Description

CLS has been written up several times so please refer to those writeups for business background/history. This writeup will focus on the AI part of Celestica’s business as it propelled Celestica to report its highest quarterly result in company history last quarter. I believe the AI tailwind is durable, and will transform Celestica’s business profile by making Celestica’s revenue more predictable/durable, higher growth, with higher margins.

At 13x NTM P/E (consensus #s), CLS is a compelling AI play. Over the past five years, CLS’s Connectivity & Cloud Solutions (CCS) team engaged in a concerted effort to build out their data center capabilities and service offerings to better serve hyperscale and data center-focused customers' needs. CLS is now a trusted provider to the top 5 global hyperscalers across core data center technologies and a prime beneficiary of AI capex spend. CCS segment revenues accounted for 68% of total company revenues last quarter and Celestica is positioned to continue to benefit from the insatiable AI capex spending from the mega cap Internet companies and hyperscalers. The massive AI investments can be viewed as a strategic "dollar auction," where companies are willing to overspend in the short term to secure long-term dominance in AI.

Tech giants are demonstrating unwavering commitment to AI investments, viewing it as crucial for long-term growth and competitiveness despite uncertain short-term returns. For example, Microsoft is taking a decidedly long-term approach to AI investments. CFO Amy Hood stated that their data center investments are expected to facilitate AI monetization "over the next 15 years and beyond". Google CEO Sundar Pichai emphasized that "the risk of underinvesting is much higher than that of overinvesting". This sentiment reflects Google's strategy to maintain its competitive edge in the AI race, even if immediate returns are not apparent. Here is Mark Zuckerberg earlier this year on AI capex spending:

“I think that there’s a meaningful chance that a lot of the companies are over-building now, and that you’ll look back and you’re like, ‘oh, we maybe all spent some number of billions of dollars more than we had to,’” Zuckerberg said. “On the flip side, I actually think all the companies that are investing are making a rational decision, because the downside of being behind is that you’re out of position for like the most important technology for the next 10 to 15 years.”

“If AI is going to be as important in the future as mobile platforms are, then I just don’t want to be in the position where we’re accessing AI through” a competitor, said Zuckerberg, who has long been frustrated with Meta’s reliance on distributing its social media apps on phones and operating systems from Google and Apple Inc. “We’re a technology company and we need to be able to kind of build stuff not just at the app layer but all the way down. And it’s worth it to us to make these massive investments to do that.”

The combined capital expenditures of Microsoft, Alphabet, and Meta grew by 60% year-over-year in the latest quarter, with expectations of 50% growth for 2024. Annualized CapEx for big tech increased from $138B to $229B year-over-year. This incremental $91B in run-rate spending is a good proxy for new AI data center construction—an enormous investment.

This is an arms race and none of the hyperscalers and internet giants want to let the others get ahead. We are now in a cycle of competitive escalation between three of the biggest companies in the history of the world, collectively worth more than $7T. At each cycle of the escalation, there is an easy justification—we have plenty of money to afford this. There is a real-sense in which only companies that have the balance sheets to withstand big write-offs can now afford to play in the AI infrastructure race. From this perspective, overbuilding may be perfectly rational.

CLS has experienced >3x growth in AI/ML compute demand in 2023 compared to 2022. Enterprise revenue growth, which is 25% of total revenue in 2023 (up from 19% in 2022 and we project will eventually reach 50% of total revenue), after declining for the past decade+ grew 33% in 2022 and 41% in 2023. CLS’s annual Hyperscaler revenue has grown at a 48% CAGR over the past six years with 2022 being the inflection point where overall Enterprise revenue returned to double digit growth. Hyperscaler revenue comes at more attractive margins because of the higher design content, IP, sourcing leverage, and unique customization offerings at scale. The fastest categories for spending growth for CLS are AI/ML servers and High End (400G/800G) Networking. In AI/ML, CLS differentiates in custom silicon due to higher technical requirements.

While we expect there will be fluctuations in demand as there are in any major technology deployment cycle, we believe the AI investment cycle is durable and will provide Celestica with a runway for future growth as technology advancement in the industry is expected to drive future hardware refresh cycles to support expanded application workloads. Unlike metaverse or crypto, the AI use-cases are tangible and with massive reach, including a knowledge worker base of 1.5 B, and large and small enterprises adopting AI. We are currently in the midst of a massive capex boom in AI acceleration. The competitive drive to build the best LLM — which is a function of data size, data quality, model size and computing power — has led to a demand for ever greater AI hardware/Data Center investments. The market believes that after a surge in AI capex for training these models and for enterprises experimenting with AI, demand will level off. However, after training comes inference, which the market underappreciates, as a much larger source of AI hardware demand. Demand can last for decades with the related inferencing and other computing ecosystem requirements. Inference is critical as it is performed real-time (i.e low latency at the prompt or the customer will search elsewhere, for example) in production unlike training --- which is performed as batch processing that can last several days. Per Amazon at their AWS:Reinvent “for every $1 spent on training, up to $9 is spent on inference.”

Accelerated computing (using GPUs) only has a 2.7% penetration rate - we are on the cusp of one of the biggest data center refreshes in two decades.

There are 3 ways we think about CLS’s CCS segment. First, the end market view, which splits the segment between their communications business comprised of networking gear and their enterprise business comprised of storage and servers, the latter of which is predominantly AI, ML compute. Second is a solutions view, which splits the segment into the hardware platform solutions business, which comprises joint design and white box offerings in Celestica’s high-value EMS business. Finally, there is Celestica’s customer view, split between hyperscalers and OEMs. It is worth noting that the hyperscaler customers account for a meaningful portion of Celestica’s HPS business, utilizing their server, networking and storage offerings.  

The hyperscaler portfolio, remains a key market, driven by increasing investments in AI capabilities and capacity and is forecasted to grow meaningfully in the years ahead. This group of customers is expected to lead spending on AI capabilities and to account for an outsized share of data center CapEx over the next several years. So why do hyperscalers choose to work with Celestica? Celestica has decades of proven leadership in engineering and manufacturing of hardware platforms, including switching, routing, wireless, storage and high-performance compute.

Momentum with hyperscalers continues to increase. From the latest quarter:

“Our CCS segment saw a 51% year-to-year increase in revenues in the second quarter and achieved segment margin of 7.2%, which is 120 basis points higher compared to the prior year period. The top line growth and margin expansion continue to be supported by large scale investments in data center infrastructure from our hyperscaler customers, including very strong demand for our HPS offering.”

“Our CCS segment saw a very strong growth in the second quarter, with revenues up 51% compared to the same period in 2023 driven by solid demand in both our enterprise and communications end markets. The broad-based strength across both our end markets continues to be supported by demand from our hyperscaler customers, which grew by 93% in the first half of 2024, compared to the same period in 2023. The strength we are seeing in our hyperscaler portfolio is broad-based as our growth is being driven by solid demand across multiple technologies, programs and customers.”

“We expect solid demand to support continued year-to-year growth in the second half. In 2024 we anticipate revenues from our programs with hyperscaler customers will surpass $4.6 billion and will account for more than 70% of our CCS segment revenues.”  

“Revenue in our enterprise end market was up by 37% year-over-year in the second quarter, above our expectation of a low 20 percentage increase driven by strong demand for AI/ML compute programs. Revenue in our communications end market was higher by 64% compared to the prior year period, which was better than our expectation of a mid-40s percentage increase. The growth in our communications end market was driven by accelerating demand for HPS networking products from our hyperscaler customers, primarily in support of their investments in AI/ML infrastructure.”

One way Celestica has sought to differentiate itself is with its HPS (hardware platform solutions) offerings, which consists of developing infrastructure platforms, hardware and software design solutions and services in collaboration primarily with CCS segment customers, as well as managing aspects of the supply chain and manufacturing, including firmware/software enablement across all primary IT infrastructure data center technologies, and after-market support. HPS differentiates itself by encompassing advanced technology design solutions that customers can tailor to their specific platform applications.

“In 2023, the business is expected to generate approximately $1.7 billion in revenue across more than 50 active programs. This business serves as an innovation engine for our customers and a demand generator with our suppliers and ecosystem partners. Our flexible engagement models enable joint design, facilitate customized hardware or provide fully realized, ready now solutions. Our HPS business operates similar to an ODM model and that we can provide white box solutions as well as full stack solutions with a Celestica hardened operating system, an integrated third-party operating system or an open-source operating system.”

“Our experience across multiple silicon and component vendors and software ecosystems provides a capability that enables first-generation innovations, which Celestica leverages in platform, customized and integrated rack solutions. We believe our competitive advantage stems from our joint design and manufacturing routes with differentiated capabilities to customize these platforms into solutions that are optimized for our customers' applications and workloads such as AI, along with a comprehensive portfolio of platforms spanning networking, storage and compute. We believe this platform and offering is unique amongst our EMS peers as we rarely compete with Tier 1 EMS players in this space and primarily find ourselves competing with ODMs.”

“Due to our end-to-end involvement from design, to build to post-deployment support, as well as our proprietary intellectual property, we realized higher margins and enjoy deeper integration with our customers, making our position harder to displace for competitors. Additionally, as design cycles accelerate, we feel that our incumbency position, coupled with our competency in the latest technology such as 400G and 800G switching and AI, ML compute provides an advantaged position to competitors working towards proficiency in these complex designs and supply chains. As a result, we believe our HPS offering will continue to be a key driver of growth for CCS.”

“Dan, as you know, not all hyperscalers are created equal. And there are some hyperscalers that lean towards more complex challenging product deployments. And that really works out well for us. The reason that you're seeing the amount of growth that we're showing in HPS is because those are the solutions that really fit the demand requirements of certain hyperscalers. And so as we see this shift towards more AI type of data centers, it implies more complex product, which is what we want to see.”

Furthermore, HPS products are typically single sourced:

“HPS products are typically single sourced for the life of the program and that's given their stringent qualification process and also next generation products typically awarded to the incumbent assuming strong initial performance because switching costs are also prohibitive. And again within our AIML data center offerings, a lot of our high value EMS programs are transitioning to HPS products over time because of the value add we're able to bring into the equation, which we think is a competitive edge versus our EMS peers.”

“HPS revenue was $686 million in the second quarter, accounting for 29% of total company revenues and was up 94% year-over-year. The very strong growth in our HPS portfolio was driven by an acceleration in demand for networking products from hyperscaler customers, including 800G switch programs.”

Total revenue grew 20% in 1Q 2024 and accelerated to 23% in the second quarter. This is despite ATS (Advanced Technology Solutions) revenue shrinking 11% in Q2 due to macro and excess channel inventory. Given the insatiable demand from the largest tech giants and Celestica’s established position with some of those AI buildout programs, I model 24% revenue growth in 2025 (helped also by a rebound in ATS in 2025) and 18% growth in 2026 before growth falls back to 10%. Given the low levels of accelerated computing market share in the data center, the data center refresh that is needed will likely drive high growth far beyond 2026. The largest tech giants have the scale, balance sheet, are highly FCF generative, and have the desire to overbuild for years to come. I don’t take into account further demand from sovereign nations who want to build their own AI systems. One large sovereign could be as large as a cloud customer.

The hyperscaler programs generate attractive margins given their highly complex and specialized design, engineering and manufacturing requirements and their demand for supporting services, such as RAC integration and IT asset disposition offering.

“One is margins with the hyperscalers in totality are accretive to our margins in CCS. And then margins in HPS are also accretive to overall CCS. And so the dynamic that you're seeing right now with hyperscalers continuing to grow and with our HPS revenue also continuing to grow, that's helping pull margins up right now. And so we do think that we continue to have a nice mix in the portfolio and opportunities for more operating leverage as we go forward.” Q2 2024 earnings

HPS products have higher margins because there is more value add.

“Typically, what happens on HPS products. As these products -- server products transition from high-value EMS into HPS, they will carry higher margins because they have higher value add. If they stay EMS and during the technology transition mode, what we see is early in the life cycle of new product introduction, the ASPs are higher and as the volumes pick up and over time, the ASPs tend to come down and normalize over a little bit. And that has been true and continues to be true as in all product life cycle pricing. But again, as products transition from EMS to HPS, the margins will increase because of the higher value add.” Q2 2024 earnings

Celestica has historically had 2-3% operating margins and 7% gross margins. Margins have expanded in recent years and Celestica now has 10%+ gross margins and 6%+ operating margins. The market is skeptical that Celestica is massively overearning. However, the business has been transformed over the last five years from AI and higher quality end markets such as Aerospace & Defense and Healthcare, all of which should provide more revenue visibility and higher structural margins. Celestica’s Lifestyle Solutions business is now 61% of total revenue and continues to become a larger percentage of total revenue.

Lifecycle Solutions portfolio revenue consists of our combined ATS segment and HPS business revenues. We disclose the combined revenue of these businesses as they share several key characteristics and commercial strategy focus, including robust long-term growth prospects, higher-value added solutions throughout the product lifecycle, and higher margins than our traditional CCS segment businesses.

Therefore, we believe that margins are sustainable, and further believe that gross margins can expand to 12% in the coming years driven by continued growth in Lifecycle Solutions revenue and significantly more scale from the AI tailwind.

Celestica’s valuation multiple should expand to 10x EBITDA, 15-16x P/E, as Celestica becomes viewed as a high quality EMS/ODM business with more revenue visibility and higher margins that in the past.

10x 2025 EBITDA (16x P/E) results in 70% upside for a 43% IRR.

Catalyst

• Rebound in Industrials business, which has been weaker than expected due to macroeconomic conditions and excess channel inventory.
• New AI related wins: “We typically get about a year of visibility I would say as programs transition from one technology for the next. So we have pretty good visibility when it comes to the transitions certainly with our largest customer, that visibility is a bit extended because we constantly share technology road maps. And then the funnel for additional AI compute programs is quite large, actually, we're engaged with a number of different customers across a number of different opportunities as they are working to develop custom silicon solutions. And again, for me, it's just a matter of time before we win several of these opportunities and add them to our backlog.” 2Q 2024 earnings