Penguin Solutions, Inc. (PENG) - Deep Dive Research Report

Research prepared May 23, 2026

1. What the Company Does

Founded in 1988 in Fremont, California as SMART Modular Technologies, what is now Penguin Solutions spent three decades as a specialty memory manufacturer - designing DRAM modules for demanding applications where commodity parts would not do. That identity has been gradually dismantled and rebuilt through a sequence of acquisitions that changed what the company is. The name SMART Global Holdings (under which it IPO'd in 2017) captured the old identity. The name Penguin Solutions, which the company formally adopted in October 2024, captures what it is trying to become.

The problem Penguin Solutions solves is the gap between "a company decides to run AI" and "a company is actually running AI in production." Buying NVIDIA GPUs is not deploying AI. You need racked clusters with the right networking topology, validated reference architectures, software to manage the cluster, memory optimized for inference latency, fault-tolerant infrastructure to keep production workloads alive, and then ongoing managed services to troubleshoot and upgrade the system across a multi-year lifecycle. Penguin Solutions provides most of these layers and partners for the rest.

The company's three acquisitions between 2020 and 2022 each added a missing capability. Penguin Computing (acquired October 2020 for up to $85M) brought HPC cluster engineering depth - 20 years of building systems that made the TOP500 supercomputer list. Cree's LED components business (acquired March 2021 for $175M in cash and notes) brought high-performance specialty components with a globally recognized brand. Stratus Technologies (acquired August 2022 for $225M plus a $50M earnout) brought fault-tolerant infrastructure for mission-critical production workloads - the Stratus system that guarantees 99.99999% uptime. The fourth building block, SMART Modular Technologies, had been there since 1988: the memory engineering capability that now underlies the company's next-generation AI inference memory products.

A concrete example of what they actually do: a tier-one US financial institution wants to deploy a large language model on-premise to process sensitive customer documents without sending data to a public cloud. They call Penguin Solutions. Penguin brings an OriginAI reference architecture - a validated blueprint specifying GPU server configurations, networking topology, storage, and management software. Penguin procures the hardware (often partnering with Dell PowerEdge servers), installs ClusterWare AI cluster management software, deploys SMART Modular DDR5 and CXL memory modules optimized for inference workloads, and wraps the whole thing in managed services to keep it running. The Q2 FY2026 earnings call confirmed this archetype: a Tier 1 US financial institution purchased Penguin's CXL-based MemoryAI KV Cache servers for exactly this use case, in a data center AI factory deployment. The company has collectively deployed nearly 100,000 GPUs and managed over 2 billion hours of GPU runtime - numbers that matter in sales conversations because enterprise buyers will not risk their AI program on a vendor with no track record.

"AI is moving from experimentation to production with workloads increasingly shifting towards real-time inference. Memory is becoming a defining constraint and a defining opportunity." - Kash Shaikh, CEO, Q2 FY2026 earnings call (April 1, 2026)

Shaikh, who took over as CEO on February 2, 2026 - replacing Mark Adams who retired after leading the transformation from SMART Global - distilled the company's market thesis better than any product description does. Memory is not a commodity in AI infrastructure. It is the bottleneck.

2. Business Segments

Advanced Computing (47% of FY2025 revenue)

This segment is the most strategically complex and the most publicly discussed, combining three distinct sub-businesses with genuinely different competitive dynamics.

HPC and AI Systems Integration (Penguin Computing brand, OriginAI): This is the soul of the company's ambition. The capability is cluster-level AI infrastructure - designing, procuring, assembling, configuring, and deploying multi-node GPU or CPU clusters for enterprise, sovereign AI, and neocloud customers. The product is not just hardware. It is the OriginAI reference architecture (validated GPU server configurations, networking, storage, and management software, pre-integrated by Penguin's engineers), the ClusterWareAI cluster management software, and a service layer that covers design, build, deployment, and ongoing management.

What distinguishes Penguin Computing from a straightforward systems integrator is the heritage. Before the SMART Global acquisition, Penguin Computing had spent 20 years building HPC clusters for national laboratories and research institutions, including systems that appeared on the TOP500 list. In 2022, the company was inducted into the Space Technology Hall of Fame for commercializing Beowulf clustering - the open-source approach to building supercomputers from commodity hardware that fundamentally changed what HPC systems could cost. This background means Penguin's engineers understand cluster-level performance optimization, workload scheduling, and infrastructure reliability in ways that a generalist reseller does not.

The segment has been going through a deliberate exit from two prior revenue streams. Hyperscale hardware sales - large cluster deployments for cloud provider AI training - were the original growth driver when the company pivoted toward AI infrastructure in 2022-2023. Management exited this business in FY2026 because the margin profile did not justify the concentration risk and capital intensity. Simultaneously, the Penguin Edge product line (SMART EC brand for edge computing and wireless computing) was discontinued by calendar year-end 2025. Together, these exits created a 14-percentage-point headwind to Advanced Computing revenue growth in FY2026.

What remains and is growing is non-hyperscale AI/HPC: enterprise financial institutions, energy companies, life sciences organizations, defense contractors, federal agencies, education institutions, and neocloud GPU rental providers. This sub-segment grew 75% in full-year FY2025, and 50% in the first half of FY2026, and has moved from 20% of segment revenue to 40% over the same period. The company secured 7 new customer logos in H1 FY2026 versus 3 in H1 FY2025 - an acceleration that drives confidence in the FY2026 back-half sales mix.

Revenue in Advanced Computing is structurally lumpy because a single multi-rack deployment for an enterprise customer can shift meaningfully between quarters based on site readiness, hardware availability, and customer procurement cycles. Management has been transparent about this - CFO Olmstead explicitly guided Advanced Computing down sequentially from Q1 to Q2 FY2026, and it came true.

Fault-Tolerant Computing (Stratus brand): Acquired in August 2022, Stratus is a 40-year-old company with a very specific identity: it builds computing platforms that do not fail. The flagship product is the Stratus ztC Endurance, a platform achieving "seven nines" uptime (99.99999%, meaning less than 3 seconds of unplanned downtime per year). A second-generation model, the ztC Endurance 9110, announced in 2025, improved performance by 46% with 64 cores for high-volume transaction processing. The Stratus ztC Edge serves industrial OT environments - factory floors, oil refineries, power grids - where IT staff are not present and the hardware must self-monitor and self-heal. The Stratus ftServer, now certified for Oracle Database applications, addresses enterprise mission-critical database workloads.

The Stratus business model is fundamentally different from the HPC integration business. Stratus sells hardware once, then collects multi-year support and service contracts. Customers who install a Stratus platform do not replace it on a normal technology refresh cycle. The qualification process to certify new computing infrastructure in a 24/7 automated manufacturing plant or pipeline control system can take months and is subject to regulatory approval in some industries. The cost of an hour of unplanned downtime in a petrochemical plant or a financial transaction processor dwarfs any conceivable price premium from a competitor. Stratus has been running since the 1980s in some customer installations.

Management has identified a new strategic direction for Stratus: extending fault-tolerant infrastructure into production AI inferencing. As enterprises move LLM inference from pilot to production, they face the same seven-nines availability requirements for AI as they do for database systems - a customer service LLM that goes down is felt immediately. Stratus's ztC Endurance platform, with its no-touch fault tolerance, is positioned for this use case.

What existed as Penguin Edge (now discontinued): SMART EC and wireless computing products formed a product line targeting edge AI inference across distributed computing environments. Management decided to exit this product line, citing insufficient returns relative to capital deployed. The wind-down creates a transition responsibility to existing customers and a sales organization distraction but eliminates ongoing investment in a product line that did not differentiate.

Why this segment exists as one unit: both Penguin Computing and Stratus are in the business of making AI and compute infrastructure work reliably at scale. Penguin Computing brings scale and performance; Stratus brings fault tolerance and mission-critical reliability. The combined sales pitch to a production AI deployment is powerful: "We build the cluster and guarantee it stays up."

Integrated Memory (34% of FY2025 revenue)

The Integrated Memory segment is SMART Modular Technologies - the company's original business, now operating as the fastest-growing segment in FY2025 and H1 FY2026. The growth has been dramatic: approximately 30% in FY2025, and 63% year-over-year in Q2 FY2026, driven by the intersection of AI infrastructure demand, DDR5 migration, and the company's increasingly differentiated memory products.

The segment operates on two structural levels. The first is memory module design and manufacturing: sourcing DRAM and NAND flash from commodity manufacturers (Samsung, Micron, SK Hynix, Kioxia) and engineering them into high-performance, high-reliability modules for demanding applications. The customers are networking equipment OEMs, telecom infrastructure builders, AI server manufacturers, and directly growing AI infrastructure buyers. Modules for these customers carry custom firmware, extended temperature ranges, specialized form factors, and long-lifecycle supply commitments that commodity modules do not offer. The Zefr line serves the highest-reliability tier: mission-critical server environments where memory failure is not tolerable.

The second level is supply chain services: procurement management, logistics, inventory, and technical support for customers who need reliable, matched memory supply across product lifetimes of 10 years or more. A telecom equipment maker cannot source memory ad-hoc as DRAM generations change - they need a partner who can source and qualify matching memory across the full lifecycle of their infrastructure. This service business creates stickiness that pure hardware sales do not.

The forward-looking products in this segment carry the most strategic weight. The Compute ExpressLink (CXL) product family represents the company's bet on the next memory architecture. CXL is an interconnect standard that allows memory to be expanded beyond what fits in a server chassis, shared across multiple servers, or made available from external memory pools at near-DRAM latency. Penguin's SMART Modular CXL NV-CMM E3.S 2T module achieved CXL Consortium compliance certification in January 2026, with early production orders arriving through Q1 and Q2 FY2026. A Tier 1 financial institution deployed CXL-based KV Cache servers (for LLM inference serving) as of Q2 FY2026.

The Optical Memory Appliance (OMA), targeted for late 2026/early 2027 first customer shipments, is the segment's most ambitious bet: an external memory appliance providing large, pooled, low-latency memory for AI inference workloads, addressing the memory bandwidth bottleneck that currently limits GPU utilization in large language model serving.

The MemoryAI product line, announced at NVIDIA GTC in March 2026, includes purpose-built servers for AI inference workloads and the MemoryAI KV Cache server optimized for LLM response acceleration. This moves Penguin from memory module supply into AI inference server design - a meaningful step up the value chain with substantially better margin potential.

Optimized LED (19% of FY2025 revenue)

The Optimized LED segment is the part of Penguin Solutions that least fits the AI infrastructure narrative. Acquired from Cree, Inc. in March 2021, the business operates under the Cree LED brand and carries 30+ years of gallium nitride (GaN) based LED development and more than 2,000 patents.

Cree LED designs and manufactures GaN-based blue and green LED chips and packages them into XLamp and J Series components for lighting OEMs, automotive lighting manufacturers, display builders, and specialty application users (horticulture, medical, entertainment). The brand's position in high-brightness, high-reliability LED components is built on decades of engineering relationships. Lighting OEMs who have qualified Cree XLamp LEDs in their fixture designs do not change easily - their UL listings, photometric performance data, and warranty commitments are all tied to the specific LED specification. Re-qualification with a competing component from Nichia or ams-Osram is not a trivial undertaking.

The challenges are structural. Chinese LED manufacturers have achieved cost parity in commodity packaging applications. Nichia, ams-Osram, and Lumileds are more scale-efficient in high-power specialty applications. Manufacturing in Huizhou, China creates tariff exposure. The segment was essentially flat in FY2025 and is guided down 5-15% in FY2026. Management cited tariff uncertainties as a specific headwind in Q3 FY2025.

The segment still generates meaningful gross profit from an installed base of qualified customers, and the brand name Cree LED carries genuine value in the specialty lighting market. But the strategic fit with an AI factory platform company is weak, and at some point capital allocation discipline should raise the question of what this asset is worth to a strategic acquirer with genuine scale in LED manufacturing.

3. Products and Business Detail

OriginAI Factory Architecture: Not a standalone product - a validated AI cluster blueprint specifying GPU servers, networking topology, storage, and management software, pre-integrated by Penguin's engineers. Available with GPU options from NVIDIA and AMD, hardware from Dell PowerEdge (via the Dell partnership announced in late 2024), and storage from Dell PowerScale. Penguin has deployed these architectures across financial services, energy, research, defense, and neocloud customers. The validation work - making sure the specific hardware combination actually performs reliably at scale - is what customers pay for. Getting this wrong in production costs a customer far more than Penguin's service fee.

ClusterWareAI (formerly ICE ClusterWare): Proprietary cluster management software provisioning, monitoring, and managing large GPU clusters. Management has been integrating ClusterWare with open-source HPC software (including NVIDIA management tools) to reduce customer lock-in risk while maintaining the software revenue stream. When a customer purchases an OriginAI cluster, ClusterWare comes with it as a software subscription, creating recurring revenue beyond the hardware transaction. The software layer is strategically crucial: if Penguin remains a pure hardware assembler, it is commoditizable. If ClusterWare becomes the cluster management standard for enterprise AI, the retention dynamics improve materially.

MemoryAI Server and MemoryAI KV Cache Server: Announced at NVIDIA GTC (March 2026), these are servers purpose-built for AI inference workloads where memory, not compute, is the binding constraint. The KV Cache server specifically addresses the "key-value cache" that LLM serving systems maintain for each concurrent inference session - as context windows grow and concurrent users increase, the memory requirement per inference grows dramatically, and general-purpose GPU servers cannot serve enough concurrent requests. A CXL-augmented server with dense, high-bandwidth memory and fast CPU access can serve substantially more users simultaneously. The Tier 1 financial institution deployment in Q2 FY2026 is the first live validation.

SMART Modular CXL NV-CMM E3.S 2T: CXL 2.0 compliant non-volatile memory module, certified by the CXL Consortium in January 2026 (listed on the CXL Consortium Integrators List). Uses the CXL interconnect standard to provide near-DRAM-latency, high-bandwidth persistent storage for memory expansion, in-memory database acceleration, and fast data recovery after power loss. The compliance certification is meaningful because it allows customers to buy in confidence that the module works with any CXL-compliant host.

Optical Memory Appliance (OMA): The most ambitious near-term product. An external appliance providing a large, pooled memory tier for AI inference clusters, targeting the GPU utilization bottleneck where expensive accelerators sit idle waiting for memory bandwidth. Management first mentioned the OMA in Q3 FY2025 (July 2025), confirmed the late 2026/early 2027 first shipment timeline in Q4 FY2025, and reiterated it in subsequent calls. This product, if it ships on schedule and performs as described, moves Penguin from component supply into AI systems architecture - a qualitative change in business model.

Stratus ztC Endurance (second generation): Seven-nines fault-tolerant computing platform for data center environments. The 9110 model (announced 2025) delivers 46% better performance than its predecessor with 64 cores for high-volume transaction processing. Fully redundant hardware, automated self-healing, predictive fault detection. Customers include industrial manufacturers, utility operators, financial transaction processors, and healthcare device builders. Second-generation introduced to extend commercial relevance into AI production inferencing workloads.

Stratus ztC Edge: Rugged edge computing for OT environments - factory floors, oil and gas operations, power infrastructure. Self-monitoring, self-protecting, designed to operate without local IT staff. Supports edge AI inferencing for industrial automation.

Stratus ftServer: Earlier-generation fault-tolerant server. Certified for Oracle Database applications in 2025, extending commercial lifecycle into enterprise mission-critical database environments.

Zefr Memory Modules: Ultra-high-reliability DRAM modules for server environments with zero tolerance for memory failure. Military-grade testing, extended temperature range, and extended lifecycle support.

SMARTsemi: High-performance discrete memory components for embedded and industrial applications. OEM manufacturers who need custom, long-lifecycle memory solutions for products that ship in volume for a decade.

Cree LED XLamp and J Series: Application-optimized LED components for general illumination, horticultural lighting, automotive applications, and specialty fixtures. GaN-based chip technology. XLamp is the flagship high-power component line; J Series targets mid-power applications. The component IP portfolio, including more than 2,000 patents, underpins both the product lines and potential licensing revenue.

Manufacturing geography: Memory module manufacturing primarily in Penang, Malaysia and Fremont, California, with additional capabilities in Asia-Pacific. LED chip and component manufacturing in Huizhou, China and New Taipei City (Taiwan). Stratus assembles in Maynooth, Ireland and Massachusetts. R&D spread across approximately 430 personnel globally. All major facilities hold ISO 9001:2015, 14001:2015, and 45001:2018 certifications.

Supply chain structure: For memory, Penguin sources DRAM and NAND from Samsung, Micron, SK Hynix, and Kioxia under purchase-order-based procurement (not long-term volume commitments). For HPC systems, Penguin sources GPU servers from NVIDIA, AMD, and Dell; networking from major vendors; and storage from Dell. The lack of long-term supply commitments creates both flexibility and risk - Penguin can adjust quickly to demand changes, but has no guaranteed priority in tight supply markets.

4. Customers

Who buys, and why:

Advanced Computing AI/HPC customers are enterprises, government agencies, and neocloud providers that need private GPU clusters. The buyer is typically a Chief Technology Officer or Head of AI Infrastructure, advised by a team of HPC or cloud architects. The decision criteria are cluster performance per dollar, reliability of the integrator's engineering, the quality of the cluster management software, and the integrator's track record in similar deployments. The sales cycle runs 12-18 months from initial engagement to hardware shipment, during which Penguin's engineers work alongside the customer's team to validate the architecture.

This extended sales cycle creates meaningful switching cost. Once a customer's architecture has been designed around OriginAI blueprints and their team trained on ClusterWareAI, switching to a Dell direct relationship or an HPE GreenLake managed service requires re-qualification, retraining, and a period of reduced operational capability. The customers who have validated Penguin's cluster design do not re-qualify competitors without a compelling reason.

Named customer relationships that management has discussed publicly: a tier-one US financial institution deploying an on-premise GenAI data center (Q4 FY2025 and Q2 FY2026 calls), SK Telecom's HAN sovereign AI system (completed deployment Q4 FY2025), Georgia Tech AI Makerspace (announced Q2 FY2026), Deepgram enterprise voice AI collaboration (Q2 FY2026), Haein sovereign AI in South Korea (Q2 FY2026), and multiple Fortune Global 500, Fortune 100, and Fortune 50 engagements that management referenced without naming due to confidentiality.

Customer concentration: The top 10 customers represented 66% of FY2025 revenue, up from 58% in FY2024. The increase reflects the concentration risk inherent in large project-based HPC deployments. As the company diversifies away from the hyperscaler hardware business (one large customer) toward many enterprise customers, concentration should decrease over time. Whether management can execute this diversification fast enough to smooth out quarterly revenue variability is an open question.

Stratus customers represent the stickiest relationship in the portfolio. Industrial manufacturers operating 24/7 automated production lines cannot replace their fault-tolerant computing platform without shutting down production and re-certifying the replacement. A Stratus installation certified in a pharmaceutical manufacturing process might require FDA approval to replace. These relationships are effectively permanent within product lifecycle windows.

Integrated Memory customers span networking OEM manufacturers needing specialized DDR5 for their product lines, AI server builders sourcing validated memory for GPU servers, telecom infrastructure providers (heavily weighted toward DDR4 for legacy infrastructure), and increasingly direct AI infrastructure buyers. Supply chain service customers are the stickiest - they depend on Penguin for lifecycle sourcing across a product that they sell for 10+ years.

Cree LED customers are lighting fixture manufacturers (residential, commercial, industrial), automotive lighting system integrators, and specialty application builders. A fixture manufacturer's relationship with Cree LED is sticky while the fixture model is in production but weakens when the fixture is redesigned. A new fixture design is a re-qualification opportunity for a competing LED supplier. The stickiness is medium-term - years, not decades.

Contract structure: Advanced Computing deployments are typically project-based: a hardware revenue event followed by software and services contracts covering management, maintenance, and support for multiple years. Stratus follows a hardware-once, support-forever pattern. Memory is a mix of spot procurement and recurring supply agreements. LED is spot and quarterly purchase orders.

5. Competitive Landscape

Advanced Computing HPC and AI Infrastructure

The market Penguin competes in for enterprise AI infrastructure has attracted every major technology vendor with a systems and services capability. The competition varies significantly by customer segment.

Dell Technologies holds roughly 20% of the AI server market per ABI Research (2024) and is simultaneously Penguin's biggest channel partner and a direct competitor. The Dell + Penguin OriginAI collaboration (announced November 2024) gives Penguin access to Dell's hardware volume and supply chain, while Dell gets Penguin's cluster engineering and software. In practice, Dell is the larger threat: it has deeper enterprise IT relationships, a larger services organization, and significant bargaining power with NVIDIA. Where Penguin wins against Dell: customers who want best-of-breed open-source software stacks rather than Dell's proprietary management tools, and customers who prefer a specialized HPC integrator's engineering depth over a generalist systems vendor.

Hewlett Packard Enterprise held approximately 22% of on-premise HPC sales globally in 2025 per Hyperion Research - the largest single share. HPE's Cray heritage gives it deep credibility in the national lab and supercomputing segments where Penguin also competes. HPE GreenLake provides a cloud-like managed services model that enterprises increasingly request. Penguin competes with HPE by being faster to deploy heterogeneous best-of-breed configurations, more flexible on software stack, and less tied to HPE-proprietary infrastructure components.

Super Micro Computer (Supermicro) has taken substantial share in AI server infrastructure by manufacturing its own servers and selling directly to hyperscalers and large enterprises at competitive prices. Supermicro is primarily a hardware vendor - it does not offer the cluster management software or managed services layer that Penguin does. In accounts that are price-sensitive and have in-house HPC engineering capability, Supermicro can undercut Penguin on pure hardware cost.

Hyperion Research's market share data for on-premise HPC in 2025 showed HPE at approximately 22%, Dell at approximately 18%, and Penguin at approximately $508M in revenue - meaningful share that placed it above IBM and Atos in the rankings. Penguin is a genuine third-tier player in HPC with growing AI identity, not a niche operation.

Barriers to entry in enterprise AI HPC integration are real but not impenetrable. The technical barriers are the validated reference architectures, the cluster management software investment, and the engineering team who can troubleshoot a 500-GPU cluster at 3am when a production AI workload fails. These take years to build. The commercial barriers are customer relationships and the track record that new logos want to see. A new entrant could build the technical capability in 3-5 years with sufficient capital; replicating 20 years of HPC deployment experience at national labs is not possible on any timeline.

Fault-Tolerant Computing (Stratus)

Stratus competes in a narrow, durable, and structurally high-switching-cost niche. Primary competitors include Schneider Electric's EcoStruxure Micro Data Center (combining ftServer technology with physical protection for edge environments), SealingTech (military and government), and IEI Integration Corp (rugged industrial computing). None of these competitors has Stratus's track record of decades in 24/7 industrial environments or the depth of the ztC product line's predictive fault tolerance. The competitive threat in Stratus is not new entrants taking share - it is the long-term question of whether next-generation cloud-native high-availability architectures make dedicated fault-tolerant hardware less necessary. So far, the industrial OT market has not embraced cloud-native software-defined reliability at the same rate as IT environments.

Integrated Memory

At the commodity end of DRAM modules, Penguin cannot compete with Samsung, Micron, or SK Hynix selling directly to hyperscalers. The company does not compete there. Its position is in specialty applications - custom form factors, custom firmware, extended temperature ratings, long lifecycle sourcing - where module engineering skill and lifecycle commitment matter. Competitors include Viking Technology (a Sanmina division), Netlist, and smaller Asian module builders. The CXL differentiation, if early production orders scale, moves Penguin into a sub-market (CXL memory expansion) where competition is thin and margins are likely to be structurally better.

Optimized LED

Nichia (Japan) has held the technical leadership in LED chip efficiency and patent coverage for decades. ams-Osram (Austria/Germany) brings Osram's automotive LED heritage combined with sensor integration capability. Lumileds (being acquired by San'an Optoelectronics and Inari Amertron) is a historical peer in high-power LEDs. Seoul Semiconductor competes broadly in packaged LEDs. Cree LED's claimed position of more than 30% global market share in high-power lighting LEDs is a legacy of the company's R&D investment over 30 years, but it is being eroded structurally by Asian manufacturers achieving cost parity in non-differentiated applications.

6. Industry

What drives demand for Penguin's most important business:

Enterprise AI infrastructure spending is the primary demand driver. Hyperscalers committed more than $320B collectively in AI-capable data center capex in FY2025 - training the models that enterprises will deploy. But a substantial fraction of enterprise AI workloads, particularly those involving sensitive data (financial, healthcare, defense), real-time inference at scale, or sovereign mandates, will be deployed on-premise rather than in public cloud. Gartner estimated worldwide enterprise spending on AI-optimized IaaS at $37.5B in 2026, with inference-focused spending growing from $9.2B to $20.6B over 2025-2026. Enterprise on-premise AI cluster deployments are where Penguin's integrated offering - architecture, software, memory, fault tolerance, managed services - is most defensible.

HPC AI infrastructure total market size is growing at an estimated 27.6% CAGR through 2030 per Technavio, with the market expanding by approximately $147B over the period. At its current on-premise HPC revenue run rate, Penguin represents a small fraction of this market - substantial opportunity to gain share if enterprise AI deployment follows the demand trajectory that hyperscaler capex is betting on.

Memory market dynamics:

DRAM and NAND markets are historically among the most cyclical in technology. Supply-demand imbalances cause dramatic pricing swings every 4-5 years. The current favorable pricing environment (stable to rising DRAM in 2025-2026) has been a major tailwind for Integrated Memory. Management acknowledged in the Q2 FY2026 call that rising memory costs are "explicitly expected to slow customer demand" - a sign that pricing has moved enough to create demand elasticity. A sustained memory price downturn, triggered by supply additions outpacing AI infrastructure demand, would simultaneously compress Integrated Memory revenue and gross margins.

The DDR5 generation transition is a multi-year structural tailwind: as networking, server, and AI infrastructure builders upgrade to DDR5, demand for specialty DDR5 modules in the complex form factors and reliability grades that Penguin serves is growing faster than the overall memory market.

CXL is the most significant emerging technology in memory. The CXL standard allows compute and memory to be disaggregated, shared across multiple servers, and expanded beyond physical server limits. This transforms memory from a server-level spec into a data center resource that can be managed and allocated dynamically. If CXL achieves mainstream adoption over the next 3-5 years, it creates a new addressable market for specialty memory module makers with CXL engineering capability - which currently means a small number of companies, including Penguin.

LED industry:

The LED component packaging market was approximately $16-17B in 2024 growing at 3.9% CAGR per MarketsandMarkets. Growth is uneven: general lighting (52% of demand) is mature and commoditizing. Automotive LED and specialty applications are growing faster. The structural trend favoring Cree LED's addressable market (high-brightness, specialty, high-reliability) is outweighed by competitive pressure from scale manufacturers in Asia who have achieved LED chip efficiency within striking distance of Cree's historical leadership.

Where the company sits in the global supply chain:

Penguin is not a chip manufacturer. In memory, it buys DRAM and NAND from Tier 1 memory manufacturers and adds engineering value through module design, firmware, and lifecycle support. In AI infrastructure, it buys GPU servers from NVIDIA OEM partners and assembles them into validated clusters with its own software layer. In LED, it designs and manufactures GaN chips through its own Huizhou facility. The company's structural position is an engineering and integration layer above chip manufacture and below the end customer.

Regulatory environment:

US government spending on AI infrastructure (DARPA, Department of Energy national labs, defense) is a demand driver that is recession-resistant and growing. Penguin's heritage in national lab HPC deployments positions it to capture some of this spending. The regulatory risk is primarily tariff-based: Chinese manufacturing exposure (Cree LED in Huizhou, memory supply chains dependent on SK Hynix and Samsung) creates sensitivity to US-China trade escalation.

Cyclicality:

Advanced Computing HPC/AI is not deeply cyclical but is project-based, creating quarterly lumpiness rather than smooth demand. Memory is highly cyclical. LED follows construction and renovation trends steadily. The portfolio mix means the company is moderately cyclical overall, with memory being the variable that can swing results in either direction most dramatically.

7. Growth Triggers

• Non-hyperscale AI/HPC customer acquisition accelerating. Seven new customer logos signed in H1 FY2026 versus three in H1 FY2025. Management has signed master service agreements with major financial institutions and oil and gas companies, with purchase orders expected "shortly" in early 2026. Pipeline spans financial services, federal, energy, education, and neocloud segments. (Q1 FY2026 concall, January 6, 2026; Q2 FY2026 concall, April 1, 2026 - accelerating)

"We closed five new AI HPC customer wins in Q2, spanning financial services, education, life sciences, and federal sectors. These wins represent the continued broadening of our go-to-market reach and diversification from our legacy hyperscaler business." - Kash Shaikh, CEO, Q2 FY2026 concall (April 1, 2026)

• MemoryAI KV Cache server deployment scaling. A Tier 1 US financial institution purchased CXL-based MemoryAI KV Cache servers for on-premise LLM inference as of Q2 FY2026, providing the first named production validation of the MemoryAI product line. Management expects this use case to grow as inference workloads scale and context windows expand. (Q2 FY2026 concall, April 1, 2026 - new)

"As inference workloads scale, we expect these types of deployments to become increasingly important." - Kash Shaikh, CEO, Q2 FY2026 concall (April 1, 2026)

• CXL product ramp. Early production orders for CXL memory cards received in Q3 FY2025. SMART Modular CXL NV-CMM E3.S 2T achieved CXL Consortium compliance certification in January 2026 and is on the Integrators List. Management noted "a substantial order for CXL cards" in Q2 FY2026, confirming the product has moved from qualification to procurement. (Q3 FY2025 concall, July 8, 2025; Q2 FY2026 concall, April 1, 2026 - repeated, progressing)

• Optical Memory Appliance (OMA) launch. First customer shipments targeted late 2026/early 2027. The OMA addresses GPU memory bandwidth constraints that currently limit inference cluster utilization. Management first named this product in Q3 FY2025 and has reiterated the timeline consistently. First orders have been placed. (Q3 FY2025 concall, July 8, 2025; Q4 FY2025 concall, October 7, 2025 - repeated, timeline unchanged)

• Sovereign AI program expansion. SK Telecom HAN system (South Korea national AI) deployed in Q4 FY2025. New sovereign AI engagement with Haein in South Korea announced Q2 FY2026. Management sees sovereign AI as a structurally distinct demand category - governments building national AI capability cannot outsource to US hyperscaler clouds. (Q4 FY2025 concall, October 7, 2025; Q2 FY2026 concall, April 1, 2026 - repeated)

• Second-half FY2026 revenue concentration. Management guided 53-54% of FY2026 revenue to fall in H2, driven by new customer deployments converting from signed agreements to delivered hardware. H2 sales mix skewed heavier by project deployment timing, not unusual for an HPC integration business. (Q1 FY2026 concall, January 6, 2026; Q2 FY2026 concall, April 1, 2026 - repeated)

• Stratus expansion into production AI inferencing. Management identified the Stratus fault-tolerant platform as naturally suited for enterprise production AI inference, where seven-nines availability is required for customer-facing LLM applications in financial services, healthcare, and critical infrastructure. Second-generation Stratus ztC Endurance 9110 positioned for this use case. (Q4 FY2025 concall, October 7, 2025 - new)

• Channel partner scale (CDW, Dell). Expanding beyond direct sales model through CDW and Dell partnerships to reach mid-market enterprises the direct sales team cannot economically serve. Early-stage but management highlighted these as proof-of-concept successes. (Q3 FY2025 concall, July 8, 2025; Q4 FY2025 concall, October 7, 2025 - repeated)

• New CPO product acceleration. Ian Colle (former AWS, named HPCWire "People to Watch 2026") appointed Chief Product Officer in Q2 FY2026. Signals a more structured product roadmap process under the new CEO. (Q2 FY2026 concall, April 1, 2026 - new)

8. Key Risks

Revenue lumpiness in Advanced Computing. The non-hyperscale AI/HPC business grows through large, multi-rack project deployments. A single enterprise customer deployment can represent meaningful revenue and shift between quarters based on site preparation, hardware delivery, and customer readiness. Management has cited 12-18 month sales cycles from engagement to hardware shipment, and revenue recognition on delivered hardware occurs separately from software and services. Any quarter where a large deployment slips produces a miss against expectations - and the pattern of wide Advanced Computing guidance ranges (-25% to -15% for FY2026) reflects management's own uncertainty about deployment timing. This is not fixable; it is structural to the project-based business model.

Memory pricing reversal. The Integrated Memory segment's dramatic growth has been partly powered by favorable DRAM and NAND pricing. Memory markets are historically among the most volatile in all of technology. The current tight supply and AI-demand-driven pricing will eventually reverse when memory manufacturers respond with capacity additions. Management explicitly warned in Q2 FY2026 that rising memory costs were already "expected to slow customer demand." A sustained memory downturn - historically a 30-40% pricing decline within 12-18 months of an upcycle peak - would simultaneously compress Integrated Memory revenue and gross margins, removing the segment that has been offsetting Advanced Computing volatility.

Customer concentration. With the top 10 customers representing 66% of FY2025 revenue, a decision by any one or two large customers to delay or redirect their AI infrastructure spending would move the revenue needle materially. The exit from hyperscale hardware reduces the most extreme concentration risk, but the enterprise AI/HPC business is still being diversified. Seven new logos in H1 FY2026 is encouraging but does not yet represent the depth of diversification that would dampen single-customer impact.

CEO transition risk. Kash Shaikh joined as CEO on February 2, 2026, replacing Mark Adams who built the current portfolio through a decade of acquisitions and operational execution. Shaikh's background is enterprise software and SaaS (CEO of Securonix, a cybersecurity SaaS company), not HPC or memory. His "AI factory platform" framing is a significant strategic repositioning - from a hardware integration business toward a software-and-services platform narrative. This direction may be correct for long-term value creation, but strategic transitions driven by new CEOs take 2-3 years to play out. The risk is organizational disruption, customer confusion, or capital misallocation during a period when the company needs to convert its expanding pipeline into signed contracts and delivered deployments. Q2 FY2026 was Shaikh's first call - far too early to assess his execution.

Penguin Edge wind-down execution risk. The discontinuation of the Penguin Edge product line (SMART EC, wireless computing) creates a 14-percentage-point headwind to Advanced Computing FY2026 growth. Customers on these platforms need migration paths. If the wind-down generates customer attrition from the broader Penguin relationship, or requires extraordinary support costs, it could distract the sales organization from winning new accounts at exactly the moment when new account growth is most important.

Cree LED structural decline acceleration. The LED segment is guided down 5-15% in FY2026. If the decline accelerates - driven by more aggressive tariff escalation on Chinese manufacturing, faster Chinese competitor share capture, or an automotive market downturn reducing demand for specialty automotive LEDs - the company faces a capital allocation decision about when to invest in restructuring versus when to divest. A rapid LED deterioration would not threaten the company's viability but would create management distraction and potentially impair the book value of the segment.

Supply chain constraints for AI infrastructure. Extended lead times for high-end NVIDIA GPUs and specific networking components constrain deployment timelines. If a customer has committed to a production AI go-live date and Penguin cannot deliver the hardware, the relationship suffers even though the constraint is not Penguin's fault. In a market where enterprise AI programs have internal political momentum and committed timelines, hardware delays create customer satisfaction risk that can affect future engagements.

Tariff and geopolitical exposure. Cree LED manufacturing in Huizhou, China is exposed to US-China tariff escalation. Memory supply chain dependence on Samsung and SK Hynix (Korean suppliers) creates sensitivity to Korean supply disruption risk. The SK Telecom / SK Hynix strategic partnership (a $200M investment from SK Telecom, closed December 2024, which brought memory and AI data center collaboration) creates relationship upside but also concentration in a single geopolitical region.

9. Walk the Talk

Concalls used in this analysis:

1. Q3 FY2025 - July 8, 2025 (CEO: Mark Adams)
2. Q4 FY2025 - October 7, 2025 (CEO: Mark Adams)
3. Q1 FY2026 - January 6, 2026 (CEO: Mark Adams, final call before retirement)
4. Q2 FY2026 - April 1, 2026 (CEO: Kash Shaikh, first call)

The Q2 FY2026 call was held 52 days before today's date - the most recent reporting period criterion is satisfied.

Starting with Q3 FY2025 (July 8, 2025): Mark Adams entered the call with precise commitments. He raised EPS guidance from a range of $1.60 plus or minus $0.10 to approximately $1.80 plus or minus $0.05 - a narrowed, higher range with limited room for miss. He also reaffirmed revenue growth of 17% for the full fiscal year (plus or minus 2%) and tightened segment guidance: Advanced Computing 15-25%, memory 25-30%, LED approximately flat. These were not vague aspirations. They were numbers with tight error bars for a company in the middle of a business model transformation.

Every one of those commitments was kept. Full-year FY2025 delivered precisely 17% revenue growth. Non-GAAP diluted EPS came in at $1.90 - above the $1.80 midpoint of the raised guidance. Advanced Computing grew approximately 17% for the full year, inside the 15-25% band. Memory grew 30% for the full year, exactly at the top of the guided 25-30% range. LED was essentially flat. Execution against guidance was complete.

"We are now seeing signs that we have entered the initial stages of that growth in corporate build outs at scale." - Mark Adams, Q3 FY2025 concall (July 8, 2025)

This prediction - that large-scale enterprise AI buildouts were beginning - was credible in July 2025 and has been validated by the Q2 FY2026 memory segment performance (+63% year-over-year) and the Q2 customer win announcements.

Q4 FY2025 (October 7, 2025): Adams opened FY2026 guidance at 6% revenue growth (plus or minus 10%) with non-GAAP EPS of $2.00 (plus or minus $0.25). The wide ranges were deliberately acknowledged by management as reflecting genuine uncertainty: the 14-percentage-point headwind from Penguin Edge wind-down and hyperscale hardware exit was being offset by enterprise AI growth and memory expansion, and the timing of enterprise deployments is inherently hard to forecast. Memory was guided at only 10-20% growth for the full year - conspicuously conservative given that Q4 FY2025 itself had just delivered 38% memory growth.

Adams also announced the $75M incremental stock buyback authorization (bringing total remaining to $112M) and flagged that H2 FY2026 was expected to be heavier than H1. He highlighted the $200M debt reduction accomplished in the quarter, bringing the company to near-zero net debt.

Q1 FY2026 (January 6, 2026): Mark Adams' final earnings call maintained FY2026 guidance exactly: 6% growth, $2.00 EPS. The Q1 delivery (revenue flat at $343M, EPS of $0.49 versus $0.41 consensus) was clean execution - not exciting but not problematic. Adams confirmed signed master service agreements with major financial and oil/gas institutions and described purchase orders as expected "shortly." He and CFO Olmstead explicitly guided Advanced Computing down sequentially from Q1 to Q2 FY2026. This proved accurate in Q2.

One notable moment of transparency: Olmstead said Advanced Computing would be "down Q1 to Q2" sequentially, and it was - moving from 44% of revenue to 34%. This kind of specific, near-term negative guidance that turns out to be accurate is a credibility signal. Management did not smooth or hedge.

Q2 FY2026 (April 1, 2026): Kash Shaikh's debut call delivered a guidance raise driven by an extraordinary memory performance (+63% year-over-year) combined with a worse-than-expected Advanced Computing result. The full-year revenue growth guidance moved from 6% to 12%; EPS from $2.00 to $2.15; memory segment from +20-35% to +65-75%. The raise is legitimate - memory is genuinely performing far above the initial guidance.

But an honest reading of the FY2026 guidance history reveals a pattern. Initial memory guidance of 10-20% growth was issued when the trailing Q4 FY2025 had just delivered 38% growth. Either management had visibility suggesting the memory cycle would decelerate sharply, or they were deliberately conservative. By Q2 FY2026, with memory up 63%, the initial guidance looks like conservative sandbagging rather than an accurate forecast.

The Advanced Computing guidance has moved in the opposite direction: from -15% to +15% initially, to -25% to -15% in Q2 FY2026. The segment continues to face timing pressure from converting signed agreements to revenue-generating deployments. Management's explanation is structural (3-6 month lags between bookings and revenue recognition) and the pipeline is growing, but the repeated downward revisions to Advanced Computing guidance suggest the business is harder to predict than management initially signaled.

Net assessment: Adams built a track record of precise delivery. EPS guidance consistently beaten, full-year revenue delivered exactly, segment ranges hit. The initial FY2026 memory guidance appears in retrospect to have been conservative relative to available evidence. Shaikh has inherited a business that was beating expectations and has chosen to raise guidance aggressively in his first call - either building genuine momentum or establishing a high bar early. With only one Shaikh call on record, assessment of his calibration must wait. Adams was a management team that did what they said. The FY2026 numbers point the same direction.

10. Shareholder Friendliness Index

Penguin Solutions has paid no dividend in its history as a public company. DPS for FY2023, FY2024, and FY2025 is $0.00. Management has not signaled any intention to initiate a dividend, and given the company's focus on AI infrastructure growth investment and the recent shift to near-zero net debt, a dividend initiation in the near term is unlikely.

On buybacks, the company has been meaningfully active. In FY2025, Penguin repurchased approximately 2.31 million shares (roughly 4.3% of outstanding stock) for $38.49M, completing its prior authorization program between June and August 2025. In October 2025, the board authorized an additional $75M repurchase program, bringing total remaining authorization to approximately $112M. The diluted share count across FY2023-FY2025 has moved in a narrow 52-55M range: approximately 52.2M diluted in FY2023, rising to approximately 54.3M in FY2024 (reflecting stock-based compensation issuance), then declining modestly to approximately 53.7M in FY2025 as buybacks began to offset new issuances. As of Q2 FY2026, the diluted count has been guided at approximately 53M shares. Net reduction is modest - buybacks are roughly keeping pace with RSU dilution rather than meaningfully shrinking the share count.

Verdict: Neutral. Capital is being returned through buybacks, the program is real, and management repaid $200M in debt in Q4 FY2025 while simultaneously accelerating buyback activity - demonstrating prioritization of balance sheet optimization over capital hoarding. But RSU-based executive compensation continuously issues new shares that offset most of the buyback benefit, leaving shareholders with only modest net count reduction.

11. Insider Activities

Source: SEC Form 4 filings via EDGAR, accessed through StockTitan, MarketBeat, and open filing search records. (Form 4 filings for each transaction.)

Recent transactions - last 12 months (most recent first):

Zero open-market purchases over the 12-month period. MarketBeat confirms 0 insider buys.

Reading the sells:

Clark Joseph Gates (SVP, President of Optimized LED) is the most frequent seller, with all transactions executing under a Rule 10b5-1 plan adopted November 11, 2025. The pre-scheduled nature of these sales removes virtually all information content - Gates set the plan up months before execution and the sales ran regardless of stock price movement (which explains sales occurring at prices from $17 to $40 per share across the period). The pattern is consistent with an executive managing a long-accumulated equity position through planned diversification, not a negative view on business prospects.

Anne Kuykendall (SVP and Chief Legal Officer) similarly adopted a 10b5-1 plan on November 11, 2025, and executed one open-market sale of 4,000 shares at $17.75 in March 2026 plus routine tax withholding on RSU vesting. The March 2026 sale, in retrospect executed at a price well below subsequent highs, is evidence that 10b5-1 plans are truly pre-scheduled and not calibrated to current views.

Jack Pacheco (EVP) sold 63,447 shares at $24.83 in July 2025, the largest single dollar transaction of the period at approximately $1.58M. No 10b5-1 plan is indicated in the available filing data. The stock was trading at approximately $25 in July 2025, significantly below current prices. This appears to be a personal liquidity event - a senior executive with years of RSU accumulation taking cash from a concentrated equity position. Not a negative signal given the price level and the personal diversification rationale.

The Straub and Nayyar sales in May 2026 warrant attention. Director Maximiliane C Straub sold 15,000 shares across three transactions (May 11, 12, 14) for approximately $684,000. Director Sandeep Nayyar sold 20,000 shares across two transactions (May 11, 12) for approximately $890,000. Neither sale was filed as being under a 10b5-1 plan. These are independent directors with no operational role, selling at prices between $43 and $49 per share - a period when the stock had risen approximately 89% over the prior 30 days per search data. The most probable explanation is profit-taking at substantially elevated prices following what appears to be a significant re-rating of the stock, not a negative view of business fundamentals. Directors who have held shares for years at $20-25 levels and find themselves at $45-49 represent a natural diversification impulse, not an insider judgment about future earnings.

New CEO Kash Shaikh: Shaikh holds approximately 376,000 shares, acquired through his employment RSU grants. No open-market purchase has been made since joining in February 2026. His equity position is substantial but entirely compensation-driven - he has not put personal capital at risk. This is a neutral signal for a CEO who joined 3.5 months ago; it is neither unusual nor meaningful at this stage.

Net assessment: Twelve months of insider activity shows zero open-market buying and approximately $4.14M in aggregate selling. The selling is predominantly pre-scheduled (10b5-1 plans) or traceable to profit-taking at elevated stock prices following a dramatic run-up. No red flags: the selling is not concentrated in the hands of the CEO or CFO, the 10b5-1 plans were set up months in advance, and the two director sales that lack 10b5-1 protection are consistent with routine profit-taking rather than informed negativity. The absence of any open-market buying by any insider, including the new CEO, is the mild concern - no one with inside access to the business pipeline and technology roadmap has chosen to invest personal capital alongside shareholders. Assessment: neutral.

12. Scenarios

Bull case

Inference computing becomes the dominant AI workload of 2026-2028, and enterprises discover that production LLM serving requires substantially more memory bandwidth than general-purpose GPU servers provide. Penguin's strategic bet - that memory is the binding constraint in inference, not compute - proves correct across multiple customer verticals. The MemoryAI KV Cache server, validated by the Tier 1 financial institution deployment, becomes a standard reference architecture for on-premise LLM serving at financial institutions, healthcare providers, and defense contractors. The Optical Memory Appliance ships on schedule in late 2026, provides a genuine performance breakthrough for GPU clusters, and enters volume production in 2027 with 20+ enterprise customers.

The 7 H1 FY2026 new customer logos convert to full deployments across H2 and FY2027, building a diversified enterprise AI/HPC base that no longer depends on any single customer or vertical. Kash Shaikh's software-platform vision gains traction: ClusterWareAI becomes the management software of choice for on-premise enterprise AI clusters in financial services and federal, generating subscription renewal revenue that compounds on top of hardware replacements. CXL memory products achieve mainstream server adoption faster than the semiconductor industry currently projects, and Penguin's early engineering investment in CXL positions it as the default supplier for specialty CXL modules.

Stratus lands a large production AI inferencing contract with a major bank, validating the "mission-critical AI" proposition and opening a new customer archetype. The LED business is either divested to a strategic acquirer at a reasonable multiple, removing the margin drag and management distraction, or stabilizes with structural cost reductions that preserve gross profit while the AI infrastructure business scales. Penguin re-rates from a memory and HPC conglomerate to an AI infrastructure platform company, with recurring software and services revenue establishing a floor under earnings volatility.

Base case

Enterprise AI deployment ramps steadily but unevenly through FY2026-2027. The H1 FY2026 customer wins convert to deployment revenue over 12-18 months as expected. Memory remains strong through FY2026 driven by continued AI infrastructure buildout demand, but pricing moderates as memory manufacturers add capacity in H2 2026 and the most acute supply tightness eases. The Integrated Memory segment's growth rate decelerates from 63% year-over-year to a more sustainable 20-30% as pricing normalizes.

The OMA ships in early 2027 to initial customer acceptance, primarily among research institutions and early-adopter financial services firms. Volume scale takes 18-24 months beyond first shipment. Advanced Computing non-hyperscale AI/HPC grows at 40-50% annually for the next two years, offsetting the ongoing LED decline. Stratus maintains its existing customer base with stable support revenue and limited new wins in production AI, which turns out to be a longer adoption cycle than management hoped. Kash Shaikh refines the AI factory platform strategy through FY2026, with concrete software product milestones emerging by the Q3 or Q4 FY2026 call. The company delivers full-year FY2026 guidance (12% revenue growth, $2.15 EPS midpoint) and gives FY2027 initial guidance in October 2026 that shows continued but not accelerating progress.

Bear case

The enterprise AI deployment cycle encounters a significant air pocket. CFOs at major enterprises, watching hyperscaler cloud AI inference pricing fall as training-focused infrastructure is repurposed for inference, conclude that on-premise AI infrastructure is capital-inefficient and delay or reduce their capex plans. Penguin's 7 H1 FY2026 new logos do not convert to material purchase orders on the expected 12-18 month timeline - they slip to 24 months, and some do not convert at all as enterprise AI programs are restructured.

Advanced Computing revenue continues disappointing not just from the intended exits but from genuine enterprise demand softness. Simultaneously, memory pricing reverses sharply in H2 2026 as memory manufacturers add capacity in response to the AI infrastructure boom - the same cyclical overcapacity pattern that has hit the memory industry every 4-5 years. With memory pricing down 20-30%, the Integrated Memory segment's revenue and margins compress simultaneously, eliminating the tailwind that offset Advanced Computing weakness in FY2026.

Cree LED's decline accelerates under tariff escalation and Chinese competitive pressure. The OMA launch slips to mid-2027 due to supply chain and engineering delays, and CXL ramp is slower than hoped as customers wait for the ecosystem to mature. Kash Shaikh, encountering a harder business than his Securonix software background prepared him for, initiates a strategic review that creates organizational uncertainty. Management misses FY2026 guidance for the first time under the Penguin Solutions brand, and investors reprice the company as a collection of moderately declining legacy businesses rather than an AI infrastructure platform.

Sources:

• Penguin Solutions Q2 FY2026 Earnings Press Release - SEC 8-K
• Penguin Solutions Q1 FY2026 Earnings Press Release - SEC 8-K
• Penguin Solutions Q2 FY2026 Earnings Call Transcript - Insider Monkey
• Penguin Solutions Q4 FY2025 Earnings Call Transcript - Motley Fool
• Penguin Solutions Q3 FY2025 Earnings Call Transcript - Insider Monkey
• Penguin Solutions Q1 FY2026 Earnings Call Transcript - Motley Fool
• Penguin Solutions 10-K FY2025 - StockTitan
• Penguin Solutions Company Website - About Us
• Penguin Solutions CEO Transition Announcement
• Penguin Solutions Q2 FY2026 Results - BusinessWire
• SMART Global Holdings acquires Penguin Computing for $85M - Seeking Alpha
• SMART Global Holdings Completes Acquisition of Cree LED - Cree LED
• SGH Completes Acquisition of Stratus Technologies - Penguin Solutions IR
• PENG Insider Trading - MarketBeat
• Penguin Solutions PENG Director Insider Sale - StockTitan
• IDC: Dell leads server market driven by AI infrastructure needs - Network World
• Gartner AI-Optimized IaaS Growth Forecast
• SMART Modular CXL NV-CMM Achieves Compliance - Inside HPC
• Stratus ztC Endurance Second Generation - Penguin Solutions IR