Investment Gems

How I found it, and why it’s timely

I’ll admit my starting bias up front, because it’s the whole reason I wrote this. I have a reflex to like anything with “AI factory” and “NVIDIA partner” stamped on it, and Penguin has both. The Q3 print on July 7 lit up my feed exactly that way: record quarter, guidance raised, AI now three-quarters of the business. My instinct was to nod along and move on.

Then I opened the segment footnote in the 10-Q, and the story fell apart in an interesting way. The company reports three segments, and only one of them is actually driving anything. That gap, between the “AI infrastructure inflection” narrative and what the segment-level P&L actually says, is the reason this is worth a look right now. The stock is up 19% on the print and re-rating on a story that the footnote only half supports.

What Penguin Solutions actually is

Penguin Solutions (formerly SGH, formerly SMART Global Holdings) is three businesses under one ticker:

• Integrated Memory (the SMART Modular brand): specialty memory and storage modules, high-density DIMMs, and increasingly a CXL-based memory product line aimed at AI inference (the MemoryAI KV Cache Server). This segment sells into servers, networking, telecom and enterprise storage, and it rides the DRAM pricing cycle.
• Advanced Computing (the Penguin Computing brand): full-stack AI and HPC clusters for sovereign AI, hyperscale, neocloud and enterprise. This is the part with the real “AI factory” story: the ClusterWareAI operating system, a new AI Factory Operations Agent, and an NVIDIA AI Factory Specialized Partner designation from June.
• Optimized LED (the Cree LED brand): application-optimized LEDs. Small, steady, not part of anyone’s thesis.

Read the marketing and you’d think Advanced Computing is the engine. Read the numbers and it’s the opposite.

The bull case, and where it actually comes from

Let me give the quarter its full due, because it deserves it. Total net sales of $478.7M, up 48% year over year and a record. Non-GAAP EPS of $0.84 versus $0.55 expected, up 79%. Adjusted EBITDA of $67.6M, up 51%. And the operating leverage was real: SG&A was basically flat in absolute dollars, $59.4M versus $59.7M a year ago, on 48% more revenue. That is a textbook incremental-margin quarter.

Management then raised the full-year FY26 outlook to 22% sales growth and $2.60 in non-GAAP EPS, and put out a preliminary FY27 framing of about 30% growth in both sales and EPS, which pencils to roughly $3.41. Analyst estimates for FY27 run $3.03 to $3.50, so management’s own number sits near the top of a range the Street already broadly agrees with. In other words, the earnings aren’t the disputed part of this story.

Here’s the thing though. Trace every one of those good numbers back to its source and you land in the same place: Integrated Memory.

Segment	Revenue YoY	Gross margin	Operating income	Op margin
Integrated Memory	+111% ($130M to $275M)	21.0% to 30.0%	$12.5M to $62.2M	9.6% to 22.6%
Advanced Computing	+3.8% (flat)	43.4% to 26.5%	$24.7M to $3.9M	18.6% to 2.8%
Optimized LED	+7.2%	29.3% to 32.6%	$1.3M to $4.1M	2.2% to 6.2%

Memory added $145M of the $154M in total revenue growth. Memory operating income more than quintupled and is now 89% of total segment operating income. And notice memory’s own gross margin went UP, 21% to 30%, which is what a pricing upcycle looks like when it drops onto a fixed cost base. That is the whole “operating leverage” story. It’s a memory cycle.

The part the headline skips

Now look at the Advanced Computing row again, because it’s the one nobody’s talking about. This is the segment that owns the AI narrative, the NVIDIA partnership, the ClusterWareAI software. Its revenue was flat. Its cost of goods rose 35%. So its gross margin collapsed from 43.4% to 26.5%, and its operating income fell from $24.7M to $3.9M. Operating margin went from a healthy 18.6% to a thin 2.8%.

Management’s explanation is that customers are moving from evaluation into production with larger but more competitively priced deals. Translated: right now Penguin is buying AI-cluster volume by giving away margin. That’s a normal thing for a systems integrator to do early in a land grab, and it might pay off later. But it means the glamorous half of the business is currently the low-margin, no-growth half, and the boring half (memory) is carrying the company on a cycle that has, historically, always turned.

That’s the crux, and it’s why I couldn’t just nod along. The consolidated non-GAAP gross margin actually fell 3.6 points to 28.1% this quarter, even with memory margins rising, because the mix shifted toward memory and the compute margin cratered at the same time.

The valuation

PENG is solidly profitable, so I’m using forward non-GAAP P/E. (I’m ignoring trailing GAAP earnings on purpose, they’re flattered by a one-time $30.9M gain on an equity-investment sale this year, so the trailing P/E of about 56x is noise.)

At $78.35 the stock trades around 24x forward earnings. To put that in context: the market values Micron, the reference memory cyclical, at about 7 to 10x forward even with its HBM capacity sold out, precisely because it refuses to capitalize peak memory earnings. It values Super Micro, the closest pure analog to Penguin’s AI-cluster assembly business, at about 10x, because box-building is thin-margin. Dell sits around 22x. Vertiv, the premium AI-infrastructure name with real differentiated margins and a huge backlog, gets about 46x. Penguin at 24x is already priced like Dell, a diversified premium-systems company, not like the memory cyclical that currently earns 89% of its profit.

Building three scenarios on FY27 earnings, over about a 1.25-year horizon:

Scenario	FY27 EPS basis	Multiple	Implied price	Total return	Annualized
Bull	$3.50 (Street high)	40x (Vertiv-like)	$140	+79%	about +59%/yr
Base	$3.27 (Street midpoint)	24x (today’s, held)	$78	about flat	about 0%/yr
Bear	$2.50 (memory rolls over, below Street)	12x (cyclical)	$30	-62%	about -54%/yr

The shape of the bet: notice the earnings aren’t really the argument. Analysts already have FY27 at $3.03 to $3.50, so the bull’s $3.50 isn’t a stretch on its own, and management’s $3.41 sits right in there. The stretch is entirely the multiple. The bull needs the market to pay a Vertiv-like 40x for earnings that are, this year, mostly a memory cycle. The bear needs one thing too, that memory cycle to roll over, which hits the earnings and the multiple at the same time. So the real question is which single thing is more likely: a durable premium re-rating on margins that are currently falling, or a memory cycle doing what memory cycles always eventually do. There’s a detail I keep coming back to: even at $3.50, the top of the analyst range, if the multiple just holds at today’s 24x you get about $84, or roughly 7% upside. Almost all of the potential return here is a multiple-expansion bet, not an earnings bet.

Risks (both directions)

Why this isn’t a Sell:

• The AI-infrastructure demand is real, the NVIDIA partner status is a real credential, and if the ClusterWareAI software attach lifts Advanced Computing’s margins, the “escape the box-builder trap” story becomes investable. The optionality genuinely exists.
• Memory pricing could stay elevated longer than the skeptics think. Cycles overshoot in both directions.

Why this isn’t a Buy:

• Concentration. 89% of segment operating income is one cyclical segment. There is no diversified profit base to cushion a memory rollover. This single fact breaks every scenario at once if it goes wrong.
• Margins are moving the wrong way for the premium-multiple story. The whole bull rests on Penguin earning an AI-infrastructure multiple, and that multiple is earned by margin, and the margin trend this quarter was down.
• Cash conversion cratered. Nine-month free cash flow was about +$3.9M versus +$173M a year ago, because accounts receivable and inventory ballooned by roughly $640M combined to fund the ramp. Growth is being financed off the balance sheet.
• Overhangs. A $202.7M convertible preferred sits in temporary equity as a dilution risk, and the CFO just left for The Trade Desk, with only an interim in the seat.

Conclusion

Penguin is executing at a high level, I don’t want to talk anyone out of that. But the market is paying a durable-AI-platform multiple for what is, this quarter, a specialty-memory company having a very good cycle with a thin-margin AI-systems business bolted on. At 24x forward the good news is in the price, and the risk/reward from here is skewed against the buyer: the upside needs a multiple the current margins don’t justify, and the downside is a memory cycle that has always, eventually, turned.

So I’m a Hold, with a re-entry around $55 to $60 (roughly a Dell multiple on consensus, the level where the memory-cycle risk is actually being discounted rather than ignored).

The one thing that would flip me to a buyer: gross margin stabilizing or expanding while the ClusterWareAI software and services mix rises. That would be evidence, in the filed numbers rather than the press release, that Advanced Computing is climbing out of the box-builder trap toward real platform economics. That’s a margin proof, not a revenue proof, and it’s the specific thing I’ll be watching in the next one or two prints. Until it shows up, the multiple is running ahead of the business.

Disclosure: I/we have no stock, option or similar derivative position in any of the companies mentioned, and no plans to initiate any such positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it. I have no business relationship with any company whose stock is mentioned in this article.