DOCNDigitalOcean Holdings, Inc.

DOCN — DigitalOcean Holdings, Inc. — dossier 2026-09-23

Slot names-ai-capex-5, cloud research routine, unattended. Model-drafted, not human-verified. Stages closed here: S1, S2 (incl. capital allocation), S3 (gate-minimum, incl. earnings quality), S5, S7 valuation table, S8, S9. S0/S4/S6/S7 are carried forward from the ai-capex cluster files.

Every figure carries the document it was read from. Quotes were located by exact substring search in a text extraction of the filing at the stated URL, or in the transcript the ROIC.ai connector returned — never re-typed. DigitalOcean's EDGAR HTML renders figures with a space inside the parentheses and after the dollar sign ($ 625,000, ( 16,269 )); the spacing is the source's. Balance-sheet and income-statement figures are in thousands as filed.

Sources

Not read (gate-minimum, §3): the FY2024 10-K, the Q1 and Q3 10-Qs, the 2026 DEF 14A (2026-04-24), transcripts beyond the two most recent, and the March and July 424B5 prospectuses themselves — the July equitisation is taken from the Q2 press release's own description of it. ROIC.ai holds only two years of annual cash-flow and per-share data for DOCN, so the five-year capital-allocation window in §2 is three years wide, and it says so.

S1 — Source

watchlist (owner, Apple Stocks, 2026-09-20), coverage sec_domestic, cik 1582961, cluster ai_datacenter. Not a Tier 0 promotion. The 2026-09-10 8-K (a $725m equipment finance facility) would have been a promotion candidate and is read here anyway.

S2 — Kill test

The operating numbers (10Q-Q2-26, PR-Q4-25, in $000 as filed)

❗ The line that matters: revenue +29%, gross profit +18%, operating income −18%, diluted EPS −26%. DigitalOcean is the only cash-generative name in this cluster and it is buying its growth acceleration with gross margin and with shares. Gross margin fell 4.9 points year over year in Q2 2026; operating margin fell from 16.3% to 10.4%; the diluted share count rose 25.8%. Q2's net income also carries an $8,295 income tax benefit where the prior-year quarter carried a $5,421 expense — without it, net income falls rather than holding flat.

The balance sheet at 2026-06-30 (10Q-Q2-26, $000)

⚠️ The debt moved rather than shrank. The $500,000 Term Loan was drawn to the full $500,000 on 2026-02-03 and repaid in cash on 2026-03-27 out of the follow-on offering, with $2,700 of unamortised issuance cost written off. In its place, finance-lease and equipment-financing obligations went from $130,514 to $577,720 in six months — a 4.4× increase, and it is the GPU fleet. On 2026-09-10 a further $725,000 equipment finance facility was signed with MUFG Bank as agent, maturing 2030-09-10, with "an accordion option of up to $300 million, that the Company intends to exercise" (8K-2026-09-10 Ex-99.1). Read the balance sheet without the leases and the company looks under-levered; read it with them and it has replaced a term loan with secured equipment paper against depreciating assets.

⚠️ The 2030 converts are deep in the money. $625,000 of 0.00% notes at a $39.17 conversion price — 15,957 thousand shares — against a stock at $146.00. The 10-Q discloses their fair value at $2,546,106 against $625,000 of principal, and records that the conversion condition has been met since 2026-04-01. Post-quarter the company repurchased "approximately $472 million" of them, "funded by a concurrent registered direct offering, reducing leverage with minimal cash usage and minimal dilution, with issued shares offset by the retired notes and an intended repurchase of approximately 500,000 shares" (PR-Q2-26, Recent Developments). That is the right trade at a 4.1× fair-value-to-par and it is still equity issued at $146 to retire a $39.17 strike.

❗ Shares outstanding: 91,947,614 (2025-12-31) → 105,002,427 (2026-06-30) → 117,579,550 (2026-07-30). The last figure is not from a DigitalOcean filing directly but from the Access Industries 13D/A, which states it is "based on an aggregate of 117,579,550 shares of Common Stock issued and outstanding as of July 30, 2026, as disclosed in the Issuer's Form 10-Q filed with the Securities and Exchange Commission (the 'SEC') on August 4, 2026" (13D/A-2026-08-06). +27.9% in seven months.

Survives or dies: DigitalOcean is not a survival question. It earns GAAP operating income, it generated $156,889 of operating cash flow in H1 2026, it has $767,026 of cash, its only near maturity is the 2026 converts ($312,322, classified current), and it has just been handed $725m of committed equipment financing plus an intended $300m accordion. The question is not whether it lives; it is whether the AI revenue it is buying is worth what it costs per share. On the evidence of Q2 2026 — 29% revenue growth converting to −26% diluted EPS — the answer is not yet.

Capital allocation and returns

Five years is not available here: ROIC.ai holds two years of annual cash-flow data for DOCN and this pass read one 10-K. The window below is FY2024–H1 2026, and it says so rather than pretending otherwise.

\* ROIC.ai reports FY2024 fixed-asset purchases as 0 with capex folded into "other investing" (−186,523); the FY2024 number is therefore not reliable at this depth and is not used for a conclusion. Sources: ROIC.ai get_cash_flow DOCN annual; 10Q-Q2-26 cash-flow statement.

• The direction of travel reversed in 2026. DigitalOcean repurchased 2,356,547 shares for $82,124 in FY2025 — "In February 2024, our Board of Directors approved additional repurchase program of up to an aggregate of $140 million of our common stock, which we completed in July
• In August 2025, we adopted the 2025 Share Buyback Program which authorizes the repurchase of up to $100 million" (10K-2025). In H1 2026 it repurchased nothing and issued 11.9m shares for $887,888 net. FY2025's average repurchase price was ~$34.85 (2,356,547 shares for $82,124); the March 2026 follow-on sold shares at roughly $74.6 net (11.9m for $887,888). The company bought its own stock at $35 and sold it at $75 — in that order, which is the right order — but the net effect over eighteen months is a share count up 28% and a $100m buyback authorisation sitting unused while the equity is issued.
• Buybacks vs. SBC. FY2025's $82,124 of repurchases against $80,315 of stock-based compensation: the buyback almost exactly offset the dilution and did nothing more. In H1 2026, $55,231 of SBC with no offset at all.
• Return on invested capital is the one bright spot and it is deteriorating. FY2025 operating income of $157,000 on total assets of $1,837,705 is ~8.5% pre-tax on assets, and equity was negative ($28,690 deficit) because of the historic buybacks, so ROIC on a conventional denominator is not meaningful. The incremental version is what matters and it is poor: between H1 2025 and H1 2026, revenue rose $109,686 and operating income fell $7,319, on net PP&E that rose $460,238. Incremental operating return on the last $460m of fixed assets is negative.
• Goodwill impairments in the window: none ($350,651 carried, essentially flat).

Capital-allocation verdict: ⚠️ neutral, trending the wrong way. The equitisation of the 2030 converts at a 4.1× fair-value-to-par was a good trade and the Term Loan repayment was sensible. But a company that spent 2024–25 buying stock at ~$35 is now issuing it at ~$75–146 to fund an asset base whose first $460m of incremental PP&E has produced less operating income than before. That is not capital destruction; it is capital reallocation into a lower-return business, on purpose, and the S9 triggers below are written to catch it if the return does not arrive.

S3 — Filings, gate-minimum

Read: the FY2025 10-K's risk-factor section and Item 5 issuer-purchases discussion; the Q2 2026 10-Q in full (Notes 5, 6, 10, the debt and lease tables, the equity roll-forward); the Q4 2025, Q1 2026, Q2 2026 and 2026-07-07 press releases; the 2026-09-10 equipment-financing 8-K and its Ex-99.1; the 2025-11-26 Item 5.02 8-K; the Access Industries 13D/A; both 2026 transcripts, Q&A-first per §3.1.

The three biggest risks management itself discloses

From the FY2025 10-K's risk factors (10K-2025, Item 1A), in the company's own words:

1. Results are hard to forecast, and the growth rate is expected to decline. "We expect fluctuations in our financial results, making it difficult to project future results, and if we fail to meet the expectations of securities analysts or investors with respect to our results of operations, our stock price and the value of your investment could decline." The body adds: "Even if our revenue continues to increase, our revenue growth rate may decline in the future as a result of a variety of factors, including the maturation of our business."
2. The 2030 converts can force cash out or reclassify the balance sheet. "The conditional conversion feature of the 2030 Convertible Notes, if triggered, may adversely affect our financial condition and operating results… we could be required under applicable accounting rules to reclassify all or a portion of the outstanding principal of the 2030 Convertible Notes as a current rather than long-term liability, which would result in a material reduction of our net working capital." The condition was triggered, from 2026-04-01 (10Q-Q2-26 Note 6).
3. Nothing stops more debt. "We are not restricted under the terms of the indenture governing the 2030 Convertible Notes from incurring additional debt, securing existing or future debt, recapitalizing our debt, repurchasing our stock, pledging our assets, making investments, paying dividends, guaranteeing debt or taking a number of other actions." Read against the $725m MUFG facility and its intended $300m accordion, this is the operative risk factor of 2026.

No going-concern language, no restatement, no auditor change, no material weakness appears in the FY2025 10-K.

Earnings quality

1. The non-GAAP bridge, and two definitional changes inside two quarters.
2. Non-GAAP diluted EPS was redefined in Q1 2026 to exclude "the in-the-money portion of our 2030 Convertible Notes as they are covered by our capped call transactions, which are expected to mitigate the dilutive effect of our 2030 Convertible Notes" (PR-Q2-26, About Non-GAAP Financial Measures). The effect is direct: GAAP diluted EPS fell to $0.29 in Q2 2026 from $0.39; non-GAAP diluted EPS was $0.45. ⚠️ Flag 1.
3. A new non-GAAP measure, "adjusted operating income", was introduced in Q1 2026 — the quarter gross margin began compressing. Asked about it on the Q1 call, the CEO said "gross margin is one input, but operating margin is a better, more holistic view of what's going on in terms of the overall profitability because the revenue growth is so rapid and it's certainly at a lower gross margin, but it comes with tremendous operating expense leverage" (CALL-Q1-26). GAAP operating income then fell 18% year over year in Q2. ⚠️ Flag 2.
4. RPO and the customer-count metrics were redefined in Q4 2025 and prior periods recast: "Beginning in the fourth quarter of 2025, the RPO amount represents all contracts regardless of the duration of their original expected term. Prior periods have been recast" and "Beginning in the fourth quarter of 2025, we redefined our total customer count and our customer category naming and disaggregation" (PR-Q2-26, footnotes 2 and 3). The headline "RPO increased to $894 million, up 12x from a year ago" runs on the new definition. ⚠️ Flag 3 — three of the four growth metrics management leads with were re-cut within three quarters of the 2026–27 outlook being raised.
5. Stock-based compensation was $55,231 in H1 2026, 10.2% of revenue (FY2025: $80,315, 8.9%). Against H1 2026 operating cash flow of $156,889, SBC is 35% of OCF. Not extreme by software standards; rising, and no longer offset by buybacks.
6. Cash conversion. OCF/net income was 1.19× in FY2025 and 3.35× in FY2024. H1 2026: $156,889 of OCF on $51,208 of net income, 3.06× — healthy, and flattered by depreciation on a fast-growing fixed-asset base rather than by accruals. No flag.
7. Working capital. The H1 2026 cash-flow statement shows accounts receivable a $32,230 use of cash against $17,064 a year earlier, on revenue up 25.5% for the half — receivables growing faster than revenue, which is what selling to committed AI customers on contracts rather than credit cards looks like. Prepaid expenses and other current assets are a further $53,906 use, which includes a "$46,043 of pre-payment obligation held in a third-party escrow account" released to the company after quarter end (10Q-Q2-26 Note). Worth watching, not yet a flag.
8. Costs moved onto the balance sheet. ⚠️ Flag 4. Capitalised internal-use software capex rose from $3,412 to $11,785 between H1 2025 and H1 2026 — +245% against 25.5% revenue growth — while amortisation of internal-use and purchased software rose from $4,399 to $7,400. The stock of capitalised cost is being built faster than it is being expensed.
9. Sector equivalents: n/a.

EQ: 4 flags — non-GAAP EPS redefined in Q1 2026 to exclude in-the-money 2030 converts (GAAP diluted EPS $0.29 vs non-GAAP $0.45); a new "adjusted operating income" measure introduced the quarter gross margin started falling; RPO and customer-count definitions recast in Q4 2025 with prior periods restated; capitalised internal-use software +245% against +25.5% revenue.

Transcripts, read Q&A-first (§3.1)

Open analyst calls with a live queue — Goldman Sachs, William Blair, Citi, Bank of America, Morgan Stanley, Cantor, KeyBanc, UBS on the Q2 call. Follow-ups are permitted and used. This is a real Q&A and the contrast with NBIS's portal format is instructive.

Guidance changes (step 2) — the full FY2026 path:

Revenue, margin and EPS were raised twice. Free-cash-flow margin was cut by up to six points and the share count guide raised by 11 million. The unlevered adjusted free-cash-flow margin — the measure that strips out the financing cost of exactly the equipment leases now on the balance sheet — was guided in February and has not been guided since.

What analysts asked about (step 3): whether 2027 should now be above "50% or more" growth (Mohan, Ader); the pricing contribution to the record $93m of net new ARR (Ader, Blakey); revenue and ARR per megawatt as the AI mix rises (Singh); scaling the go-to-market for nine-figure customers (Borges, Zhang); AMD versus NVIDIA unit economics (Sultan); and, twice, free cash flow including lease payments (Ader, both quarters).

Dodges (step 4) — two, and they are the same question asked twice by the same analyst:

• Q1 2026 · free cash flow including lease payments · Jason Ader (William Blair). Asked: "for 2027, you talked about adjusted free cash flow margin in the mid- to high teens, I believe. Could you give us a sense of what it would be, including lease payments?" Matt Steinfort: "It's hard to answer, though, because it will depend entirely on the lease terms that we have… That's why we're not guiding to that at this point," then moved to balance-sheet flexibility and "the return on investment that we're underwriting for these new facilities." Locator: CALL-Q1-26.
• Q2 2026 · free cash flow including lease payments · Jason Ader again. Asked: "could you provide any specific guidance on the free cash flow for the year, including all the leases?" The answer went to net leverage first; Ader had to prompt — "And just on the free cash flow." — and got: "free cash flow, as we said, would be 11% to 13% for the year. That's on an adjusted free cash flow basis… And if you take all of the principal payments and everything, we'll still generate cash in 2026. So we expect to be free cash flow positive on any metric that you use." A direction, not a number, where a number was asked for. Locator: CALL-Q2-26.

The thread to pull: the same analyst asked the same question in two consecutive quarters and got no figure either time, in the same window in which the adjusted-FCF margin guide was cut from 15–17% to 11–13%, the unlevered adjusted-FCF margin guide was dropped entirely, finance-lease and equipment-financing obligations went from $130,514 to $577,720, and a further $725m facility was signed. All-in free cash flow is the number this business will be judged on and it is the one number management will not put a range around. Carried to the scorecard's ⚠️ line, with an S9 trigger.

A third exchange, logged but not scored as a dodge because the refusal was explicit rather than evasive: asked by Radi Sultan (UBS) for "the unit economics on a per megawatt basis for AMDs compared to NVIDIA GPUs", Paddy Srinivasan said "we certainly don't want to discuss the unit economics of different hardware throughputs" (CALL-Q2-26). Saying you will not answer is not the same as answering something else.

S5 — Ownership

13F register at 2026-06-30 (13F-Q2-3S): 386 filers, down from 456 — the filer count fell 15% in a quarter in which the stock rose. 64,083,701 shares reported.

⚠️ The active money sold and the dealers bought. BlackRock, T. Rowe, Ameriprise, Wellington, State Street and Bank of America all trimmed; Susquehanna, Jane Street, Citadel, Two Sigma, Walleye and D. E. Shaw all added. Hood River, a small-cap growth manager, is the one discretionary buyer of size. The pattern is the same one the converts create at NBIS — market-making and convert-arbitrage inventory replacing long-only holders.

⚠️ Latest-quarter 13F ingestion can lag (README warning, carried); the script reported no separate "not-yet-ingested" list for DOCN, so the filer-count fall of 70 should be read as partly ingestion and partly real until the next quarter settles it.

❗ Access Industries holds 15.86%. The Blavatnik group — Access Industries Management LLC, Access Industries Holdings LLC, AI Droplet Holdings LLC, AI Droplet Sharing LLC, AI Droplet Subsidiary LLC and Len Blavatnik personally — reported 18,653,451 shares, 15.86%, on a Schedule 13D/A with event date 2026-08-04 (13D/A-2026-08-06). Three 13D/As were filed in 2026 (2026-05-11, 2026-05-15, 2026-08-06). A 13D, not a 13G: this is a control-intent filer, it is the largest holder by a wide margin, and any decision about further equity issuance passes through it.

Insider / officer change. ⚠️ The Chief Product and Technology Officer resigned on short notice in November 2025: "On November 20, 2025, Bratin Saha notified DigitalOcean Holdings, Inc. (the 'Company') of his intent to resign as Chief Product and Technology Officer of the Company to pursue other opportunities. The parties mutually agreed Mr. Saha's departure would be effective November 26, 2025" (8K-2025-11-26). Six days from notice to effective. Two senior hires followed in 2026 — Kevin Van Gundy as CRO and "Leo" as CMO (CALL-Q2-26). Form 4 and Form 144 detail was not pulled for this name; gap.

Shelf / ATM / buyback. An S-3ASR was filed 2026-03-24 and has been used twice — 424B5s on 2026-03-24/26 (the 11.9m-share follow-on) and 2026-07-15/17 (the registered direct offering that funded the convert repurchase). The 2025 Share Buyback Program authorises up to $100,000 and expires 2027-07-31; nothing was repurchased under it in H1 2026. So: a live automatic shelf that has been used twice in six months, and an unused buyback authorisation. Short interest was not pulled for DOCN (Equibles budget spent on the other names); gap.

S7 — valuation table (FY2028), and its assumptions

At $146.00 (PX-2026-09-22) on 117,579,550 shares (2026-07-30), market value is ~$17.2bn. Against the FY2026 guide that is ~104× guided non-GAAP diluted EPS ($1.35–1.40) and, on H1's $0.45 of GAAP diluted EPS annualised, ~160× GAAP. ADV is ~$397m (2.72m shares × $146), thin relative to NBIS and worth stating.

Net debt is small and getting smaller: ~$153m of 2030 converts left after the ~$472m repurchase, $312m of 2026 converts, $578m of finance leases and equipment financing at 2026-06-30, against $767m of cash plus the July raise. Management's own framing: "just take the Q2 balance sheet and… simply subtract the amount of debt we retired in the equitization, puts us at 0.7x net leverage" (CALL-Q2-26, Steinfort) against a stated guideline of staying "well below that 4x net leverage".

Assumptions: FY2027 takes management's "exceed 50% growth in 2027" (PR-Q2-26, CEO quote) at the guide in the base and haircuts it in the bear; FY2028 grows 40% (base), 27% (bear) and 64% (bull). Margins start from the FY2026 guide of 38.5–39.5% adjusted EBITDA and hold (base), compress with gross margin (bear) or expand as the software attach rises (bull). Diluted shares of 135m extends the FY2026 guide of 122–123m by SBC and the remaining converts, and assumes no further equity issuance — which the last seven months argue against. Multiples are judgement.

What would have to be true for $146 to be right: FY2028 revenue near the bull case and a 20×+ multiple, or the base case plus a re-rating on the software attach rate. On the base case the model puts the shares at ~$107, 27% below the current price — this table does not support the current quote, and that is the finding rather than a rounding problem. The bear case is not a distress case: it is simply the current gross-margin trend continuing while the multiple normalises.

No buy price and no position size (owner decision 2026-09-23).

S8 — Management credibility

Eight rows, all with both locators. DigitalOcean is a domestic filer that guides quarterly and annually in an 8-K Item 2.02 exhibit, so the record is unusually checkable.

Two more, operational, both met: the Q1 commitment of "approximately 60 MW of incremental committed data center capacity that will come online throughout 2027" (PR-Q1-26) followed by a further "incremental 20 MW… bringing DigitalOcean's total committed data center capacity to approximately 155 MW" (PR-2026-07-07, restated in PR-Q2-26); and the second-half capacity schedule, confirmed on the Q2 call as "we've got 15 megawatts that are left… it's on track" (CALL-Q2-26, Steinfort). The CEO's claim of "capacity delivered on or ahead of schedule" (PR-Q2-26) is consistent with the filings this pass read.

Are the promisers still there? Srinivasan (CEO) and Steinfort (CFO) made and still hold every row; nothing here is departed. But the product organisation is not the one that set the 2026 plan: Bratin Saha resigned as Chief Product and Technology Officer effective 2025-11-26, six days after giving notice (8K-2025-11-26), three months before the FY2026 guide was issued, and a new CRO and CMO arrived during 2026. The 10-K names this as a risk in its own words: "our ability to effectively integrate and retain new members of our executive leadership team and senior management" (PR-Q2-26 forward-looking factors, item 11).

Dodge log rolled up: 2 dodges, both free cash flow including lease payments, both from the same analyst, in consecutive quarters. One explicit refusal (AMD vs NVIDIA unit economics), not scored.

Verdict: ✅ conservative — on everything they put in the headline. Six of the eight rows are beats or in-progress raises, and the quarterly beats are large (non-GAAP EPS roughly double the top of the Q2 guide). No revenue or profit guide has ever been cut. But all three misses and withdrawals are cash and share count: the adjusted-FCF margin guide was cut, the unlevered version was dropped, and the diluted share guide rose 10% — and it is the cash question, twice asked, that the CFO will not put a number on. The honest one-liner is conservative on the P&L, evasive on the cash flow statement, and S7's base case takes the revenue guide as given while taking nothing on trust about free cash flow.

S9 — Monitoring

• Next earnings: Q3 2026 results, expected early November 2026 — derived from the cadence (Q3 2025 on 2025-11-05, Q4 2025 on 2026-02-24, Q1 2026 on 2026-05-05, Q2 2026 on 2026-08-04). ROIC.ai lists no scheduled call. Derived, not sourced. Note the company pre-announced Q2 on 2026-07-07, a month early — a repeat of that would be the first signal of the quarter.
• Filing feed: the Tier 0 EDGAR sweep covers cik 1582961. A domestic filer, so 10-Q, 8-K Item 2.02 and Form 4 all arrive on the sweep with their substance visible.
• Leading indicator between quarters: the finance-lease and equipment-financing obligation balance ($577,720 at 2026-06-30, plus the $725m MUFG facility and its intended $300m accordion). It is disclosed every quarter, it is where the GPU spend lands, and it is what the adjusted free-cash-flow measure excludes. A second indicator, free and monthly: whether the 2030 converts' conversion condition stays met (the 10-Q discloses it), since that governs the working-capital reclassification risk management itself names.
• The trigger that converts this name to a buy is gross margin stabilising while revenue growth stays above 30% — one quarter in which gross margin is flat or up sequentially, operating income grows year over year, and the diluted share count is flat. That combination has not happened in 2026: Q2 delivered 29% revenue growth with gross margin down 4.9 points, operating income down 18% and diluted shares up 25.8%. It is the single quarter that would settle whether the AI mix is accretive per share or only in aggregate. Failing that, an all-in free-cash-flow guide — the number Jason Ader has now asked for twice — would do the same job.
• Kill criteria (specific, testable):
• Gross margin keeps falling. 59.9% (Q2 2025) → 55.0% (Q2 2026). Below ~50% while revenue growth decelerates means the AI revenue is bare-metal economics wearing a platform's name, and the premium multiple goes with it.
• Diluted EPS keeps shrinking while revenue accelerates. $0.39 → $0.29 in Q2. Two more quarters of that and the growth is being bought, not earned.
• The share count guide rises again. 111–112m → 118–119m → 122–123m. Another step, or an equity raise that is not paired with retiring the converts, breaks the "minimal dilution" framing the July equitisation was sold under.
• The adjusted-FCF margin guide is cut a second time, or the company stops guiding it as it stopped guiding the unlevered version. That would make three cash metrics withdrawn or reduced in a year.
• Access Industries files a 13D/A reporting a sale. A 15.86% holder with control intent reducing is a different fact from BlackRock trimming.