Penguin Solutions
This is the archived full research. Read the story
Penguin Solutions builds memory modules for servers and designs, builds and runs AI computer clusters for smaller cloud companies and enterprises.
The scores
Growth3/4 ↑
Sales rose 26%, but memory doubled on price while computing fell 14%; the growth narrows to one driver.
Business quality2/4 →
Modules are built to standards from makers' chips and the makers sell modules too; memory profit followed the cycle.
Financial position3/4 ↓
Cash nearly covers the 0% notes due 2031, but cash from operations was minus $151.9M as inventory tripled.
Execution record3/4 ↑
Fiscal 2026 beat its guide by 20 points of growth, mostly memory prices, under a new chief executive and finance chief.
Per share capture2/4 →
Diluted shares rose 17.8% in a year as notes and preferred moved into the count, against $74.4M of buybacks.
Potential2/4 →
Growth rests on a memory cycle and neocloud orders nobody has valued; nothing beyond purchase orders is signed.
Risk profile3/4 →
Top ten customers took 66% of sales and supply comes from three memory makers, one buying a stake.
Valuation4/4 ↓
About 15.7 times operating profit, 62% of a three-stage value near $67.91 at an 8% return: cheap, past the 20% margin.
Evidence quality3/4 →
Segment profit and the cycle are filed or TrendForce's, but memory's share of the guide is an inference.
If it works 1.9× · if it fails you keep about 100%.
The Verdict
We rate Penguin Solutions Neutral. It builds memory modules and AI computer clusters; memory earned nearly all its recent profit because chip prices soared, while the cluster business lost its biggest customer and has not shown it can earn on new ones, so the business is mixed. The price is under two-thirds of our value, but that value treats memory sold at shortage prices as the normal year; on a normal memory price our gloomy case sits near the price, so the discount is no margin of safety. As a bet on memory-expansion cards it is defensive. Clusters that earn would make it Attractive.
The long caseAI inference needs more memory than servers hold; CXL cards add it, and Penguin already sells them.
How we weighed it
The price against our value
Cuts both ways
Price at 62% of our value
The price is under two-thirds of our value, but that value treats shortage prices for memory as normal; the bear case sits close to the price.
The business
Against it
Mixed
A mixed business: memory earns nearly all the profit at peak prices, and the cluster business is not yet earning on its new customers.
Growth against what the price assumes
Against it
The growth the value needs comes from memory at shortage prices, which history says do not last.
What could go wrong
Against it
Memory prices can ease, cash went into stock bought at peak prices, a chip maker takes a stake, and many bets run against the shares.
Against competitors
Cuts both ways
Memory makers' own prices ride the same shortage, so the comparison does not settle whether the profit lasts.
How its group is behaving
Against it
Memory names move together on the shortage; its profit is the group's cycle, not its own advantage.
The latest news and report
Cuts both ways
The latest report showed memory earning more than guided and clusters growing with new buyers, but the cash drain showed the cost of stocking memory at peak prices.
Where this departs from the fair value
The fair value takes the analysts' path, which runs on memory sold at shortage prices. Treated as normal, that profit flatters the value; on a normal memory price the bear case sits near the price, so the discount is no margin of safety.
What moved it from Attractive
The view is now a judgement of the whole thesis, not the price band alone: the value rests on shortage prices for memory, which the band could not weigh.
The evidence under this - 6 claims
The guide may assume a lower memory margin and a higher computing one; the company gives no segment profit guide.
Analysts' fiscal 2028 EPS of $5.26 is above fiscal 2027's, so they expect the guide to be a floor, not a peak (C35).
- Part of the fall is Penguin Edge, a loss-making line now closed, not the AI business (C11).
- Q4 sales grew 11% with non-hyperscale up 99%, and the Q4 segment split is not yet filed (C13).
The 10-K describes CXL cards sold mainly inside its own AMD-based servers, a small base.
The company
Servers need memory, and memory chips come from three makers: Samsung, SK hynix and Micron. Penguin buys those chips and builds them into modules designed and tested for a particular customer's servers and network gear, made in Malaysia.
Its second business builds clusters of AI computers for smaller cloud companies, enterprises and governments, then runs them for three to five years with its own software. Those buyers would otherwise order from Dell, Super Micro or another server maker.
The evidence under this - 6 claims
Services fell 11% in Q4 to $57M on lower hyperscale services (C48).
How it makes money
It is paid per module, per cluster and per year of service, on purchase orders rather than long contracts. In fiscal 2026 memory was about 53% of sales, clusters 32% and a lighting business, Cree LED, 14%. Services were about a tenth of sales in the latest quarter. The ten biggest customers bought 66% of everything in fiscal 2025, and one cloud giant alone 18%.
The evidence under this - 5 claims
Where the money comes from, stream by stream
| Stream | Share of sales (fiscal 2026) | How it earns it |
|---|---|---|
| Memory modules, flash and SSDs (Integrated Memory) | 53% | DRAM modules, CXL cards and flash products built from memory makers' chips for server, network and industrial makers. |
| AI clusters, software and services (Advanced Computing) | 32% | GPU clusters, ClusterWare, managed services and Stratus fault-tolerant servers; services were about 10% of Q4 sales. |
| LEDs (Optimized LED) | 14% | Cree LED chips and components made in China. |
Who runs it
Kash Shaikh became chief executive in early 2026, and Stephen Cumming, previously finance chief at Edgio and Cambium Networks, became finance chief in October after an interim spell. Directors and officers owned 2.9% of the shares in late 2025. Half of executives' share awards pay only if the stock beats a broad index of smaller companies over three years.
The evidence under this - 2 claims
Why it wins, and who could beat it
Why customers come back
Nothing structural keeps customers here. Modules are built to industry standards from chips the makers also sell as their own modules. Penguin's edge is being qualified on a customer's servers and getting chips when they are scarce, which TrendForce says module makers kept doing in 2026.
The record shows what that edge is worth: memory profit was $73.6M in fiscal 2023, $22.4M in 2024 and $62.2M in a single quarter of 2026. It follows memory prices.
The evidence under this - 5 claims
Backlog covering four quarters and a new supply arrangement suggest buyers value the allocation (C24).
The threat
Every part of the business has a bigger rival. The chip makers sell modules straight to server makers, and Samsung and SK hynix make the new memory-expansion cards too. In clusters, Super Micro, Dell, HPE and Lenovo compete for the same buyers, and the largest cloud companies design their own.
The evidence under this - 5 claims
The 10-K describes CXL cards sold mainly inside its own AMD-based servers, a small base.
The forecast is two years old and read through a newspaper, not Yole's report.
The rivals: MU, MU, SMCI, DELL, HPE, and 2 more that are private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| MU | The memory makers' own module lines: Samsung, SK hynix and MicronPriced against it: One segment only | Integrated Memory: DRAM and flash modules and SSDs for server, networking and industrial makers | On sale | One segment only |
| The memory makers' own CXL modules from Samsung and SK hynix, and Micron | CXL memory expansion and MemoryAI | On sale | ||
| Other module houses: Kingston, ADATA and NetlistPriced against it: Listed abroad | Integrated Memory: DRAM and flash modules and SSDs for server, networking and industrial makers | On sale | Listed abroad | |
| SMCI · DELL · HPE | AI server makers: Super Micro, Dell, Hewlett Packard Enterprise and LenovoPriced against it: One segment only | Advanced Computing: AI factory clusters, ClusterWare software, managed services and Stratus servers | On sale | One segment only |
| Contract manufacturers that build clusters to design: Foxconn, Quanta and WiwynnPriced against it: Listed abroad | Advanced Computing: AI factory clusters, ClusterWare software, managed services and Stratus servers | On sale | Listed abroad |
How big it could get
How big this could get
Nobody counts the specialty module market, so its size cannot be built. What TrendForce does measure is the memory market it sells into: server memory prices up about 270% in 2026, supply 1% to 2% short of demand with the gap wider in 2027, and the big cloud companies spending about $1.3 trillion in 2027. Much of that money goes to buyers Penguin is moving away from.
The evidence under this - 3 claims
TrendForce also flags cloud providers' negative free cash flow as a variable that could cut spending (S09).
Most of that spending is the hyperscalers' own, the customer type Penguin is moving away from (C11).
Who has signed
Nothing beyond purchase orders. Management names wins with smaller cloud companies, one holding $10B of contracted computing from an AI lab, but does not say what Penguin's share is worth. It says memory orders cover at least four quarters, though orders can be cancelled.
The evidence under this - 4 claims
The $3B and $10B are the customers' own contracts, not Penguin's; its share is not disclosed.
Services fell 11% in Q4 to $57M on lower hyperscale services (C48).
Sales are on purchase orders without long-term agreements (C22), so backlog can be cancelled or repriced.
The results
The numbers
Sales grew 26% in the year to Aug 2026, faster than the 17% in the year before.
Operating margin is 9.6%, up from 0.6% in the year to Aug 2023.
Running costs take 18¢ of each sales dollar, down from 23¢ in the year to Aug 2022.
Each dollar put into the business earns 15% a year after tax, up from 4.7% in the year to Aug 2022.
It used $163M of cash in the latest year, so the $647M it holds lasts about 4.0 years at that pace.
It owes $142M more than the cash it holds, and its operations bring in no cash to repay it.
The share count rose 8.0% in 4.0 years, and pay in shares costs 1.8% of sales.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-11-10. Fiscal years end in August.
| Aug 2022 | Aug 2023 | Aug 2024 | Aug 2025 | Aug 2026 | |
|---|---|---|---|---|---|
| Size and growth | |||||
| Sales | $1.40B | $1.44B | $1.17B | $1.37B | $1.73B |
| Sales growth | +32% | +3.3% | −19% | +17% | +26% |
| Profit | |||||
| Gross margin | 28% | 29% | 29% | 29% | 28% |
| Operating margin | 4.8% | 0.6% | 1.6% | 4.2% | 9.6% |
| Net margin | 4.8% | −13% | −4.5% | 1.9% | 10% |
| What growth costs, per sales dollar | |||||
| Research and development | 5.6% | 6.3% | 7.0% | 5.8% | 4.7% |
| Selling, general and admin | 15% | 18% | 20% | 17% | 13% |
| Pay in shares | 2.7% | 2.7% | 3.7% | 3.0% | 1.8% |
| Capital spending | 1.5% | 2.7% | 1.7% | 0.7% | 0.7% |
| Cash | |||||
| Cash from operations | $105M | $104M | $77.2M | $109M | −$152M |
| Free cash flow | $84.6M | $65.0M | $57.8M | $100M | −$163M |
| Balance sheet | |||||
| Cash and short-term investments | $313M | $391M | $389M | $454M | $647M |
| Debt | $584M | $790M | $657M | $462M | $789M |
| Net cash | −$271M | −$400M | −$268M | −$8.08M | −$142M |
| Returns and shares | |||||
| Return on capital | 4.7% | - | - | 3.6% | 15% |
| Shares (diluted) | 54.4M | 51.3M | 52.4M | 54.4M | 58.8M |
What growth costs
Costs barely moved while sales grew, so operating profit nearly tripled in fiscal 2026. But the extra profit came from higher memory prices, not from a cheaper way of working, and the company's own filing says memory earns less on each sale than the rest.
The evidence under this - 3 claims
Part of the leverage is the end of Edge's losses, not scale.
The last report (6 Oct)
Quarter ended August 2026, filed figures.
Q4 fiscal 2026 brought sales of $566.7M, up 68% on a year earlier. Memory did the work: up 158%, while clusters grew 11%. Reported earnings flattered the quarter with a one-off tax benefit.
Underneath, the company spent cash to grow: $163.1M went out in the quarter as stock and money owed by customers swelled. Management guided fiscal 2027 sales up about 40%, to $2.43B.
What it did to the thesisIt showed memory is earning more than anyone guided, and that the cluster business grew again with new buyers. It left open whether clusters earn a margin on them, and the cash drain showed the cost of buying memory at peak prices.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $1.00 against $0.77 expected.
The evidence under this - 5 claims
- The ranges are wide: plus or minus 10 points on sales and plus or minus $0.70 on EPS.
- The CEO said memory 'continues to be strong, stronger than we expected', so the memory guide may prove low (C24).
What management did with the money
It bought back $74.4M of shares in fiscal 2026 while borrowing $750M at no interest, repayable or convertible into shares in 2031, to retire older convertible bonds and fund the memory it was buying.
Insiders in 2026Q3: 4 purchases and 9 sales, from their Form 4 filings.
The evidence under this - 3 claims
The price, and where we disagree
In short. Cheap. The price is about 62% of what we think the business is worth, at an 8% yearly return.
The shares have tripled in a year while sales per share rose 7%: almost all of the gain is buyers paying more for each dollar of sales. They bottomed at $16.24 in March and peaked at $81.39 in July.
Two tests are close. The annual report, due late October, shows what the cluster business earned in the fourth quarter. The first-quarter results on 5 January show whether it earns on its new buyers while memory prices are still rising.
The evidence under this - 4 claims
TrendForce also flags cloud providers' negative free cash flow as a variable that could cut spending (S09).
- Part of the fall is Penguin Edge, a loss-making line now closed, not the AI business (C11).
- Q4 sales grew 11% with non-hyperscale up 99%, and the Q4 segment split is not yet filed (C13).
No margin is disclosed for the non-hyperscale business, and the segment's nine-month margin fell (C12).
How we value it
- Method
- Cash the business will produce, in three stages
- Required return
- 8%
- Long-run growth
- 4%
- Moat
- No moat
| Bear | Base | Bull | |
|---|---|---|---|
| Sales growth, years 1 to 3 | 20.8%, 6.1%, 11% | 41.7%, 12.2%, 22% | 52.1%, 15.2%, 27.5% |
| Normal operating margin | 9.1% | 11.4% | 12.6% |
| Return on new capital after year ten | 8% | 8% | 11% |
| Value a share | $35.85 | $67.91 | $118 |
- Price
- $41.83
- Priced on
- 6 Nov
- Price as a share of the base value
- 62%
- Starting growth the price pays for
- 12.1%
- Yearly return the price offers on the base case
- 11.2%
- Risk check: valued at 10%, the rate for a cyclical or indebted business
- $49.71
Where each figure comes from
| Figure | Source | Where exactly |
|---|---|---|
| Sales, the last full year | Filed report | Annual report (10-K) for the period to 28 Aug 2026, filed 20 Oct 2026; the full year. |
| Operating margin today | Filed report | Annual report (10-K) for the period to 28 Aug 2026, filed 20 Oct 2026; the full year. |
| Tax rate | Our method's rule | The normal rate where no rate is filed to read: 25% for a US company, 21% elsewhere. |
| Sales per dollar of new capital | Filed report | Annual report (10-K) for the period to 28 Aug 2026, filed 20 Oct 2026: the year's sales over its operating capital. |
| Shares | Filed report | Annual report (10-K) for the period to 28 Aug 2026, filed 20 Oct 2026; the diluted share count. |
| Cash and investments | Filed report | Annual report (10-K) for the period to 28 Aug 2026, filed 20 Oct 2026; the balance sheet on that date. |
| Debt | Filed report | Annual report (10-K) for the period to 28 Aug 2026, filed 20 Oct 2026; the balance sheet on that date. |
| Sales growth, years 1 to 3 | Analysts' forecasts | FMP's consensus for the years to 28 Aug 2027, 28 Aug 2028 and 28 Aug 2029: the average of 1 to 5 analysts' revenue estimates, 41.7%, 12.2%, 22%. FMP publishes how many analysts, not who they are. |
| Normal operating margin | Analysts' forecasts | FMP's consensus operating margin for the year to 28 Aug 2029, the average of one analyst's estimates, 13.2%, less stock pay of 1.8% of sales. FMP publishes how many analysts, not who they are. |
| Return on new capital after year ten | Our method's rule | Set by our reading of the moat: the 8% rate with none, 12% with an average one, 15% with a strong one; the bear case earns the rate. |
| The bear and bull cases | Our method's rule | Our method's spread: the bear case takes half the growth and a fifth off the margin, the bull a quarter more growth and a tenth more margin. |
| Price | Market price | FMP's closing price on 6 Nov 2026. |
What each case rests on
Bear caseGrowth runs at half the analysts' path, 21%, 6% and 11%, the margin settles a fifth lower at 9.1%, and new capital earns only the 8% rate: the downturn test.
Base caseRevenue grows 42%, 12% and 22% over three years, the analysts' path (C54), then fades toward 4%, closing 30% of the gap each year; the margin moves to 11.4%, the analysts' year-three figure after stock pay.
Bull caseGrowth runs a quarter above the analysts' path, 52%, 15% and 28%, the margin reaches 12.6%, and new capital earns 11%.
What moves the base value
| Lower | Higher | |
|---|---|---|
| Growth in years 1 to 3, a quarter lower or higher | $55.92 | $82.15 |
| The normal margin, 15% lower or higher | $56.15 | $79.68 |
| Capital per dollar of growth, a quarter more or less | $63.22 | $70.73 |
| Return on new capital after year ten, at the rate or one step up | $67.91 | $85.54 |
| The rate, a point higher or lower | $52.27 | $94.14 |
| Long-run growth a point lower | $66.36 | - |
Value a share as growth and margin change
| Normal margin ↓ · growth → | 50% | 75% | 100% | 125% | 150% |
|---|---|---|---|---|---|
| 9.1% | $35.95 | $43.37 | $52.22 | $62.72 | $75.08 |
| 10.3% | $40.91 | $49.64 | $60.07 | $72.43 | $87.01 |
| 11.4% | $45.88 | $55.92 | $67.91 | $82.15 | $98.95 |
| 12.5% | $50.84 | $62.19 | $75.75 | $91.86 | $111 |
| 13.7% | $55.80 | $68.46 | $83.60 | $102 | $123 |
Growth is shown as a share of the base case's. Green: the price is at or under 80% of that value. Red: at or over 120%.
How we read the multiple
Counting debt and cash, buyers pay about 13 times the operating profit guided for fiscal 2027.
To earn 6 cents a year after tax on each dollar paid, a common rule of thumb for what a share should earn, Penguin must:
- Earn about $327M of operating profit every year.
- That is exactly the profit guided for fiscal 2027, a year of record memory prices.
- About three-quarters of it is memory; if memory fell back to its 2025 profit, operating profit would be near $60M.
Against its own history. Much dearer than its own normal, because buyers now pay for a memory boom as if it lasts.
The evidence under this - 4 claims
- If the 63M guided shares do not include all of the preferred, the equity is understated by up to $0.4B.
- On the guide's non-GAAP operating income of about $405M the multiple is near 11.
Analysts' fiscal 2028 EPS of $5.26 is above fiscal 2027's, so they expect the guide to be a floor, not a peak (C35).
The guide may assume a lower memory margin and a higher computing one; the company gives no segment profit guide.
Memory profit has never fallen below $22.4M a year in the last three years, and a data-centre mix may hold it higher (C04, C24).
Priced against its rivals
None of the listed rivals above sells the same thing to the same buyers as a large part of what it does. A price comparison needs at least two, so the price is judged against its own history instead.
The bet: what it pays if it works
The bet is on memory-expansion cards that let AI servers hold more working memory than their slots allow. If they grow into a real line, they add a modest amount to a company valued mostly on everything else.
If they fail, shareholders keep the memory, cluster and lighting businesses, worth most of today's price on next year's guided profit, though less if memory prices turn.
| CXL memory expansion and MemoryAI operating profit in two to three years | $100M blocks of CXL and MemoryAI sales a year needed | At the stated multiple |
|---|---|---|
| $51.00M (the big win) | about 3 at 100% kept | 1.9× |
| $102.00M | about 6 at 100% kept | 2.3× |
To pay 2× (a Fair bet) it would need about $63.82M of CXL memory expansion and MemoryAI operating profit (about 4 $100M blocks of CXL and MemoryAI sales a year); to pay 5× (a Cheap bet), about $486.56M of CXL memory expansion and MemoryAI operating profit (about 29 $100M blocks of CXL and MemoryAI sales a year), at the same multiple.
Each row is CXL memory expansion and MemoryAI alone, plus the rest of the company (Memory modules, AI clusters and LEDs: everything outside CXL) counted at about $3.27B.
The steps it has to take
- Done CXL cards sold inside Penguin's own servers
- Done An AI inference provider expands its CXL deployment
- Not started A MemoryAI KV-cache product shipping to a named customer due 2027-08
- Not started CXL or MemoryAI revenue disclosed as its own figure due 2027-10
- Not started CXL and MemoryAI sales near $300M a year due 2029-08
The detail behind the label
The evidence under this - 6 claims
The forecast is two years old and read through a newspaper, not Yole's report.
The 10-K describes CXL cards sold mainly inside its own AMD-based servers, a small base.
Sales are on purchase orders without long-term agreements (C22), so backlog can be cancelled or repriced.
- The ranges are wide: plus or minus 10 points on sales and plus or minus $0.70 on EPS.
- The CEO said memory 'continues to be strong, stronger than we expected', so the memory guide may prove low (C24).
Where we disagree with the market
Penguin is an AI winner whose cloud wins and memory demand keep compounding, with sales reaching $2.75B in fiscal 2028.
Memory's profit is a shortage price, and the cluster business has not yet shown it earns on the buyers replacing its biggest customer.
Analysts carry the guide forward. The filed numbers show memory earning nearly all of the third quarter's segment profit, and the same business earned just $22.4M in all of fiscal 2024.
We would be wrong if clusters earn a tenth of their sales as profit for two quarters while sales to the new buyers keep growing.
What analysts say, and the evidence
The evidence under this - 6 claims
The guide may assume a lower memory margin and a higher computing one; the company gives no segment profit guide.
- Part of the fall is Penguin Edge, a loss-making line now closed, not the AI business (C11).
- Q4 sales grew 11% with non-hyperscale up 99%, and the Q4 segment split is not yet filed (C13).
What could go wrong, and how we would know
The risks
- The shortage lasts longer. TrendForce sees memory supply short into 2028; memory profit could keep beating the guide for years.
- Clusters start earning. Sales to new cloud buyers grew 99% in the latest quarter; if they earn like the old customer did, there is a second engine.
- Memory orders run deep. Management says memory orders cover at least four quarters, which could carry profit through fiscal 2027.
- A chip maker as owner. SK hynix agreed to buy a 10% stake, which could secure supply, or give a supplier a say over a customer.
- Many bets against it. 13.8% of the shares available to trade are sold short, so good news could move the price sharply.
The evidence under this - 6 claims
TrendForce also flags cloud providers' negative free cash flow as a variable that could cut spending (S09).
Sales are on purchase orders without long-term agreements (C22), so backlog can be cancelled or repriced.
- The ranges are wide: plus or minus 10 points on sales and plus or minus $0.70 on EPS.
- The CEO said memory 'continues to be strong, stronger than we expected', so the memory guide may prove low (C24).
No margin is disclosed for the non-hyperscale business, and the segment's nine-month margin fell (C12).
The sale needs regulatory approval and had not been reported as closed by 15 October 2026.
What we still do not know
- What the cluster business earned in the fourth quarter, and what sales to the new cloud buyers earn on their own.
- How much of next year's memory growth is higher prices and how much more modules.
- What Penguin's share of its cloud customers' large contracts is worth.
- Whether SK hynix's stake purchase has closed, and whether it comes with a supply deal.
The evidence under this - 4 claims
- Part of the fall is Penguin Edge, a loss-making line now closed, not the AI business (C11).
- Q4 sales grew 11% with non-hyperscale up 99%, and the Q4 segment split is not yet filed (C13).
Sales are on purchase orders without long-term agreements (C22), so backlog can be cancelled or repriced.
The $3B and $10B are the customers' own contracts, not Penguin's; its share is not disclosed.
The sale needs regulatory approval and had not been reported as closed by 15 October 2026.
What would change our view
- Neutral→AttractiveClusters earn, so profit stops resting on memory alone
It needs all of:
- ○The cluster business earns a tenth of its sales as profit for two quarters in a row, so profit no longer rests on memory alone.checked at the next results, due by 5 Jan
- ○Sales to new cloud and enterprise buyers keep growing 40% or more on a year earlier in those quarters.watched as it happens
- Neutral→UnattractiveThe price runs 20% above fair value
It needs:
- ○The share price rises to about $81.49, 20% above what we think the business is worth.about ≈$79.40 today, 90% above the last close
We would also be wrong if
- Memory profit falls from one quarter to the next while memory prices are still rising.
- Stock keeps rising while sales fall from one quarter to the next.
What is coming
- 13 December 2026 Window to force conversion of the preferred opens: Whether the preferred becomes 6.1M common shares, and whether SK hynix's purchase has closed.
Every dated event ahead
- 13 December 2026 Window to force conversion of the preferred opens: Whether the preferred becomes 6.1M common shares, and whether SK hynix's purchase has closed.
- 5 January 2027 Q1 fiscal 2027 results: Memory sales and segment profit against TrendForce's prices, computing's margin with neocloud deals, inventory and cash from operations.
- 30 June 2028 TrendForce's expected new DRAM output: Whether DRAM prices turn, which tests memory's profit and the inventory.
Since we wrote this
- 10 NovThesis reviewJudged Neutral on the whole picture, at 62% of our valueThe view is now a judgement of the whole thesis, the fair value its main evidence and no longer the decider.
- 9 NovThesis reviewAttractive: priced at 52% of a value near $80.27Re-valued in three stages at an 8% return, the method now drawn on every thesis: the price sits past the 20% margin of safety.
- 9 NovBet moveBet re-struck: Asymmetric to Safe, little upsideRe-priced at the 6 November close; the rating moved from Fair bet to Expensive bet.
- 6 NovBet moveBet re-struck: Safe, little upside to AsymmetricRe-priced at the 5 November close; the rating moved from Expensive bet to Fair bet.
- 15 OctThesis writtenUnattractive, medium convictionMemory modules earning shortage prices, priced as if the shortage were the normal year.
Also checked 3 times with nothing new to act on.
Sources 18
Sources - 18
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2025-10-21business and segments, products, CXL cards, suppliers, competition classes, customers and concentration, purchase orders, R&D and patents, segment profit for three years, the SKT preferred termsbacksC01C02C03C04C10C20C22C29C31C43C46
- S02primarycompany filing2026-10-06fiscal 2026 and Q4 income statement by segment, balance sheet, cash flows, GAAP to non-GAAP reconciliation, fiscal 2027 outlook, customer wins and supply arrangementsbacksC32C05C06C07C13C14C17C18C19C24C33C34C35C40C46C49
- S03primarytranscript2026-10-06Q4 fiscal 2026 earnings call transcript (FMP), held 2026-10-06data/cache/transcripts/PENG-2026Q4.jsonnon-hyperscale AI infrastructure growth and share, neocloud unit economics, memory pricing and backlog, CXL and MemoryAI, the product and services split, the notes exchange
Non hyperscale AI infrastructure net sales grew 99% year-over-year and represented 66% of advanced computing net sales.
backsC02C07C09C10C13C15C16C19C24C29C48 - S04primarycompany filing2026-07-07segment profit on the segment measure and on non-GAAP, the reasons for each segment's change, the hyperscaler transition, memory's lower gross margin, the convertible notes and the preferred stockbacksC08C09C11C33C43
- S05primarycompany filing2026-07-17the notes' size, coupon, maturity, conversion rate, capped call, redemption terms and use of proceedsbacksC19
- S06primarycompany filing2025-12-19board, the SKT designee and standstill, beneficial ownership of directors and officers, and pay design including relative-return PSUsbacksC44C45
- S07primarycompany filing2026-07-28Schedule 13D Amendment No. 1 by SK Telecom Co., Ltd. and Astra AI Infra LLC SK Telecom Co., Ltd.the sale of all 200,000 preferred shares to SHIFTIX1, an SK hynix subsidiary, for $380.3M, and the shares they convert intobacksC23C43
- S08primaryindustry data2026-09-30the fourth-quarter DRAM contract price forecast, server DRAM undersupply and allocation to module makers
DRAM suppliers have been fulfilling agreed volumes to PC OEMs and module makers every quarter throughout 2026.
backsC26 - S09primaryindustry data2026-07-30the 2026 DRAM sufficiency ratio, the wider 2027 gap and when new capacity arrives
TrendForce estimates the DRAM sufficiency ratio at approximately -1% to -2% in 2026.
backsC27 - S10primaryindustry data2026-08-25server DRAM contract price rises in 2H25 and 2026, and price ceilings in long-term agreements
Server DRAM-a key procurement category for CSPs-saw contract prices rise by a cumulative 64% in 2H25, with a further jump of approximately 270% expected in 2026.
backsC25 - S11primaryindustry data2026-08-03the top nine cloud providers' 2026 and 2027 capital spending
Looking to 2027, TrendForce estimates that the combined CapEx of the top nine CSPs will hit about $1.3 trillion, showing nearly 50% year-over-year growth.
backsC28 - S12secondarynews2024-07-18Yole's CXL market forecast and Samsung's view of when CXL grows
According to market tracker Yole Intelligence, the global CXL market is forecast to grow from $1.7 million in 2022 to $15 billion by 2028.
backsC30 - S13primarycompany filing2026-08-31Super Micro Computer Form 10-K for the fiscal year ended June 30, 2026 Super Micro Computer, Inc.who an AI server maker names as its competitorsbacksC21
- S14secondaryvendor data2026-10-15Research pack frozen beside this thesis, 2026-10-15content/PENG/research-pack.jsonthe 14 October close, the 252-day return and its decomposition, estimates, the rating split and its history, earnings history, insider rows and their base rate, and FINRA short interestbacksC34C35C36C37C44
- S15primarycompany filing2026-10-06Stephen Cumming's appointment, the interim CFO's dates, and his pay including relative-return PSUsbacksC44C45
- S16primarycompany filing2025-10-07fiscal 2025 results by segment and the fiscal 2026 outlook issued with thembacksC03C32
- S17primarycompany filing2026-11-09trailing revenue, operating profit, stock pay, tax, debt, cash and the diluted share countbacksC55
- S18secondaryvendor data2026-11-09the analysts' average revenue and operating profit for the next three fiscal years, and the closebacksC54C55