2023-12-28 08:04:39 ET
Summary
- Arista's main competitive advantage lies within a stable, adaptative, and efficient operating system leveraging high-speed switches.
- Ethernet switches are gaining momentum within large IA clusters and could gain IT budget versus InfinityBand. Arista is well-placed to dominate its competitors.
- Financials are outstanding, with a 40% cash flow conversion and a highly net cash balance sheet allowing for regular and increasing buybacks.
- However, my DCF model indicates the stock is fairly valued. I am waiting for a better entry point.
My thesis
The goal of this article is to assess Arista Networks (ANET) positioning in the high-performance switch market, to look at its competitors, and to discuss how the firm can further gain share in large cluster workloads used for artificial intelligence (namely AI). We will see that Arista seems to be well positioned to grow double-digit per year in the medium term while benefiting from operating leverage. My company valuation will complete the writing and scenario analyses will add more ranges of potential outcomes. While the business case is appealing, positive development seems to be priced in the stock. I'm rating the stock as a HOLD: I'm waiting for a better entry point.
Investment overview
The Ethernet network switches market is estimated to be close to $45bn, where Cisco (CSCO) is the leader with a 40% market share (if annualized from IDC's recent data, see the graph below). Arista is second with a 10% share followed by Huawei with 8% which has been pushed away over recent years from critical US/EU infrastructure for security reasons. Other important firms are HP (HPE), Juniper (JNPR), and the Chinese company H3C. The rest of the market is highly fragmented. Note that outside Ethernet solutions, Nvidia (NVDA) offers a competitive offer with its InfinityBand switches, originally designed by Mellanox. While InfinityBand seems to have the best performance in terms of latency, the rise of AI large clusters could offer new generations of Ethernet switches (200G to 800G) an advantage: a sub-segment where Arista seems to dominate. Arista's management estimates its addressable market count grow from $37bn in 2023 to more than $60bn by 2027, leading to a CAGR of 13%.
Arista was created at its origins for high-performance networking and had as its first main client Google in 2004 . The company benefits from a successful team as founders Kenneth Duda and Andy Bechtolsheim are still active as board members. Jayshree Ullal joined Arista in 2008 to become CEO, with previous experience as a senior executive at Cisco in the data center branch. In her early career, she was a semiconductor engineer within notable US tech companies. They all have a substantial stake in the company which makes them aligned with shareholders' interests.
The firm key competitive advantage lies within its operating system but also its ability to design superior architecture . The high programmability of its EOS allows for the integration of external silicon chips mainly sourced from Broadcom (AVGO) and partially from Marvell (MRVL). Able to rapidly leverage up-to-date best chips of the industry without integration friction gives Arista a competitive advantage versus Cisco which uses a mix of in-house developed chips (ASICs) and delayed integration of Broadcom's chips. Arista manufacturing is taken care of through external contractors.
Today, it dominates the high-end performance 100G/400G switching market with more than 35% of the share ahead of Cisco (20%) and other smaller competitors such as Juniper, Nvidia (Mellanox), and Chinese companies H3C/Huawei/ZTE (outside the US). We can see in the following picture that the data center Ethernet switch market is expanding through high-speed segments (>100Gb per second), where Arista has an edge. The company has already shipped some 800Gbps systems that could be the next growth driver.
Meta (META) and Microsoft (MSFT) represented 46% of its $4.4bn FY2022 revenues, progressing from 25% in 2021 and 32% in 2020 . 70% of Arista revenues come from Cloud hyper-scalers and data center providers. While this dependence on two customers could make me worry, in the meantime, they are sticky and have growing CAPEX plans. Also, Arista's ongoing development plants into an enterprise campus network could help to diversify the margin of the future revenue stream. The firm sees $750m in campus sales by 2025 which should represent 10% of its revenues.
Arista's recent developments look promising. First, it settled partnerships with cyber-security firms such as Zscaler (ZS) and CrowdStrike (CRWD) to offer Network-as-a-Service solutions. NaaS represents a fast-growing market as enterprise budgets are diversifying away from firewalls and VPNs. In 2024, the firm will showcase a new 800Gps Ethernet Spine network able to handle a connection with 10-30k GPUs. The growth of large language models for AI requires exponential performance improvement. Indeed, as stated by Nvidia GPU performance to increase 2X each generation and cluster size by X4, leading to an overall X8 trajectory at every new advanced AI model. The rise of large language models requiring numerous parameters and bigger cluster sizes leveraged by thousands of GPUs could help Ethernet networking gain a share on InfinityBand and NV-Link.
What valuation can we expect?
I broadly aligned myself with the management's financial targets. I modeled 16% revenue CAGR from 2024 to 2027, slightly above the addressable market expected growth of 13% as Arista should continue to gain market share.
On operating expenses: I converged toward an R&D/sales ratio of 14% (EST 12%-14% seen by the management), 6.5% for S&M (seen at 6%), and G&A below 2% (inline). The effective tax rate should gradually increase as deferred tax assets are consumed. CAPEX should stay low as Arista is outsourcing its production. I do expect a working capital release in the coming years as the cash conversion cycle (record high) should moderate along with more inventory shipments (post-build-up). With this scenario, I find that the firm can generate: between $3.4bn and $3.8bn in the years 2025-2026, leading to cash-flow yields of 5% (P/E 20X) to respectively 5.5% (P/E of 18X) which looks reasonable.
Given this model, a terminal growth rate of +3%, and a WACC of 10%, I obtain a share price of $221/share that is slightly below the current market price: a lot of good news has already been priced in. To give more perspective, I implemented a sensitivity analysis, varying the discount rate and mid-term growth.
Considering the relative value, we can see Arista is priced at similar financial ratios to its peers having elevated exposure to data centers.
Balance sheet analysis
The company has a net cash balance sheet. The cash and short-term investments position increased from $3bn to $4.5bn over the nine months. Stock repurchases have reached a cumulative $1.6bn since 2020 with no dividends paid. M&A activity was modest over the period: Arista seems to have all the building blocks required to grow organically. Looking back at financials, we see that Arista's FCF margin on revenues reaches 40%: it is a cash generator. Therefore, I do believe it has more room to increase its shareholder buyback program.
Technical analysis
We can see that Arista's stock price has been highly correlated with the previously stated peer group. This is rational as its revenue exposure is significantly dependent on the data center segment. Arista's YTD performance has been strong and driven by solid quarterly beats, not so much on speculation that AI will be a key driver.
Risks
The first risk that comes to my mind is related to the competitive landscape. While Arista continuously gained market share against Cisco and Juniper, this trend could reverse if the firm doesn't always keep up with the most innovative solutions. Also, hyperscaler cloud clients could further in-source part of their CAPEX by developing "white-boxes". I however think that the shift toward AI-driven workloads should strengthen the firm's competitive advantage as best-in-class solutions are required (high stability and customizability, low latency and error rates). A second risk relates to Arista's customer concentration: Microsoft and Meta represent close to 40% of its sales and could induce an elevated cyclicality of the firm's results (which are highly dependent on their data center spending).
Conclusion
Two decades of investments in high-performance networking have placed Arista Networks in the right spot to benefit from the rise of AI workloads. An experienced management team has built a powerful suite of connectivity solutions leveraged by a stable, adaptable operating system favored by data center operators. The company's solid balance sheet and strong cash-flow generation make me confident in the ability of the company to continue to innovate and redistribute any excess liquidity to shareholders. What can we add? Well, all that comes with a price, and the current stock level makes me stay on the sideline. I'm rating the stock as a HOLD: I'm waiting for a better entry point.
For further details see:
Arista Networks: Enabling AI Data Center Workloads