(NAS:SGH) Smart Global Holdings: Pivoting to an AI Specialist

Neutral rating on Smart Global Holdings (NAS:SGH) this month. Through extensive research, analyses, and speaking with IR, our initial hypothesis for a “buy” rating did not materialize. Our principle: if we don’t add it to our portfolio, no “buy” rating. Despite lacking excitement, neutral ratings are essential for credibility and track record.

Thesis

Smart Global Holdings is made up of three subsidiaries and although the company’s past shows a diversified holding company, it is now evolving to an operating one. SGH is in the early innings of their turnaround, namely focusing the company on High Performance Computing (HPC) around their brand Penguin Solutions (this will also be the name of the company in the near future). While we are optimistic about the company’s prospects, we require further evidence of execution of their HPC segment along with increased enterprise adoption of AI solutions, either on-premise or at the edge, before considering initiating a position.

Background

SGH began as a private equity-owned memory module company, initially based in Brazil. The company strategically diversified into HPC and memory solutions as part of its preparation for an IPO. This diversification allowed SGH to position itself as a more comprehensive technology solutions provider. Since 2018, the company’s share price has remained relatively flat, largely due to challenges stemming from Brazilian regulations that have significantly impacted their memory segment, as well as turnover within several of their acquired businesses. They have since divested their Brazilian segment and it is likely that SGH will continue to divest their other non-core segments in the near future, allowing them to concentrate their efforts on the core driver of the business going forward: Penguin Solutions.

Penguin Solutions

Penguin Solutions initially focused on providing hardware solutions for research laboratories and government agencies and has over 25 years of experience in system deployments. Today, Penguin specializes in building and managing high-performance computing clusters—networks of servers that work together to perform complex calculations at high speeds. Historically, HPC was primarily used by academia and government labs, but it has now entered the commercial spotlight, driven by the rise of AI. Enterprises are increasingly leveraging HPC to train large language models (LLMs) and unlock AI’s potential. Penguin Solutions offers a comprehensive, end-to-end service, from designing and building server racks to installing and managing them for uptime and security. Its offerings are hardware-agnostic, sourcing equipment from competitors like Dell and HPE, as well as partners such as Nvidia and AMD. And for those customers that want a managed solution, Penguin can host customers servers on their own Penguin Cloud.  

Penguin’s expertise in HPC is what gives them an advantage compared to other players such as HPE and Dell and is why companies such as META hired them to build out certain parts of their AI super cluster.  To date, Penguin Solutions has deployed and manages over 75k GPUs and recently won a contract with Voltage Park, a GPU service provider to manage a cluster of 24K H100 GPs.

Penguin Solutions’ business is fundamentally tied to on-premise data centers, which are located on-site rather than in the cloud. Year-over-year, revenues excluding acquisitions have remained flat, while companies like Supermicro, which provide hardware and services to major cloud service providers (CSPs), have reported triple-digit growth (see graph below). CSPs have been the key players in the AI race, with most enterprises remaining largely on the sidelines. The trillion-dollar question is: How will AI influence cloud versus on-premise adoption? Will it reverse the trend toward cloud migration, or further accelerate it? We explore the dynamics between cloud and on-premise infrastructure in the discussion below.

Cloud vs On-Prem

Enterprises today have several options for hosting their workloads: the cloud, on-premise (“on-prem”), or colocation facilities, where a company owns hardware housed in another provider’s data center. The leading cloud service providers (CSPs) are Microsoft, Amazon, and Google. CSPs have gained market share over on-prem solutions due to their scalability and convenience. They handle maintenance, updates, and security, alleviating the burden on in-house IT teams, while offering features that can be cost-prohibitive, particularly for smaller organizations. The chart below illustrates the growing penetration of CSPs, which now account for approximately 50% of IT capacity.

For larger companies, the dynamics shift as cloud costs begin to weigh heavily on margins, often outweighing the benefits once enjoyed at a smaller scale. Companies like Dropbox nearly doubled their gross margins by consolidating workloads to on-premise infrastructure and moving away from the cloud. For large-scale organizations, it is not uncommon for cloud costs to account for more than 50% of their cost of goods sold (COGS).

So, does AI change the equation? AI has raised concerns about uncontrolled data leakage, with large language models (LLMs) potentially being trained on proprietary company data. This has led many enterprises to consider “repatriating” workloads back to on-premise infrastructure. In a recent CIO survey, 83 out of 100 respondents indicated plans to repatriate, up from around 60 three years ago. However, while the intent is clear, we believe AI workloads have actually widened the gap between cloud and on-prem offerings. Managing data centers equipped with GPUs is increasingly complex, with GPU utilization often dropping below 50% due to networking issues and system failures. So far, CSPs have proven superior to handling this complexity due to hardware innovation, API integration, model flexibility, privacy, and integration with other best-in-class tools like database management.

Opportunities

Inference workloads will push compute to happen at the edge

Edge data centers—smaller facilities typically with fewer than 50 racks located in urban areas—represent a highly fragmented industry. However, as inference workloads grow and take on a larger share of computational tasks, more of this processing could be pushed to the edge. The edge is critical due to the physical limitations of the speed of light and the latency caused by long distances. A transformation in edge data centers could expand the total addressable market for on-premise and hybrid cloud providers like Penguin Solutions. The edge market in the U.S. is currently valued at approximately $10 billion.

Defense Spending and Sovereign AI will be primarily on-prem

As AI advances and takes on military and political significance, defense and government agencies will increasingly seek to keep the majority of their data and compute on-premise, rather than relying on cloud providers. Penguin Solutions, with its strong operational history—particularly in the U.S.—and extensive experience in designing and installing supercomputers for the Department of Defense, is well-positioned to benefit from this shift. The increased government spending, projected to reach nearly $1 billion in 2024, up from $600 million in 2023, presents a significant opportunity. In contrast, competitors with more troubled histories, such as Supermicro Computer, which is currently under a DOJ investigation, will likely face greater challenges entering this market.

SK Telecom partnership can bolster Penguin’s offerings

SK Telecom, a subsidiary of SK Hynix—one of the world’s largest DRAM chip suppliers- recently acquired a stake in SGH through a $200 million preferred equity investment. While the go-to-market strategy has not been fully detailed, it can be inferred that SK aims to leverage partners like SGH, which specialize in designing and installing solutions, to promote their offerings. Other potential areas for partnership include the further development of software and service offerings, as well as expanding growth beyond North America.

Risks

AI and Cloud Adoption

With the rapidly evolving ecosystem of GPUs, networking, and other cutting-edge technologies, cloud providers are best positioned to stay ahead, given their access to talent, capital, and a culture focused on delivering best-in-class solutions. The speed of technological advancements may drive enterprises to adopt cloud offerings more quickly, allowing them to concentrate on their respective core competencies. This would shrink the TAM for on-prem solution providers such as Penguin Solutions.

Leadership

SGH remains largely under the same management as it had during its acquisition spree four years ago. While the company is now refocusing, much of the effort involves addressing challenges created during that period. There has also been significant turnover within its subsidiaries, including Penguin Solutions. Whether the management team can successfully revitalize the company and implement the right strategy remains uncertain.

Conclusion

Although SGH is cheap at the current levels, we are not taking a position given the state of flux the company and the industry is in. SGH is undergoing a significant transformation from a diversified holding company to an operating entity focused on high-performance computing through its Penguin Solutions brand. While the company’s refocusing efforts are encouraging, particularly in the growing AI space, challenges remain around execution and turnover within acquired businesses. However, before initiating a position, we need to see more concrete evidence of execution and increased enterprise repatriation on-premise or at the edge.

INVESTMENT DISCLAIMERS & INVESTMENT RISKS
Past performance is not necessarily indicative of future results. All investments carry significant risk, and it’s important to note that we are not in the business of providing investment advice. All investment decisions of an individual remain the specific responsibility of that individual. There is no guarantee that our research, analysis, and forward-looking price targets will result in profits or that they will not result in a full loss or losses. All investors are advised to fully understand all risks associated with any kind of investing they choose to do.