion into consideration in their projections, but they expect to see a more holistic network connecting back-end clusters in the future. Customers are looking favorably at Arista for this reason.

The speaker explains that the transition from AI pilots to full deployments is a matter of time and familiarity. There are various factors that need to be addressed, such as power and cooling considerations, infrastructure setup, and foundational discovery of networking needs. The operating system also plays a crucial role in determining the necessary features and connectivity. The deployment process is similar to that of the cloud, with various use cases emerging and the need for high-performance and efficient training clusters. The upcoming 400 and 800 gig technologies are not the main bottlenecks, but rather a part of the overall process.

The availability of all the necessary components is currently the main obstacle in deploying AI clusters, but there is a high demand for them. The company has experience in building large storage clusters which can be applied to AI clusters. Ethernet has a significant advantage over other networking technologies due to Metcalfe's law. It is uncertain how long it will take for InfiniBand to surpass fiber channel.

Jayshree Ullal, CEO of Meta, discusses the company's visibility in terms of customers and their plans for the future. In November, they had limited visibility of only three to six months, but now they have improved to at least six months. The Cloud and AI titans are shifting their focus to a more balanced approach, with an increase in AI and expansion of cloud regions. In the Enterprise sector, there is a strong activity in Q1, which leads to an increase in guidance for the rest of the year. Ullal believes that the company's conservative team will see even better results in the future.

The speaker discusses the potential growth of cloud providers in the AI market, specifically tier two providers. They mention an increase in visibility and activity, with sales cycles taking longer for enterprise clients. They also mention ANET's production rollout to support GPUs and the potential for 10,000 to 100,000 GPUs by 2025. However, the speaker notes that there are many factors that need to come together for this growth to occur, including the development of AI centers and facilities.

The speaker discusses the key factors in AI production, including GPU availability, network scale and performance, and power considerations. They note that GPU lead times have improved, which will allow for more customers to build out their networks. They also mention the importance of power, GPU, locality, and performance in predicting a positive rate of production in 2025. The speaker also mentions that there may be a ramp towards reaching the 750 million goal next year and that there may be some contribution to this year's outlook.

Jayshree Ullal and Chantelle Breithaupt discuss their plans for AI and reaching their 750 million target in 2025. They clarify that this is a glide path and not a hockey stick growth. They also mention their investments in go-to-market strategies and their focus on improving lead times. They plan to continue investing in AI and cloud, but do not see significant changes in their go-to-market approach.

The speaker discusses the company's approach to the campus market and its focus on larger enterprise customers. They mention the use of channel partners and their plans to expand into the mid-market and target specific verticals. The speaker also addresses the decline in server unit numbers and the potential for a server refresh cycle to be a catalyst for the company.

Jayshree Ullal, CEO of Arista Networks, believes that there will be another cloud cycle due to server refresh and associated use cases. She also notes that regional expansion by major cloud companies is contributing to this trend. When asked about the impact of software revenue on enterprise activity, Ullal attributes it to the high quality and innovation of their software and the universal topology they offer. She also mentions that while the enterprise use cases for AI are currently small, there is some activity in that area.

The deployment schedule for BlackRock in AI is slower than expected, with initial customer delivery delayed until late this year. However, this is not expected to cause a pause in cloud spending as customers still need to build a good scale out network regardless of the GPUs or BlackRock.

The company is preparing for the network to connect to a BlackRock or H100, and they are prioritizing training workloads and job completion time. They are willing to invest in the network regardless of the GPU being used. A presentation with additional information on their results can be accessed on their website. The call has now ended.

This summary was generated with AI and may contain some inaccuracies.