Last price
$69.62
22 Jun 2026 · NasdaqGS
The former SMART Global Holdings has reinvented itself as an AI-infrastructure systems integrator and memory specialist. The stock is up roughly 4x off its lows on a genuine memory-wall thesis. The business is real. The price has run ahead of it.
Penguin Solutions sits at the unglamorous but structurally critical layer of the AI build-out: it bolts GPUs, memory, networking, power and orchestration software into working clusters, and it sources and configures the DRAM/NAND modules those clusters are starving for. The secular tailwind is real and aligns with the situational-awareness frame of a trillion-dollar compute mobilisation. The problem is not the business; it is that at $69.62 the market is already paying 32x forward non-GAAP earnings and sitting ~40% above the sell-side mean target of $50, for a company whose headline AI-infrastructure segment is shrinking and whose blended gross margin is ~28%, not the 70%+ a software multiple would imply.
Forget the buzzword. For Penguin, Advanced Computing is the literal name of its largest reporting segment, and the work is concrete: take a customer's AI or HPC requirement, then design, build, deploy and manage the full cluster. That means racking thousands of NVIDIA GPUs, wiring high-bandwidth networking, integrating memory and storage, laying the power and cooling, and running it all on Penguin's own orchestration software (Scyld / ICE ClusterWare). The flagship proof point is Haein, Korea's largest sovereign AI cluster, 1,000+ NVIDIA B200 GPUs taken from order to production in roughly two months.
In plain terms: Penguin is an AI-factory builder and memory supplier, not a chip designer and not a cloud. It earns margin on systems integration, attached software and managed services, plus a high-volume, lower-margin memory-module business. Think "the company that assembles and runs the supercomputer," sitting one layer below the hyperscaler and one layer above the silicon.
Reported revenue has been lumpy, not linear: $1.40B (FY22), $1.44B (FY23), a cyclical trough at $1.17B (FY24), then $1.37B (FY25, +17%). FY26 is guided to roughly $1.53B (+12% midpoint), with consensus modelling closer to $1.58B. The shape tells the real story: this is a cyclical hardware-and-memory business with a software-and-services overlay, not a smooth SaaS compounder. GAAP gross margin has held in a tight 28–29% band; non-GAAP runs ~31%.
The current re-rating is driven less by these blended numbers and more by where the growth is coming from: a memory segment compounding at 60%+ and a thin but high-value CXL franchise the market is capitalising at a premium.
The sequential picture is flat at the top line (~$343M for three straight quarters) while the mix churns violently underneath: memory roaring, advanced computing falling. Non-GAAP EPS has been remarkably sticky at $0.49–$0.52 across six quarters, and the company has beaten EPS consensus for six consecutive prints. Q2 FY26 GAAP EPS of $0.58 flattered by below-the-line items; the cleaner read is the ~$0.52 non-GAAP.
Q3 FY26 (reported 7 Jul 2026) is the next catalyst, guided to ~$375M revenue and ~$0.55 non-GAAP EPS, the first quarter the company expects sequential top-line recovery.
President & CEO since 2 Feb 2026. 30+ years enterprise tech: ex-CEO Securonix & Virtana, GM at Dell, senior roles at HPE and Cisco. Succeeded long-time CEO Mark Adams.
Nate Olmstead departs 8 Jul 2026; Aaron Johnson (VP Finance) steps in as interim. Permanent search underway. Timing, one day post-earnings, is a watch item.
Founded 1988 as SMART Modular. IPO'd 2017 as SMART Global Holdings (SGH). Rebranded Penguin Solutions Oct 2024; ticker SGH to PENG.
There is no single charismatic "founder-owner" here. This is a roll-up (SMART Modular memory + Penguin Computing HPC + Stratus fault-tolerant computing + an LED business) now being repositioned around the Penguin Computing AI brand under a brand-new management team. Strategic anchor investor: SK Telecom ($200M convertible preferred, board seat), with SK hynix supplying memory.
Penguin reports three segments. The interesting tension: the segment everyone buys the stock for (Advanced Computing) is shrinking, while the one nobody talks about (Integrated Memory) is carrying the growth. The third (Optimized LED) is a legacy harvest asset most analysts model toward divestiture.
| Segment | FY25 rev | % mix | Q2'26 rev | Q2'26 YoY | Est. gross margin | Read |
|---|---|---|---|---|---|---|
| Advanced Computing | $648M | 47% | $116M | -42% | ~32-38% | Highest value, declining |
| Integrated Memory | $464M | 34% | $172M | +63% | ~28-33% | Growth engine, lower margin |
| Optimized LED | $256M | 19% | $56M | -7% | ~28-33% | Legacy, divest candidate |
Segment gross margins are not formally disclosed; ranges are analyst estimates. The mix shift from AC toward IM is the central margin story (see Section 05).
Two of the questions worth answering precisely: what part of the AI stack does Penguin occupy, and what is "Integrated Memory"? Hover the layers below. Penguin's value-add concentrates in the integration and memory layers, between the silicon and the workload.
It is the descendant of the original 1988 SMART Modular business: Penguin buys raw DRAM and NAND from the fabs (SK hynix, Micron, Samsung), then designs, assembles, tests and sells finished memory modules, server DDR5 DIMMs, and specialty memory, to OEMs and enterprises. It is a classic value-added distribution and module-build model, and right now it benefits from a price-and-volume DRAM supercycle as HBM crowds out commodity DRAM capacity.
The newer, higher-IP layer is CXL "MemoryAI": appliances that pool large external memory (up to 11TB/server) to hold the KV-cache that AI inference workloads are starved of. That is the strategic crown jewel, co-developed with SK hynix.
Mostly a headwind to operating margin today, a potential tailwind tomorrow. The raw memory-module business is structurally lower-margin (commodity passthrough, gross margins in the high-20s%), so as it grows from 34% toward ~45% of revenue, it drags blended gross margin down — exactly why FY26 non-GAAP gross-margin guidance of ~28% sits below FY25's 28.8% even as revenue grows.
But it is working-capital and asset-light at the margin and, crucially, the CXL/MemoryAI slice carries 40%+ gross margin. The bull case is mix: if CXL and attached software scale, the same "integrated memory" label flips from a margin drag into the highest-return product in the portfolio. Today it is the former.
This is the question that looks alarming on the surface, "an AI-infrastructure company whose AI-infrastructure revenue is falling 42%", and it deserves a precise answer. It is mostly deliberate, partly structural, and partly a timing lag.
A single large hyperscaler drove FY24-25 AC strength through huge, lumpy, low-margin GPU-cluster orders with zero pricing power. Management chose to walk away, guiding to roughly zero hyperscale hardware revenue in FY26. Top-10 customers were 66% of FY25 sales — concentration they are intentionally unwinding.
The Penguin Edge managed-edge line is being discontinued, removing a revenue stream from the segment with no clean one-for-one replacement yet. Optics-negative, margin-neutral-to-positive.
New enterprise / sovereign / neocloud wins carry 12-18 month sales cycles. Non-hyperscale AI revenue actually grew ~50% YoY and is now 40%+ of AC, but it has not yet scaled enough to offset the hyperscaler it replaced. Recovery is a FY27 event.
The honest read: management is trading volume for quality — swapping one giant, low-margin, unpredictable customer for a broader book of higher-margin enterprise and sovereign logos. That is the right long-term call, but it creates an air-pocket: the segment shrinks for 12-18 months before the new pipeline converts. The market is paying today as if the recovery is already de-risked. It is not.
Gross margin sits at 28.8% GAAP (~31% non-GAAP) and has been impressively stable through a violent revenue cycle. The trouble is below it: GAAP operating margin swung from 7.7% (FY22) to a trough of 1.6% (FY24) and back to 5.4% (FY25). This is a structurally low-operating-margin model — heavy on cost of integration and memory passthrough, with opex (R&D + SG&A) eating most of the gross profit.
FCF margin is the redeeming feature: free cash flow of $100M in FY25 on $1.37B revenue is a 7.3% FCF margin, and FCF has been positive every year ($85M / $65M / $58M / $100M FY22-25), comfortably converting EBITDA. Capex is light ($9M in FY25), so this is a genuinely cash-generative model despite thin accounting margins.
Memory mix is bad for gross/operating margin, neutral-to-good for FCF. Growing the lower-margin Integrated Memory segment compresses reported margins (hence sub-29% guidance), but it is asset-light and cash-converting. The lever that raises margin is the opposite of volume: CXL/MemoryAI software-and-IP mix at 40%+ gross margin, plus services attach in Advanced Computing. The bull case is a mix-driven path toward a ~40% long-term gross-margin aspiration; the base case is margins stuck in the high-20s for the next 1-2 years.
You flagged inventory growth, and you are right to. Inventory has climbed from roughly $151M (FY24) to $255M (FY25) to $322M at Q2 FY26 — a +61% YoY jump, pushing days-inventory to ~99 from the high-30s/70s.
Management frames it as strategic pre-buying: "using the balance sheet to purchase ahead where we can." With DRAM prices rising and HBM crowding out commodity supply, locking memory inventory now ahead of H2 AI demand is a rational, margin-protecting move. Crucially, accounts payable rose in lockstep (to ~$401M), so much of the build is vendor-financed, not cash out the door — operating cash flow stayed positive ($55M in Q2).
Rising inventory plus flat sequential revenue plus a declining Advanced Computing segment is the classic setup for markdown risk. If the H2 enterprise pipeline slips, that memory was bought into a falling price and a softening demand curve. This is the single cleanest bear KPI to track each quarter: inventory should fall, or revenue should rise to absorb it, by Q4 FY26 / Q1 FY27.
At $69.62, PENG carries ~2.4x EV/Sales and 32x forward non-GAAP EPS for a business growing the top line ~12% with high-20s% gross margins. That is a premium to commodity integrators (SMCI, Dell) and a discount to power/cooling darlings (Vertiv). The premium is the market capitalising the memory-wall / CXL story, not the blended financials.
| Company | EV/Sales | Fwd P/E | Gross margin | Rev growth | The comp |
|---|---|---|---|---|---|
| Penguin (PENG) | 2.4x | 32x | ~31% | +12% | AI integrator + memory + CXL IP |
| Super Micro (SMCI) | 0.8x | 12x | ~9% | +41% | Closest pure-play AI assembler |
| Dell (DELL) | 2.1x | 22x | ~19% | +22% | Scaled AI-server momentum |
| HPE | 2.1x | 12x | ~34% | +15% | Cleanest margin comp |
| Vertiv (VRT) | 12.4x | 51x | ~37% | +28% | The premium the bulls dream of |
Reverse-DCF read: to justify $69.62 on a 5-year DCF at an 11% WACC, you need roughly 14-15% revenue CAGR and operating margin expanding from ~5% GAAP toward 9-10% and a ~20x terminal multiple holding. That is the bull case priced as the base case. My base case (13% CAGR, ~7% operating margin, 18x terminal) discounts to roughly $58.
Drag the probability sliders. The model recomputes the blended 12-month target live. Defaults reflect my house view: a wide, two-sided distribution because both the memory cycle and the AC recovery are genuinely uncertain.
Probabilities auto-normalise to 100%. Each scenario price is a 12-month forward target from the DCF + multiple framework below.
| Scenario | 12-mo price | Rev CAGR | Op margin | Term. multiple | FY27 EPS | Implied return |
|---|---|---|---|---|---|---|
| Bear · 30% | $34 | +4% | 4% | 14x | $2.40 | -51% |
| Base · 45% | $58 | +13% | 7% | 18-21x | $2.80 | -17% |
| Bull · 25% | $92 | +18% | 10% | 28x | $3.30 | +32% |
Bear: memory cycle rolls, AC recovery slips, inventory written down. Base: IM strong through FY27, AC stabilises, margins high-20s. Bull: AC inflects + CXL/MemoryAI scales + short squeeze on 20% float. Blended ~$59, below today's price: the stock has discounted the base case and then some.
| Risk | Prob | Impact | Mitigant |
|---|---|---|---|
| Customer concentration (top-10 = 66%) | H | H | Deliberately diversifying into enterprise/sovereign |
| AC recovery slips past FY27 | M | H | Non-hyperscale book +50%, 7 new logos in H1 |
| Memory cycle peaks / DRAM prices fall | M | H | SK hynix allocation, vendor-financed inventory |
| Inventory markdown ($322M, 99 DIO) | M | M | Payables-funded; OCF still positive |
| Valuation de-rate (32x, +40% vs street) | H | M | Genuine CXL franchise; FCF support |
| Governance: CFO exit + new CEO | M | M | Interim CFO is 16yr public-co veteran |
| SK preferred dilution ($32.81 conv, ~6M sh) | H | M | Deep ITM; signals strategic commitment |
| Short squeeze / volatility (20% SI) | M | M | Two-sided: cuts both ways on prints |
| NVIDIA / supply allocation dependency | M | H | Multi-vendor, Dell alliance partner of year |
HOLD, conviction 3/5. Probability-weighted 12-month target $59 (-15%). This is a business worth owning at a price worth waiting for: accumulate on a pullback toward the mid-$40s to $50 (roughly the street mean and where forward P/E compresses to the low-20s), trim into strength above the high-$70s, and hard-stop the thesis if Q4 FY26 inventory does not normalise or AC fails to inflect by H1 FY27.