Battery Park, Manhattan

research@frothcity.com

← All reports
MU Long Semiconductors / Memory

The Bullish Case for Memory: Micron, SK Hynix, and SanDisk

A long, led by the multiple. These are profitable memory makers, not a bet on which AI lab wins. This note is dated November 22, 2026, a Sunday, so the prices are the November 20 closes. Sandisk figures are split-adjusted for the 8-for-1 split. The earnings are from filings and releases public by November 21.

Froth City is not receiving any compensation for mentioning or promoting any of the ETFs in this report, but we reserve the right to do so in the future. (Roundhill, our inbox is open.)

This note is opinion, not investment advice. Froth City Research and its affiliates may hold positions in securities discussed. Do your own diligence.

Executive summary

We are long the memory complex: Micron Technology (MU), SK hynix (000660.KS), and Sandisk (SNDK). The positive case is that memory is a toll on computing. AI servers, ordinary data centers, PCs, phones, cars, and networks all need DRAM, high-bandwidth memory, or NAND. You do not have to pick the winning model, the winning chip designer, or the winning cloud. You have to believe the bits get stored and moved.

The lead is valuation, and it has to be computed, not borrowed. On November 20, Micron closed at $594.51. Fiscal 2026 diluted GAAP earnings were $74.33 a share, so the trailing multiple is 8.0 times. Yahoo shows no Micron split in 2024 through 2026, and the 10-K earnings are on that same share basis. Sandisk closed the November 20 session at a split-adjusted $79.24. Fiscal 2026 diluted GAAP earnings were $73.76 before the 8-for-1 split, or $9.22 after it, so the completed year is 8.6 times. A trailing twelve months that includes the October quarter is 5.1 times on that same post-split basis. SK hynix closed at 1,113,000 won. Its October 28 release does not state earnings per share. Using about 729 million shares from its January 2026 disclosure, trailing net earnings are about 4.3 times, and that figure is flattered by a second-quarter gain the company itself reported as a 118% net margin.

The cyclical risk is the whole second half of this note. A 4 times multiple on earnings that get cut in half is an 8 times multiple. An 8 times multiple becomes 16 times. Memory has done that before. The bullish claim is that contracted supply, a broader demand base, and balance sheets that are net cash make this peak longer than the last ones. That is a claim. It is not a promise.

How we rank the long. Likelihood and impact are Froth City's judgment. Prices are November 20, 2026.
# Reason Likelihood Impact Timing
1 The multiples on reported earnings are low High Severe The November 20 prices
2 These are profitable businesses, already High Severe Fiscal 2026 and the September quarter
3 The balance sheets are net cash, and cash is coming back High High Year-end and third-quarter balance sheets
4 Customers are signing up for years of supply High High Agreements already signed for 2027 and 2028
5 The stocks have already fallen a long way High High From the June peaks to November 20
MU trailing P/E
8.0×
SNDK FY P/E, split-adjusted
8.6×
SK hynix implied TTM
4.3×
MU FY2026 revenue
$133.2B
SK hynix Q3 revenue
₩104.6T

Contents

  1. 1 The multiple is the thesis
  2. 2 Profits, not a story stock
  3. 3 Net cash, buybacks, and a dividend
  4. 4 Demand is wider than one AI lab
  5. 5 What the selloff actually was
  6. 6 Four ways to own the theme
  7. 7 Risks
  8. 8 Other considerations

01 The multiple is the thesis

We compute every multiple below. We do not use a data-vendor screenshot. Forward earnings in the usual sense are a consensus for the next year. None of these three companies published a full-year earnings-per-share consensus in the documents we used, and we will not invent one. What they did publish is the last year's earnings and, for Micron and Sandisk, a guide for the next quarter. We show both, and we label the annualized guide as our arithmetic, not as the street.

Table 1. Price, earnings per share, and the multiple. Micron has no 2026 split, so price and EPS are on the same basis. Sandisk price and EPS are split-adjusted for the 8-for-1 split: the company's reported per-share figures are divided by 8. SK hynix earnings per share are implied from reported net profit and about 729 million shares, not a reported EPS. See the notes under the table.
Name Price Price date Earnings used EPS P/E
Micron $594.51 Nov 20, 2026 close FY2026 diluted GAAP $74.33 8.0×
Micron $594.51 Nov 20, 2026 close FY2026 diluted non-GAAP $75.52 7.9×
Micron $594.51 Nov 20, 2026 close 4 × FQ1 2027 non-GAAP guide midpoint $152.60 3.9×
Sandisk $79.24 Nov 20, 2026, split-adjusted FY2026 diluted GAAP, divided by 8 $9.22 8.6×
Sandisk $79.24 Nov 20, 2026, split-adjusted Trailing four diluted quarters, divided by 8 $15.59 5.1×
Sandisk $79.24 Nov 20, 2026, split-adjusted 4 × FQ2 2027 non-GAAP guide midpoint, divided by 8 $28.00 2.8×
SK hynix ₩1,113,000 Nov 20, 2026 close Implied trailing net, four quarters, 729 million shares ₩259,810 4.3×
SK hynix ₩1,113,000 Nov 20, 2026 close Company forward EPS Not stated Not computed

Sources for the earnings. Micron's fiscal 2026, ended September 3, 2026, is the Form 10-K filed October 9, 2026, and the September 30 earnings release: revenue $133.188 billion, GAAP diluted EPS $74.33, non-GAAP diluted EPS $75.52. The first-quarter fiscal 2027 guide in that release is non-GAAP diluted EPS of $38.15 plus or minus $1.00. Four times the midpoint is $152.60. That is not a full-year forecast. It is what the multiple would be if that quarter repeated four times.1

Sandisk's fiscal 2026, ended July 3, 2026, is the Form 10-K filed August 17, 2026: diluted EPS $73.76 on revenue of $20.248 billion. The four-quarter sum uses that year's second, third, and fourth quarters ($5.15, $23.03, and $43.97) plus the first quarter of fiscal 2027 ($52.60), from the October 29, 2026 earnings release, which is $124.75. The second-quarter fiscal 2027 non-GAAP guide is $55.00 to $57.00 a share, midpoint $56. That release says every per-share figure in it is pre-split. The 8-for-1 split became effective after the close on November 19, and trading opened split-adjusted on November 20. Yahoo's chart, including the days before the split, is on that split-adjusted scale: a pre-split buyback of about $1,562 a share in the October quarter ($5.0 billion for 3.2 million shares) is about $195 after dividing by 8, and that sits inside the October range of the Yahoo series. The November 20 regular-market price of $79.24 is therefore a post-split price. We divide the pre-split earnings by 8: fiscal 2026 becomes $9.22 (8.6 times), the trailing four quarters become $15.59 (5.1 times), and four times the next-quarter guide midpoint becomes $28.00 (2.8 times). All Sandisk per-share figures below are split-adjusted.25

SK hynix, on October 28, 2026, reported third-quarter revenue of 104.6155 trillion won, operating profit of 83.3338 trillion won, and net profit of 39.7756 trillion won. It did not state earnings per share. On January 28, 2026, it said a cancellation of 15.3 million treasury shares was about 2.1% of shares outstanding, which is about 729 million shares. The third-quarter dividend is 14 trillion won at 20,000 won a share, which implies about 700 million shares received that dividend. The gap is the right size for treasury stock. We use 729 million, the count tied to a company percentage, not the 700 million dividend implication. Four quarters of reported net profit, the fourth quarter of 2025 through the third quarter of 2026, are 189.2901 trillion won, or about 259,810 won a share, and 4.3 times the November 20 close. On 700 million shares the same earnings would be 4.1 times. The second-quarter release states net profit of 93.9226 trillion won and a net margin of 118%, against operating profit of 60.5426 trillion won and an operating margin of 76%. Net profit above operating profit is the company's own comparison, not a unit error. It is the quarter in which the Kioxia gain sits. We do not have a company forward EPS.3

Multiples on reported or guided earnings MU, FY2026 GAAP 8.0× MU, annualized Q1 guide 3.9× SNDK, FY2026, split-adjusted 8.6× SK hynix, implied TTM 4.3×
Figure 1. Bars are scaled to 8.6 times. Green bars are trailing results, and the Sandisk bar is split-adjusted. The blue Micron bar is four times the company's own next-quarter guide, not a street forward P/E. We did not verify a forward P/E for the S&P 500, the SOX, SMH, or SOXX from a primary source as of November 20, 2026, so those indexes are not drawn.

A November snapshot that puts Micron at 4.0 times forward earnings is in the neighborhood of our 3.9 times on the annualized guide. It is not the trailing multiple, which is 8.0 times. A snapshot that puts Sandisk at 3.3 times forward earnings is close to our 2.8 times on the annualized next-quarter guide, once price and EPS are both split-adjusted. It is not the trailing multiple. The trailing four quarters are 5.1 times, and the completed fiscal year is 8.6 times. Treating the November 20 price as post-split and leaving the earnings pre-split was the mistake that produced multiples under 2 times. Those are withdrawn.

Table 2. What happens to the multiple if the earnings in Table 1 are cut in half and the November 20 price does not change. This is the cyclical risk, stated as arithmetic.
Case EPS P/E at the November 20 price
Micron FY2026 GAAP$74.338.0×
Micron FY2026 GAAP, cut in half$37.1716.0×
Micron annualized Q1 guide$152.603.9×
That guide, cut in half$76.307.8×
Sandisk FY2026, split-adjusted$9.228.6×
Sandisk FY2026, split-adjusted, cut in half$4.6117.2×
Sandisk trailing four quarters, split-adjusted$15.595.1×
Sandisk trailing four quarters, cut in half$7.8010.2×
Sandisk annualized Q2 guide, split-adjusted$28.002.8×
That guide, cut in half$14.005.7×
SK hynix implied trailing net₩259,8104.3×
SK hynix implied trailing net, cut in half₩129,9058.6×

Halving Micron's completed year takes an 8.0 times stock to a 16 times stock without the price moving. That is no longer a single-digit multiple. Halving Sandisk's split-adjusted trailing four quarters takes 5.1 times to 10.2 times. Halving the completed fiscal year takes 8.6 times to 17.2 times. The annualized next-quarter guide, cut in half, is still 5.7 times, but that guide is one hot quarter repeated, not a year the company has reported. Likelihood that the cycle can do this is high. Impact is severe. Timing is whenever prices for DRAM, HBM, or NAND turn.

02 Profits, not a story stock

Micron's fiscal 2026 revenue was $133.188 billion, against $37.378 billion in fiscal 2025. GAAP net income was $84.969 billion, against $8.539 billion. Non-GAAP net income was $86.758 billion. Operating cash flow was $89.68 billion, against $17.53 billion. Gross margin for the year was 80.7% on a GAAP basis. The fourth quarter alone was $54.229 billion of revenue and $37.701 billion of GAAP net income. These are not the figures of a speculative lab. They are the figures of a supplier that was capacity-constrained into a price spike.1

Micron fiscal 2025 and fiscal 2026 revenue and GAAP net income FY2025 revenue $37.4 FY2026 revenue $133.2 FY2025 GAAP net income $8.5 FY2026 GAAP net income $85.0
Figure 2. Billions of dollars. Bars are scaled to $133.188 billion of fiscal 2026 revenue. Source: Micron earnings release, September 30, 2026, and the Form 10-K filed October 9, 2026.

SK hynix's third quarter was the first in which quarterly revenue cleared 100 trillion won. Revenue was 104.6155 trillion won, up 32% from 79.3187 trillion won in the second quarter. Operating profit was 83.3338 trillion won, an 80% operating margin, up 38% from 60.5426 trillion won. Net profit was 39.7756 trillion won, a 38% net margin. The first quarter was already a record on the company's telling: revenue 52.5763 trillion won, operating profit 37.6103 trillion won, net profit 40.3459 trillion won. Full-year 2025, reported January 28, 2026, was revenue of 97.1467 trillion won and net profit of 42.9479 trillion won. One quarter of 2026 net profit, if you use the second quarter's 93.9 trillion won, is larger than the entire prior year, and that quarter is the one that includes investment gains. The operating line is the cleaner read, and it is still enormous.3

Sandisk's fiscal 2026 revenue was $20.248 billion, against $7.355 billion, up 175%. The 10-K says the increase was a 437% rise in Datacenter revenue, to $5.153 billion from $960 million, a 195% rise in Edge, and a 29% rise in Consumer. The first quarter of fiscal 2027, ended October 2, 2026, was revenue of $11.031 billion, up 378% from $2.308 billion, and GAAP diluted EPS of $52.60. Datacenter revenue in that quarter was $4.802 billion, against $269 million a year earlier. The company marks that year-ago comparison as not meaningful, which is fair: $269 million is a base, not a run rate. Consumer revenue in the same quarter fell 39%, to $397 million from $652 million. The profit is real. It is not evenly spread across every end market.2

03 Net cash, buybacks, and a dividend

Micron ended fiscal 2026 with $38.364 billion of cash and, on the company's release, $73.48 billion of cash, marketable investments, and restricted cash. Long-term debt was $2.792 billion and current debt was $491 million, about $3.28 billion together. Even on the smaller cash figure, net cash is about $35 billion. Capital expenditures, net, were $27.37 billion for the year. Adjusted free cash flow was $62.31 billion. The company bought back $650 million of stock and paid $610 million of common dividends. On September 30 the board declared a quarterly dividend of $0.15 a share. The return of cash is small next to the earnings. The balance sheet is not.1

SK hynix said cash and cash equivalents at the end of the third quarter were 114.5 trillion won, up 26.5 trillion won in the quarter, and borrowings were 15.8 trillion won, down 2.8 trillion won. Net cash was 98.7 trillion won. For 2026 it plans shareholder returns of about 100 trillion won: cash dividends of 30 trillion won or more, and buybacks and cancellations of 60 trillion won or more, including 40 trillion won already completed. The third-quarter dividend is 14 trillion won, or 20,000 won a share, half of it described as a special dividend from the Kioxia gain. A return that large is a bullish fact and a reminder that part of the earnings being capitalized are investment gains, not DRAM.3

Sandisk ended fiscal 2026 with long-term debt and current debt at zero in the Form 10-K, and with $4.762 billion of cash. At October 2, cash was $5.136 billion. In the October quarter it repurchased about 3.2 million shares, pre-split, for $5.0 billion. Over two fiscal quarters it says it repurchased 6.0 million shares, about 4% of shares outstanding. Capex discipline here is not "they stopped spending." Micron spent $27 billion. It is that operating cash flow covered the spend and left a net cash balance. Likelihood is high because these are balance-sheet facts. Impact is high in a downturn, because a net-cash supplier can wait. Timing is the latest reported quarter.

04 Demand is wider than one AI lab

Micron's 10-K describes an industry in which customers want committed long-term access to advanced memory, and in which supply growth is structurally constrained. In the third and fourth quarters of fiscal 2026 it entered strategic customer agreements, and it expects to enter more. They are take-or-pay. The customer gets supply and a view of price. Micron gets visibility. The 10-K also says generative models have driven demand for HBM and other advanced products, and that the long-term trajectory is unknown. We do not upgrade that sentence into "HBM is sold out." The company did not use those words in the passages we read. What it did say is contracted supply, not a hope.1

SK hynix said it has signed long-term agreements with 14 major customers, including global cloud providers, covering more than 50% of total production capacity for 2027 and 2028. The contracts are typically five years, and the price sits inside a range that still moves with the market. HBM4 shipments began in earnest in the third quarter, and the company expects HBM4 to pass HBM3E in both volume and revenue in the fourth quarter. It also said it expects the shortage to last through 2030, because leading-edge processes are hard and new fabs take a long time. That is management's forecast. It is also the mechanism that would make a low multiple wrong: if the volume is already spoken for, a spot-price crash has less to bite on.3

Sandisk's 10-K describes multi-year agreements with Datacenter and Edge customers that commit both sides to volume. The Datacenter end market was $5.153 billion in fiscal 2026, a quarter of company revenue, not the whole company. Edge was $12.160 billion. Consumer was $2.935 billion and, in the latest quarter, it shrank. Memory demand is not only AI. It is also not immune to a weak phone or PC year. The bullish point is the mix: the profit surge is coming from the end markets that are signing multi-year supply, and the companies do not need a single model lab to be the winner.

05 What the selloff actually was

The draft said that, as of this note, Micron had fallen roughly 32% from its early-July level, and that Sandisk had fallen by more. On the Yahoo series, that 32% is too small for early July. Micron has no split in that history. It closed at $1,032.28 on July 1 and at $594.51 on November 20, a decline of 42%. The June 25 close was $1,213.56, a decline of 51% to November 20. Sandisk's series is split-adjusted on both ends. The split-adjusted close was $254.03 on July 1 and $79.24 on November 20, a decline of 69%. The split-adjusted high was $294.30 on June 22, a decline of 73% to November 20. The June 25 high was $293.50, the same 73% once it is rounded. SK hynix has no split in the Yahoo history. It closed at 2,560,000 won on July 1 and at 1,113,000 won on November 20, a decline of 57%. Its June 22 high was 2,919,000 won, a decline of 62%.5

A lower price is not automatically an entry. It is an entry if the earnings in the denominator survive. The stocks have already priced a large cut relative to the June peaks. They have not priced a return to fiscal 2025 earnings. Micron at $594.51 on fiscal 2025 diluted EPS of $7.59 would be 78 times. The bull case is not "buy the 2025 multiple." It is "the market is paying 8 times a year that already happened, and less than 4 times if the next quarter is the new run rate." The bear case is that the next quarter is the peak print.

06 Four ways to own the theme

We checked the sponsor pages. Where the page did not give a figure in the HTML we retrieved, we say so.

Roundhill Memory ETF (DRAM). The fund exists. Roundhill's page describes it as exposure to companies producing HBM, NAND, and DRAM. The gross expense ratio on that page is 0.65% a year. Top exposures are shown as of September 30, 2026, with Micron, Samsung, and SK hynix named. The weights were not in the page HTML we retrieved, so we do not draw them.4

Kurv Memory Select ETF (KMEM). The fund exists. Inception on the page is June 30, 2026. The expense ratio shown next to that date is 0.65%. A 0.22% figure on the same page is the 30-day median bid-ask spread, not the fee. Look-through holdings as of September 30, 2026, were SK hynix 33.4%, Micron 26.5%, Samsung 19.2%, Sandisk 11.4%, CXMT 2.5%, Seagate 2.2%, Western Digital 1.7%, Kioxia 1.6%, Nanya 1.2%, Winbond 0.6%, and GigaDevice 0.5%. The page also lists an "Other" of negative 11.9%. We do not force that residual into the pie.4

KMEM positive look-through holdings SK hynix, 33.4% Micron, 26.5% Samsung, 19.2% Sandisk, 11.4% Other positive names, 10.3%
Figure 3. Kurv Memory Select ETF look-through weights as of September 30, 2026, for the holdings with a positive weight. The slices are those weights scaled to 100% of the positive book. The sponsor page also shows Other at negative 11.9%, which is not drawn. Source: kurvinvest.com/etf/kmem.

VanEck Semiconductor ETF (SMH) and iShares Semiconductor ETF (SOXX). Both are real, broader semiconductor funds. Memory is only a slice of them. The iShares page states an expense ratio of 0.33%, and as of November 19, 2026, it shows semiconductors at 83.74% of market value and semiconductor equipment at 16.21%. The top-holding weights were not in the page HTML we retrieved. VanEck's SMH page redirected before it returned a document, so we do not state an expense ratio or a holding weight for SMH. DRAM and KMEM are the direct expressions of this note. SMH and SOXX are a bet on the whole chip industry, which includes the customers and the competitors of the memory makers.4

07 Risks

Memory is cyclical. The earnings in Table 1 are what a shortage looks like: Micron gross margin of 87% in the August quarter, SK hynix operating margin of 80%, Sandisk gross margin of 85% in the October quarter. Those are not mid-cycle margins. If AI spending slows, or if the fabs being built for 2027 arrive into a softer tape, prices fall, earnings fall, and the forward multiple rises even if the stock price falls with them. Table 2 is that risk with the price held still. In a real downturn the price does not hold still.

The suppliers can overbuild. Micron's 10-K says that if HBM demand weakens and suppliers move capacity back to conventional DRAM, conventional DRAM can be oversupplied. SK hynix said next year's investment will rise as Yongin Phase 1 comes on, even while it says it will time that spend. A long-term agreement that covers half of 2027 and 2028 capacity is not a sale of all of it, and the price inside those contracts still moves with the market.

The three names are not the same cycle. Sandisk is NAND and SSDs. Its Consumer business just fell 39% year over year. Micron and SK hynix carry DRAM and HBM. A NAND glut and an HBM shortage can happen in the same quarter. Owning all three diversifies the product. It does not diversify the cycle out of existence.

SK hynix's trailing net earnings are flattered. The second-quarter release itself puts net profit above operating profit, with a net margin of 118% against an operating margin of 76%. The 4.3 times implied multiple uses that net profit and about 729 million shares from the January disclosure. If the Kioxia gain does not repeat, the trailing number was never the run rate. The October release still does not state a share count. The dividend implies about 700 million participating shares, which would make the same earnings 4.1 times. We would rather show that range than borrow a 3.1 times figure we cannot recompute.

Sandisk per-share figures in this note are split-adjusted. The November 20 price of $79.24 is post-split. The fiscal 2026 diluted EPS of $73.76 in the 10-K is pre-split, and we divide it by 8 before we divide the price by it. Dividing only one of those numbers produces a multiple that is off by a factor of eight. That is the mechanical risk in the first days after an 8-for-1 split, and it is why the table says "split-adjusted" on every Sandisk row.

08 Other considerations

These are real, and they are not the numbered long.

There is no street forward P/E in this note. We did not find a full-year consensus EPS for these three, or for the S&P 500 or the SOX, in a primary document public by November 22, 2026. The 3.9 times Micron figure and the 2.8 times Sandisk figure are four times a one-quarter guide, and the Sandisk guide has been split-adjusted. They will be wrong if the guide is a peak quarter. They are still the only forward arithmetic we can tie to a company document.

Capex is not small. Micron's net capital spending was $27.37 billion in fiscal 2026. That is covered by $89.68 billion of operating cash flow. It is also a promise of more bits later. Discipline here means the cash flow covered the spend. It does not mean the industry has stopped adding capacity.

The cash returns at Micron are modest. A $0.15 quarterly dividend and $650 million of buybacks against $85 billion of GAAP net income is a start, not a capital-return story. SK hynix's 100 trillion won plan is the capital-return story, and part of it is a special dividend from an investment gain.

Customer concentration cuts both ways. Take-or-pay agreements and five-year LTAs make demand visible. They also tie the supplier to a few buyers who will renegotiate the moment they can. Related-party and hyperscale concentration is support until it is bargaining power.

09 Notes

  1. Micron Technology, Inc., earnings release furnished as Exhibit 99.1 to the Form 8-K filed September 30, 2026, and Form 10-K for the year ended September 3, 2026, filed October 9, 2026. Revenue, GAAP and non-GAAP earnings, the first-quarter fiscal 2027 guide, cash and investments, debt, dividends, buybacks, capital spending, and the strategic customer agreements are from those documents. sec.gov. 10-K.
  2. Sandisk Corporation, Form 10-K for the year ended July 3, 2026, filed August 17, 2026, and the fiscal first-quarter 2027 earnings release furnished as Exhibit 99.1 to the Form 8-K filed October 29, 2026. Revenue by end market, the 437% Datacenter increase, quarterly diluted EPS, the buyback, and the second-quarter guide are from those documents. The 8-for-1 split is the Form 8-K filed November 19, 2026. 10-K. earnings release.
  3. SK hynix Inc., "SK hynix Announces 3Q26 Financial Results," October 28, 2026; "SK hynix Announces 2Q26 Financial Results"; "SK hynix Announces 1Q26 Financial Results"; and "SK hynix Announces FY2025 Financial Results," January 28, 2026, all on the company's newsroom. Quarterly revenue, operating profit, net profit, cash, borrowings, the dividend, the long-term agreements, and the shareholder-return plan are from those posts. news.skhynix.com.
  4. Fund pages retrieved November 21, 2026, and still the latest pages public on November 22: Roundhill Memory ETF, roundhillinvestments.com/etf/dram; Kurv Memory Select ETF, kurvinvest.com/etf/kmem; iShares Semiconductor ETF, ishares.com. The VanEck Semiconductor ETF page redirected before it returned a document, so SMH fees and holdings are not stated.
  5. Yahoo Finance chart closes, split-adjusted where a split is in the series. Micron $594.51 and SK hynix 1,113,000 won are the November 20, 2026, closes. Neither name has a split in that history. Sandisk $79.24 is the regular-market price at 21:00 UTC on November 20, 2026, on the post-split scale. The daily close field for that session was blank. The chart also records the 8-for-1 split. July 1 and June peak levels for Sandisk are from the same split-adjusted series. This note is dated Sunday, November 22, 2026, so November 20 is the last close.

This note is opinion, not investment advice. Froth City Research and its affiliates may hold positions in securities discussed. Do your own diligence.