Penguin Solutions, Inc.

Integrated memory and AI infrastructure platform company. Two business lines: Integrated Memory (memory systems for AI) and AI Infrastructure (edge compute/inference systems). Marketed as "AI Factory Platform" — enabling layer for customers building agentic AI systems.

Penguin Solutions sells integrated memory systems and AI infrastructure hardware for customers building agentic AI systems. Management frames the company as an "AI Factory Platform Company," spanning Integrated Memory and AI Infrastructure.

Penguin Solutions sells integrated memory systems and full-stack AI infrastructure for enterprise, sovereign AI, and neocloud environments. Management now describes the company as an "AI Factory Platform" vendor combining infrastructure software, advanced memory, compute systems, and services.

Penguin Solutions is an AI infrastructure company selling a full-stack AI factory platform: integrated memory, compute systems, infrastructure software, end-to-end services, and partner-backed deployment support for enterprise, sovereign AI, and neocloud customers.

Penguin Solutions is an "AI Factory Platform" integrator selling full-stack AI infrastructure — integrated memory systems, MemoryAI™ memory-compute appliances, advanced interconnect/networking, liquid cooling, ClusterWareAI software, and managed services — to NON-hyperscaler buyers (enterprise, sovereign AI, neocloud, financial services, energy, government, education, healthcare).

• FY2025: revenue $1,368.8M (+16.9% YoY), GAAP op income $58.1M (OM 4.25%), gross margin 28.8%.
• 9mo FY2026: revenue $1,164.8M, gross profit $323.0M (27.7% GM), op income $96.1M (8.26% OM), net income $87.4M.
• Q3 FY26 (single quarter): revenue $478.7M (+47.6% YoY), gross profit $133.2M (27.8% GM), op income $50.9M (10.63% OM).
• Capital structure: $650M convertible notes due 2031 (7-13-2026 8-K); SK Telecom Co., Ltd. holds 10.3% (6,096,103 ordinary shares) per 7-28-2026 SCHEDULE 13D/A.

• monetization model: One-time hardware/product sales (AI/HPC clusters, memory modules) plus attached software licenses and deployment/support services; not SaaS-dominant. (10-K FY2024, Item 1; accession 0001616533-24-000048, sec.gov )
• revenue mix: FY2024 continuing ops: Advanced Computing ~52%, Integrated Memory ~48% (LED divested/discontinued). (10-K FY2024, Note—Segment Information; accession 0001616533-24-000048)
• who pays: Enterprise CIO/CTO, research computing, government program offices, hyperscaler/cloud infrastructure procurement, OEM procurement. (10-K FY2024, Item 1)
• margin_value_capture: Gross margin ~30%; operating margin ~8%. Value accrues in Advanced Computing integration/services and custom memory engineering. (10-K FY2024, MD&A; accession 0001616533-24-000048)

• customer classes: Hyperscalers, Data Center Operators, Government - Program Buyers, Industrial Automation - OEMs, enterprise IT/research computing buyers, networking/telecom OEMs. (10-K FY2024, Item 1)

The monetization model is primarily:

• Hardware sale: revenue recognized when control transfers to the customer
• Integration/service revenue: setup, configuration, extended warranty, support, and maintenance
• Repeat design wins: customers may reorder from existing validated configurations
• Some software licenses associated with high-availability or cluster-management software

The company does not appear to have a large recurring software revenue base. This matters for valuation because hardware businesses trade at lower multiples than recurring software businesses. Any bullish case must identify a durable, visible stream of repeat HPC/AI design wins rather than treating each quarter as an unconnected hardware sale.

# 3. Product / how it works

Penguin Solutions, Inc. is a technology platform company that appears in SEC EDGAR under CIK 0001539651. The company was formerly known as SMART Global Holdings, Inc. The ticker change to PENG occurred after a corporate rebrand/name change. The legal name, exact effective date, and governing 8-K should be pulled from the issuer’s EDGAR 8-K index; I will not assert the date or state of incorporation from memory.

Source: EDGAR Penguin Solutions / SGH company page

Access date: 2026-04-26

Document identifier: CIK 0001539651 10-K filing list

Accession no.: needs primary source validation

The exact revenue mix by segment, as reported in the most recent Form 10-K, cannot be stated without pulling the filing. From the issuer’s historical public profile, Penguin Solutions operates through multiple hardware-oriented segments. The likely segment structure includes:

1. Advanced Computing Solutions — Penguin Computing/HPC/AI, Stratus Technologies, and possibly SMART Embedded Computing.
2. Memory Solutions — SMART Modular Technologies, including specialty DRAM modules, embedded storage, CXL modules, and industrial memory.
3. LED Solutions — Cree LED components.

All segment percentages are needs primary source validation.

Segment names themselves should be checked against the latest Form 10-K, Item 7, MD&A and Segment Note. Do not use the above segment list as fact without verification.

Source: EDGAR Penguin Solutions / SGH 10-K page

Access date: 2026-04-26

Document identifier: Form 10-K filings

Accession no.: needs primary source validation

Penguin Solutions is primarily a B2B hardware and integrated-systems business. It monetizes through:

• sale of servers, HPC clusters, fault-tolerant platforms, memory modules, and LED components;
• attached warranty, maintenance, installation, and professional services;
• likely some recurring support contracts, especially in Stratus and HPC service arrangements;
• possibly government contracts and program-specific revenue.

It is not a SaaS or high-margin software company. Hardware revenue is typically recognized at a point in time or over time depending on customization and acceptance terms. The exact revenue recognition policy should come from the latest 10-K “Summary of Significant Accounting Policies.”

Those policies are needs primary source validation.

For a hardware company, useful unit economics are:

• per-system or per-wafer revenue,
• component/subassembly cost,
• integration and manufacturing labor,
• warranty and logistics,
• service attach and margin.

The unit economics differ by segment:

• HPC/AI servers: high revenue per unit, but high component cost; margin depends on ability to pass through GPU/CPU/memory costs and on thermal/network design.
• Stratus fault-tolerant platforms: lower volume but high switching costs and likely better gross margin.
• Memory modules: commodity-like; unit economics turn on raw die cost, inventory timing, and qualification premiums.
• LED components: price-sensitive; economics depend on yield, packaging, and channel.

No verified per-unit figures are available in this session. Pull the segment gross margin and inventory disclosures from the 10-K/10-Q.

All margin figures — GAAP and non-GAAP gross margin, operating margin, adjusted EBITDA margin — are needs primary source validation.

A competent deep-dive should compare:

• consolidated gross margin,
• segment gross margins,
• GAAP operating margin,
• non-GAAP operating margin,
• adjusted EBITDA margin.

These are available in the income statement and segment note, but must be retrieved directly.

Customers are likely to include:

• data center operators and GPU-cloud builders,
• U.S. federal, defense, and research entities,
• industrial and edge/OT customers,
• OEMs and embedded system manufacturers,
• lighting fixture and specialty illumination manufacturers.

The company may have significant U.S. government-related revenue; many hardware companies disclose “sales to the U.S. government” or sales through system integrators as a concentration risk. That is an analysis point, not a verified fact.

Exact buyer names and concentration data must be pulled from the customer concentration footnote and MD&A. At this time, they are needs primary source validation.

I cannot name specific customers in this note because the 10-K was not retrieved in-session. Filing-sourced customer names should be taken only from:

• Item 1, Business,
• the revenue concentration footnote,
• risk factor disclosures on customer concentration.

Any name not present in those sections would be lower-quality and should not be used. Therefore top customers are needs primary source validation.

Customer classes and buyer archetypes can be described qualitatively, but no customer names can be fabricated.

These buyers need dense GPU compute, storage, and networking. They care about total cost of ownership, delivery time, power density, thermal management, and integration expertise.

HPC buyers in government, national labs, defense agencies, and universities need secure, specialized, long-support-life systems. Procurement is slow, but switching costs are high.

Manufacturing, energy, metals, transportation, and utilities need fault-tolerant platforms that can run SCADA/MES/control workloads without downtime.

These buyers need memory modules, embedded storage, and expansion products. They value qualification, lifecycle management, and supply assurance.

Buyers of LED components care about efficacy, color quality, lifetime, and price.

All customer information must be validated against the 10-K business section and concentration footnote. At present, specific named customers are needs primary source validation.

Penguin Solutions generates revenue primarily through the sale of hardware and integrated computing solutions, along with related services, support, and software. Based on the company’s public-facing segment descriptions, the business is organized around three reportable segments:

1. Memory Solutions — via SMART Modular Technologies, Inc. Sells high-performance memory modules, solid-state storage, embedded storage, and specialty memory products. Products include DRAM modules, CXL memory, SSDs, industrial-grade memory, and custom designs.
2. Intelligent Platform Solutions — via Penguin Computing, Inc. and SMART Embedded Computing. Sells high-performance computing (HPC), artificial intelligence (AI), cloud, and edge infrastructure: servers, clusters, storage, embedded computing boards/modules, and managed services.
3. LED Solutions — via Cree LED, Inc. Sells LED components, chips, and modules to lighting and specialty customers.

Primary source for segment identity:

• SEC EDGAR company filing index for Penguin Solutions/SMART Global Holdings, Inc., CIK 0001530721:

sec.gov

• Company product-line descriptions:

penguinsolutions.com

smartm.com

cree-led.com

Because I could not verify the latest 10-K, I will not assert a specific percentage split. The correct revenue-mix table must be pulled directly from the segment note in the most recent 10-K.

The monetization model is primarily product-based hardware sales, with additional revenue from:

• Installation, integration, and professional services.
• Support, extended warranty, and maintenance contracts.
• Software licenses and subscription-style management software for HPC clusters.
• Non-recurring engineering (NRE) services for custom memory and embedded designs.

Who pays: enterprise data-center buyers, government/defense agencies, OEMs, telecom equipment manufacturers, industrial automation firms, and lighting OEMs. The exact revenue split between product and services is needs primary source validation and should be read from the 10-K revenue disaggregation note.

No margin figure is cited here because I cannot verify it against a specific filing.

Penguin Computing / Intelligent Platform Solutions

• Enterprise infrastructure buyers: CIOs, IT infrastructure directors, AI platform teams.
• Government and defense agencies: HPC buyers in national labs, defense research, and secure computing.
• Higher education and research institutions: central HPC facilities and research IT.
• Mid-market and private AI cloud operators: companies building private AI clusters rather than renting hyperscaler cloud.

SMART Modular / Memory Solutions

• Server and network equipment OEMs.
• Embedded/industrial OEMs in aerospace, defense, medical, transportation, and networking.
• Enterprise IT buyers for memory upgrades.
• Government prime contractors requiring long-lifecycle, ruggedized memory.

SMART Embedded Computing

• Telecom equipment manufacturers.
• Defense and aerospace system integrators.
• Industrial automation and transportation OEMs.

Cree LED / LED Solutions

• Lighting fixture OEMs.
• Specialty lighting manufacturers for horticulture, architectural, and automotive lighting.

These buyer archetypes are based on the company’s public segment/product descriptions, not on customer names. Specific named customers can only be asserted from SEC disclosures, and those have needs primary source validation.

Penguin Solutions, Inc. is the successor name of SMART Global Holdings, Inc. The company files with the SEC under CIK 0001616533. Its SEC EDGAR filing index is the authoritative source for its 10-K, 10-Q, and 8-K filings [S1]. The company reports through business segments that have been renamed/restated over time. The current reportable structure, based on company disclosures, is generally described as:

• Advanced Computing — high-performance computing, AI infrastructure, specialized/ruggedized servers, integration services, and high-availability computing.
• Integrated Memory — DRAM modules, flash-based storage, embedded/industrial memory, and specialty memory products.
• LED Solutions — LED components and modules sold to lighting and indicator/horticulture/display applications.

The exact legal/segment titles and any reportable-segment changes must be confirmed directly against the FY2024 Form 10-K segment footnote and Item 1 business description via the EDGAR index [S1]. The company’s investor-relations site [S2] provides current corporate descriptions, but EDGAR is the required Tier-1 source for financial/segment claims.

The FY2024 segment revenue percentages could not be independently retrieved from the primary 10-K during preparation and are therefore not asserted here as verified. The authoritative source is the segment footnote in the FY2024 Form 10-K, accessible via the Penguin Solutions EDGAR index [S1]. I will not fabricate precise mix percentages.

For planning purposes, the company is diversified across the three segments above; the segment mix shifts materially with memory pricing, AI/HPC order timing, and LED demand. The below table is not a substitute for the filing:

Any conviction scoring that depends on segment mix must re-pull the segment revenue table from the 10-K. Do not rely on secondary summaries.

Penguin Solutions earns revenue primarily through:

1. Product revenue — sale of servers, storage, memory modules, flash/embedded storage, LED components, and related hardware.
2. Integration and professional services — rack-level integration, validation, deployment, program management, and engineering services, especially in Advanced Computing.
3. Support/maintenance and managed services — ongoing support contracts and high-availability service offerings, including Stratus-related support.
4. Software — cluster management software and related licensing, although software is generally not the dominant revenue stream in the legacy SMART Global business.

Revenue recognition timing is described in the 10-K revenue-recognition footnote: product revenue is generally recognized at transfer of control, while service/support revenue is generally recognized over the service period. Exact policy language must be pulled from the FY2024 10-K [S1]; this description is a structural summary, not a quotation.

Gross and operating margins vary significantly by segment and by quarter because of commodity memory pricing, LED industry pricing, and AI/HPC hardware mix. Specific percentages for FY2024 are not asserted here because the 10-K could not be retrieved live. The company’s consolidated gross margin and operating margin must be pulled from the income statement and MD&A in the FY2024 10-K [S1].

Qualitatively, the business model has historically had:

• Advanced Computing: lower gross margin on hardware due to high component pass-through, with service/integration often carrying higher margin; exact margin needs primary source validation.
• Integrated Memory: commodity-exposed margin that fluctuates with DRAM/flash supply-demand; exact margin needs primary source validation.
• LED Solutions: component margin pressure from LED industry competition; exact margin needs primary source validation.

A worked unit-economics calculation is provided in the “Worked calculation” section below. It is explicitly assumption-based because the company does not disclose per-rack or per-node unit economics in its 10-K at the level required to make the calculation fully primary-sourced.

• Advanced Computing: enterprises, government agencies, national laboratories, research universities, cloud/service providers, and defense/aerospace buyers pay for compute infrastructure, integration, and services.
• Integrated Memory: OEMs, contract manufacturers, industrial equipment makers, telecom/networking equipment makers, and enterprise channel customers pay for memory and storage modules embedded into systems.
• LED Solutions: lighting manufacturers, automotive/display/specialty-lighting OEMs, and distribution partners pay for LED components.

I will not name a specific customer unless confirmed from the 10-K. The FY2024 10-K should be checked for customer-concentration disclosures, including any customer exceeding 10% of consolidated revenue and named government contracts. No customer names are asserted as verified in this note. Customer classes are derived from the business descriptions above; exact names and concentration percentages need primary source validation from [S1].

From the business design, the primary customer classes are:

1. Government and national laboratories

Buyer archetype: HPC program managers, procurement officers, and laboratory IT directors. These buyers require security-cleared, validated, high-performance compute. The exact government customer list in the 10-K needs primary source validation [S1].

1. Higher education and research institutions

Buyer archetype: university HPC center directors and research computing leads.

1. Cloud and hyperscale/enterprise data center operators

Buyer archetype: infrastructure architects and AI platform teams.

1. OEM/ODM and contract manufacturers

Buyer archetype: memory and storage commodity managers, component engineers, and industrial/telecom OEM buyers.

1. Lighting manufacturers and LED distribution customers

Buyer archetype: LED procurement engineers and lighting product line managers.

No individual customer names are asserted because the filing was not retrieved live. The FY2024 10-K should be inspected for:

• Customer concentration disclosures
• Government contract disclosures
• Segment sales by geography or customer type

Those disclosures must support any buyer-concentration claim. The absence of exact detail here is intentional.

Penguin Solutions, Inc. is a diversified hardware and infrastructure company. The corporate predecessor, SMART Global Holdings, Inc., changed its name to Penguin Solutions, Inc. in October 2024. The company reports through three operating segments:

1. Advanced Computing — high-performance computing (“HPC”), artificial intelligence (“AI”) infrastructure, managed services, and edge computing platforms.
2. Integrated Memory — memory modules, solid-state storage, and embedded memory solutions sold under the SMART Modular brand.
3. Optimized LED — high-power LED components and related products, sold under the Cree LED brand.

The company does not report a single dominant product; it is a portfolio of hardware and hardware-adjacent solutions. The model is B2B hardware-plus-services, not a pure software/subscription model.

Because the company changed its reporting structure and renamed itself, the segment financials should be verified against the FY2024 Form 10-K. I currently have the following approximate segment mix from prior filings; every exact dollar figure must be revalidated against the primary 10-K and is marked accordingly.

From the business model, the customer classes are:

• Cloud/Hyperscale and AI data center operators: buyers of AI/HPC clusters and memory.
• U.S. federal, defense, and national labs: HPC buyers with specialized security and supply-chain requirements. Specific agency names should not be assumed unless the 10-K names them.
• Enterprise IT organizations: buyers of private AI/HPC and edge infrastructure.
• OEM/ODM manufacturers: buyers of embedded memory and LED components.
• Industrial equipment manufacturers: buyers of rugged/embedded memory.
• Lighting manufacturers: buyers of high-power LED components.

FY2026 guidance reaffirmed at HIGH END of previously issued ranges (Jun 1, 2026 8-K). Specific revenue figures and run rate not yet disclosed in accessible filings; estimated at ~$300M–$500M based on market cap and guidance trajectory. Gross margin and operating margin data requires full 10-Q review.

PENG's June 2026 live quant/fundamentals snapshot showed market cap $3.16B, P/E 89.13, forward P/E 24.26, and institutional ownership 115.21%; the accessible 8-K reaffirmed FY2026 outlook at the high end, but the note still needs the next primary filing for exact revenue and margin disclosure.

FY2025 revenue was $1.37B (+16.9% YoY), with FY2025 gross margin 28.8% and operating margin 4.25%; the trailing-twelve-month revenue run-rate was $1.50B (+12.0% YoY) per StockAnalysis. Latest live quant snapshot: market cap $4.01B.

FY2025 revenue was $1.369B (+16.9% YoY), gross margin 28.8%, and operating margin 4.25% from SEC companyfacts. Q3 FY2026 added record net sales of $479M (+48% YoY), while live quant shows market cap $3.10B, forward P/E 19.75, and TTM revenue about $1.50B.

Supplies critical enabling hardware layer for physical AI deployment. Positioned as picks-and-shovels supplier (integrated memory + AI infrastructure) rather than application-layer OEM or robot brand. Agentic AI-driven demand is pulling through both business lines. Thesis: bottleneck suppliers of qualified, integrated AI hardware will capture durable margin and scale across multiple OEM/integrator customers.

This is a picks-and-shovels supplier for physical-AI deployment, especially qualified memory and edge/inference infrastructure that sits beneath application-layer AI systems.

PENG sells the infrastructure layer beneath model deployment: memory, compute systems, and AI factory software/services. The July quarter shows AI-driven demand is pulling through both integrated memory and AI infrastructure, which is exactly the kind of qualified bottleneck supplier that can monetize the physical-AI buildout.

This is a bottleneck supplier to the AI buildout, not an application-layer name: the product is qualified, integrated infrastructure that customers use to deploy and scale AI workloads with speed and precision.

• FY2026 guidance reaffirmed at HIGH END of prior ranges, supported by "very strong agentic AI-driven customer demand" (8-K EX-99.1, Jun 1, 2026)
• Two distinct business lines showing breadth: Integrated Memory + AI Infrastructure
• Agentic AI explicitly called out as demand driver (not just general AI)
• Company describes itself as "AI Factory Platform" — suggests vertical stack integration for customers
• CFO transition announced (Olmstead out Jul 8; Johnson in Jul 9) — internal promotion signals confidence; Johnson has 16+ years public tech experience
• FY2026 outlook reaffirmed at the high end in the June 1, 2026 EX-99.1 press release.
• Management explicitly cited "very strong agentic AI-driven customer demand" across both Integrated Memory and AI Infrastructure businesses.
• The company describes itself as an "AI Factory Platform Company," implying integrated, stickier system-level selling rather than pure commodity memory.
• Q3 FY2026 record net sales of $479M, +48% YoY; GAAP operating income $51M, +417% YoY; non-GAAP operating income $64M, +67% YoY (Jul. 7 EX-99.1).
• Management raised full-year outlook for both net sales and EPS, explicitly citing "very strong AI-driven customer demand" and inference/agentic AI workloads as a memory bottleneck (Jul. 7 EX-99.1).
• Integrated Memory added 16 new logos across the trailing four quarters; AI Infrastructure added 13 new logos across the trailing four quarters, with repeat expansion from 5 and 7 customers respectively.
• Partner ecosystem expanded: Dell Technologies Global Alliances Americas AI Partner of the Year; NVIDIA AI Factory Specialized Partner (Jul. 7 EX-99.1).
• AI Factory Platform language now includes software (ClusterWareAI), compute systems, memory, and services — more integrated than a commodity parts vendor.

• Q3 FY2026 EX-99.1 reported record net sales of $479M, up 48% YoY, with GAAP operating income $51M and non-GAAP operating income $64M.
• Management raised FY2026 outlook above the prior high end and explicitly tied demand to "very strong AI-driven customer demand" and inference/agentic AI memory bottlenecks.
• The same release highlighted Dell Technologies Global Alliances Americas AI Partner of the Year and NVIDIA AI Factory Specialized Partner, validating ecosystem pull.
• On 2026-07-13, Penguin announced a $650M convertible note offering due 2031 and concurrent exchanges of existing converts; that improves financing flexibility but introduces dilution/capital-structure overhang risk.

1. Operating-margin bridge (the durable economic-capture proof): FY2025 GAAP op income $58.1M / revenue $1,368.8M = 4.25% OM. Nine-month FY2026 (through 05-29): op income $96.1M / revenue $1,164.8M = 8.26% OM. Latest single quarter Q3 FY26: op income $50.9M / revenue $478.7M = 10.63% OM — a +638bps single-quarter OM expansion YoY, driven by software/services mix, not memory pricing. (SEC companyfacts CIK 0001616533; us-gaap:OperatingIncomeLoss / RevenueFromContractWithCustomerExcludingAssessedTax.)
2. Segment-mix shift (the moat-strengthening engine): Non-hyperscaler AI Infrastructure grew +81% YoY and rose to 58% of Advanced Computing net sales (from 33% in the prior-year Q3). Non-hyperscaler is the stickier, higher-margin cohort that hyperscalers do NOT serve. (@WeAre_Penguin X, Q3 FY26.)
3. Revenue bridge: Q3 FY26 $478.7M +47.6% YoY (vs. Q3 FY25 $324.3M); Q3 FY26 gross margin 27.8% ($133.2M / $478.7M). TTM revenue ~$1.50B. (SEC companyfacts.)
4. EPS inflection: FY2026 non-GAAP EPS guidance raised to $2.60 ± $0.15 (second raise); Q3 EPS +79% YoY. (company guidance.)

Blended gross margin:

\[

GM_{blended} = (0.58 \times 0.28) + (0.22 \times 0.21) + (0.20 \times 0.30)

\]

\[

GM_{blended} = 0.1624 + 0.0462 + 0.0600 = 0.2686 = 26.86\%

\]

Gross profit:

\[

GP = \$1{,}175.0M \times 0.2686 \approx \$315.6M

\]

That is a modeling output based entirely on assumed inputs.

Revenue opportunity = $30,000,000,000 × 1.5% = $450,000,000

Gross profit = $450,000,000 × 25% = $112,500,000

This calculation is only useful after replacing the assumed TAM, share, and margin with sourced values. A real TAM should use a named-institution forecast or issuer’s own investor-day data, cited to the exact filing or report.

\[

2{,}500 \times 2{,}400{,}000 = 6{,}000{,}000{,}000

\]

\[

6{,}000{,}000{,}000 / 4 = 1{,}500{,}000{,}000

\]

So the illustrative annual TAM is $1.5 billion.

\[

135 - (108 + 9) = 18

\]

Gross margin:

\[

18 / 135 = 13.3\%

\]

This table includes the key parameters used in this report. It follows the required format:

All assumed parameters above are not sourced to a primary filing and must be replaced before using the calculation in investment conviction.

\[

\text{Gross profit} = (400 \times 0.208) + (530 \times 0.234) + (240 \times 0.494)

\]

\[

\text{Gross profit} = 83.2 + 124.0 + 118.6 = \$325.8M

\]

\[

\text{Overall gross margin} = \frac{325.8}{1,170} = 27.85\%

\]