DOCN · Analyze — Financebot

2026-08-12 · $134.07 · Mkt cap $15.67B · EV ~$15.8B · Technology / Software Infrastructure

First file ever on a position that has been held without one. kb.py find DOCN returned no matches — nothing in Knowledge/, no prior report in Output/. The watchlist line carried Trim $70+, an instruction that has been 91% below the market and was flagged as decoration in the 8/11 scan. Replacing it is the main deliverable here.

Verdict up front

TRIM — conviction 4.5 / 10.

The business is executing better than almost anything else in this portfolio: revenue +29% and accelerating, AI ARR +212% to 21% of the total, net dollar retention at a three-year high, guidance raised twice this year, and a Rule-of-40 score of 69. None of that is in dispute and the re-rating from $30 to $134 was earned, not imagined.

What is in dispute is the price and the accounting through which the price is being justified. At 13.4× EV/forward revenue, ~97× forward non-GAAP EPS and ~112× the company's own adjusted free cash flow, DigitalOcean is priced for the bull case being delivered in full. And the adjusted FCF metric that makes the multiple look survivable explicitly excludes equipment acquired under financing arrangements and finance leases — while gross property, plant and equipment grew $958M in twelve months against roughly $400M of operating cash flow. The capital intensity did not go away; it moved off the capex line.

Hold the stub, do not add, and put a working sell rule where a dead one was.

0. Data integrity — three traps, all firing

1. The vendor annual balance sheet is stale and materially misleading. fin.py reports the FY2025 (Dec-25) position: debt $1.60B, equity −$28.7M, debt/assets 87.2%. The prior watchlist line repeated it as "87% D/A leverage." That was true in December and is not true now. A ~$900M equity raise in Q1 2026 (additional paid-in capital $16.0M → $916.9M, shares 91.9M → 104.3M) retired ~$360M of converts and flipped the balance sheet.

	6/30/2025	12/31/2025	3/31/2026	6/30/2026
Cash	$387.7M	$254.5M	$741.4M	$767.0M
Total debt	$1,763.6M	$1,602.1M	$1,299.9M	$1,529.9M
— of which finance leases	$274.4M	$306.3M	$380.2M	$608.9M
Net debt	$1,101.4M	$1,041.3M	$178.4M	$154.0M
Stockholders' equity	−$175.2M	−$28.7M	+$887.4M	+$930.7M
Tangible book	−$633.2M	−$476.9M	+$439.0M	+$486.7M

Net debt is $154M, not $1.0B. Equity is +$931M, not negative. Any thesis built on the leverage is out of date. (This is the same failure mode as Knowledge/Playbook/pitfall-spinoff-carveout-balance-sheet-persists-in-vendor-feeds — an annual vendor snapshot surviving past a transformative capital action.)

2. Trailing EPS is inflated by a tax item, and the forward P/E sitting above the trailing P/E is the tell. PE(ttm) 60.94 against PE(fwd) 72.40 fires Knowledge/Playbook/pitfall-forward-pe-above-trailing-pe-flags-an-inflated-base. The cause: FY2025 operating income was $157.0M and interest expense $17.9M, so pretax income was roughly $139M — yet reported net income was $259.3M. The ~$120M gap is a tax benefit, almost certainly a valuation-allowance release (pitfall-tax-valuation-allowance-round-trip-breaks-eps). It also trips pattern-net-margin-above-operating-margin-is-a-tripwire: net margin 23% against an operating margin of 10%.

Normalised FY25 EPS ≈ $139M × 0.79 ÷ 105.3M = $1.04, not the reported $2.52. True trailing P/E is ~129×, not 60.9×.

3. Reported capex no longer measures capital intensity. Q2 2026 purchases of PP&E were only $41.6M — about 15% of revenue, against a FY2025 run-rate of $268.5M. That is not a spending slowdown. Gross PP&E went $1,290.6M (6/30/25) → $2,249.1M (6/30/26), +$958.5M, while finance-lease obligations rose $274.4M → $608.9M (+122%, and +60% in Q2 alone). CFO Steinfort, on the call: "we closely match our cash outflow with our revenue by financing equipment." Detail in §4 — this is the central finding of the report.

1. Fundamentals

$M	FY22	FY23	FY24	FY25
Revenue	576.3	692.9	780.6	901.4
Gross profit	364.4	397.5	465.9	539.6
Gross margin	63.2%	57.4%	59.7%	59.9%
Operating income	(25.7)	32.8	91.0	157.0
Net income	(27.8)	19.4	84.5	259.3 ⚠️
EPS (dil)	(0.24)	0.20	0.89	2.52 ⚠️
R&D	143.9	136.9	142.5	161.6
OCF	195.2	234.9	282.7	309.6
Capex	(120.2)	(124.8)	(186.5)	(268.5)
FCF	74.9	110.1	96.2	41.1
Buybacks	(600.0)	(488.5)	(59.8)	(82.1)
Diluted shares (M)	100.8	96.4	94.5	105.3

CAGRs (3yr): Revenue +16.1% · OCF +16.6% · FCF −18.2% · Diluted shares +1.5%/yr (rising)

That FCF line is the whole argument in four numbers: $110M → $96M → $41M while revenue grew 30%. Operating cash flow compounded at 16.6% and free cash flow fell 18% a year, because capex more than doubled. Then in 2026 reported capex collapsed to 15% of revenue — not because the spending stopped, but because it changed form.

Note also gross margin. 59.9% in FY25, ~55% in Q2 2026 — roughly 500bps of compression as GPU and inference mix grows. Per §2 of analysis_notes.md, margin compression is the standard moat-erosion signal. Here it is mix rather than price, which is less alarming — but it means every incremental dollar of the AI revenue everyone is celebrating arrives at a structurally lower margin than the legacy droplet business.

Buybacks are over as a story. $600M and $488M in 2022–23, then $60M and $82M. The company stopped retiring shares and started issuing them: diluted count is rising 1.5%/yr, and two 2026 equity issuances took shares outstanding from 91.9M to 116.9M — +27% in eight months.

Scorecard: revenue growth A · revenue acceleration A+ · gross margin trend C− · cash conversion D · balance sheet B+ (much improved, see §0) · capital allocation incomplete — pivoted from buybacks to a capacity land-grab mid-thesis · earnings quality D (tax-inflated base, plus an adjusted-FCF definition that excludes the main funding channel). Flagged gaps: FY21 absent from the feed (3yr CAGRs only, not the framework's 5–8); no 2026 capex guidance is given in dollars or as a % of revenue — management guides capacity in megawatts instead; no GPU utilisation figure has ever been disclosed.

2. Q2 2026 — the print that earned the re-rating

Reported 2026-08-04; call transcript 08-11.

	Q2 2026
Revenue	$281M, +29% YoY (roughly 2× the ~14% of a year ago)
ARR	$1,125M, +29%; record $93M incremental ARR, +191% YoY
AI customer ARR	$234M, +212% — ~21% of total ARR
Net dollar retention	102% — a three-year high
Gross margin	~55%
Adjusted EBITDA	$114M, 40% margin
GAAP net income / EPS	$35M / $0.29
Non-GAAP EPS	$0.45
Operating cash flow	$110M (39% of revenue)
Adjusted FCF	$60.6M, ~22% margin (TTM 17%)
Purchases of PP&E	$41.6M
RPO	$894M ($366M next 12 months) — up >10× YoY
Cash	$767M

Guidance, raised twice:

	Feb 2026	May 2026	Aug 2026
Revenue	$1.075–1.105B	$1.130–1.145B	$1.170–1.180B (+30–31%)
Adj. EBITDA margin	—	37–39%	38.5–39.5%
Adj. FCF margin	—	9–12%	11–13%
Non-GAAP EPS	—	—	$1.35–1.40
Diluted shares	—	—	122–123M

Q3 guided to $304–307M (+32–34%). Management reaffirmed ≥50% revenue growth for FY2027.

The AI business is real and it is quantified — credit where due. AI ARR $234M growing 212%, and critically 85% of it is inference services and attached core cloud, not bare-metal GPU rental. Inference revenue grew ~800% YoY; 6,000+ customers onto the inference engine within 90 days of its late-April launch; token volume up 30× in 60 days; open-weight models went from ~15% to ~75% of token volume. Named ecosystem references: OpenCode, Daytona, Vercel, OpenRouter. The first nine-figure annual commitment was signed. That is not slideware, and it is a better business than renting GPUs by the hour.

But look at what is happening to the customer base underneath it.

Cohort	Share of ARR	Change
$100k+ customers	35%	ARR +98%, count +9%
$500k+ customers	26%	ARR +160%, count +35%
$1M+ customers	23%	up from 9% a year ago, ARR +214%, count +73%
Top 25 customers	20% of ARR	—

DigitalOcean is converting from a 680,000-customer long-tail self-serve business into an AI-whale business, fast. Tiered NDR makes the split explicit: 102% at $100k+, 106% at $500k+, 115% at $1M+ with zero churn in that cohort — the top compounds while the tail leaks.

And at exactly that moment, CFO Steinfort said net dollar retention "will no longer be highlighted as a key financial metric." A company retires its headline retention metric in the quarter it prints a three-year high, while customer concentration roughly doubles. That is a disclosure retreat at the least convenient moment, and it removes the single best early-warning indicator an outside holder had.

3. The re-rating, reconstructed

Monthly closes: Sep-25 $34 → Dec $48 → Feb-26 $56 → Mar $86 → Apr $96 → May $156 → Jun $157 (intramonth high $187.50) → Jul $117.50 → Aug $134.07. Roughly +295% in twelve months, and −28% from the June peak.

Drivers in order: (a) the Feb Q4 print — $1B ARR, AI revenue +150%; (b) the March GPU Droplets / AI-native cloud launch, the biggest single leg at +53% in the month; (c) the May Q1 print, after which sell-side targets were re-based violently — Morgan Stanley $75→$175, Goldman $78→$179; (d) Russell 1000 inclusion on June 29, forced passive buying; (e) a July drawdown on the equity offering used to retire converts.

There was probably a squeeze component in the first leg. Short interest was 16.35% of float, 4.1 days to cover, as of 2025-09-15 — ample fuel for a February–May move. ⚠️ A more recent read of ~11.9% of float could not be tied to a settlement date; treat current short interest as unverified.

Worth stating plainly: index inclusion and a short base are not fundamentals. Two of the five legs of a 295% move were mechanical.

4. Where the capital actually goes — the central finding

DigitalOcean guides to an adjusted FCF margin of 11–13% and reports 22% in Q2. Its own definition is operating cash flow less purchases of PP&E, capitalised internal-use software and intangibles, excluding restructuring cash — and excluding equipment acquired under financing arrangements and finance leases.

That last exclusion is doing enormous work.

	6/30/2025	6/30/2026	Change
Gross PP&E	$1,290.6M	$2,249.1M	+$958.5M (+74%)
Net PP&E	$734.7M	$1,555.0M	+$820.3M (+112%)
Finance-lease obligations	$274.4M	$608.9M	+$334.5M (+122%)
Total assets	$1,719.8M	$3,123.1M	+82%

Against that, twelve-month operating cash flow is roughly $400M and reported capex over the same span was a fraction of the PP&E added.

The company added nearly $1B of productive assets in a year while generating ~$400M of operating cash flow, and reported a positive free cash flow margin throughout. The gap was bridged by finance leases, equipment financing, and a ~$900M equity raise.

This is the same economic event as Akamai's 40%-of-revenue capex (see the companion Output/Stocks/Infrastructure/AKAM/analyze-2026-08-11.md and Knowledge/Playbook/pattern-ai-build-inflates-earnings-while-destroying-fcf) — but Akamai puts it on the capex line where a screener can see it, and DigitalOcean routes it through leases where one cannot. Akamai's disclosure is uglier and more honest.

To management's credit, they do disclose a partial correction: "adjusted FCF less lease principal payments" was $154M / 16% of revenue on a TTM basis at Q1. That is the more honest figure, it is buried, and it still does not capture the full lease-financed asset additions.

The scale commitment is the other half. Committed capacity is now 155MW, +20MW added in Q2, with sites coming online through late-2027/2028. For a company guiding to $1.18B of revenue, 155MW is an enormous fixed forward obligation. Management says most new capacity is pre-allocated to specific customers or the token fleet before launch — which is reassuring if true and unverifiable, because no utilisation figure has ever been disclosed.

5. Moat

Quantitative base. ROIC = FY25 NOPAT ($157.0M × 0.79 = $124M) ÷ invested capital $1,267M = 9.8%. Against a beta of 1.59 and a WACC comfortably in the 11–12% range, DigitalOcean is earning at or below its cost of capital — and invested capital grew 46% year over year to $1,852M, so the denominator is running ahead of the numerator. Gross margin is compressing ~500bps on AI mix.

Moat source	Assessment
Distribution / SEO funnel	The real asset — a decade-deep tutorial library that ranks for a huge share of "how do I configure X on Ubuntu" queries. Near-100% product-led, self-serve. Hard to buy, hard to copy.
Switching costs	Weak at the base, real at the top. A bare droplet is an image and a DNS change. Managed Postgres + Spaces + inference tokens together create genuine data gravity.
Cost advantage	None. Negative. Hetzner's CX22 is ~€4.50/mo against DigitalOcean's ~$24 for the same spec, with 20TB of egress versus 2TB. That is a 3–5× gap.
Brand / simplicity	Real with individual developers; not a procurement criterion.
Efficient scale	None — this is a commodity capacity market.
Network effects	None.

The pricing direction is the tell, and it cuts both ways. DigitalOcean raised NVIDIA and AMD GPU Droplet prices effective 2026-08-01, with management describing ~30% list increases largely realised through renewals, after a ~20% list increase in 2022. Raising price into a market where Hetzner is four times cheaper is either pricing power in inference or an acknowledgement that the price-sensitive tail is already lost. On the current evidence — tail leaking, whales compounding — it is both.

Adversarial stress-test. 1. Hetzner wins by doing nothing. It is already 3–5× cheaper on raw compute. 2. Cloudflare bundles Workers/D1/R2 compute into an existing free CDN funnel and prices egress at zero. Compute becomes an add-on rather than a purchase decision. 3. AWS cuts Lightsail 30% and extends startup credits, starving the Builder cohort at the source. Hyperscaler credits capture the exact customer DigitalOcean needs to grow into, at zero cash cost to the founder. 4. The deepest one: the simplicity premium is a UI premium. DigitalOcean's moat is "easiest control panel, best tutorial." When an AI coding agent provisions infrastructure from a Terraform file, neither is a purchase criterion, and the ranking collapses to price and API quality — where DigitalOcean loses to Hetzner on the first and the hyperscalers on the second. The AI wave that is currently driving the revenue is the same wave that erodes the acquisition channel.

On the GPU/inference side specifically, DigitalOcean has no supply advantage (no NVIDIA preferred allocation, no anchor hyperscaler contract like CoreWeave's), no cost-of-capital advantage, and no proprietary silicon. Its answer — sensibly — is to not compete on raw GPU-hours at all, but to sell a serverless token endpoint plus the surrounding cloud to developers already in its funnel. That is the right strategy. It is also a software-and-distribution edge that Nebius (Token Factory), Together, and Akamai's 4,400-location AI Grid are actively replicating.

Evergreen rating: 4/10. A share-taker in a commoditising market, currently riding a genuine inference cycle. Five years: probably still here and larger. Ten years: only if the inference-plus-attach flywheel hardens into real data gravity before the funnel decays.

6. Sentiment

Analysts are bullish and were dragged there. Feb: Cantor $68→$83. Mar: Citizens $83→$105. Apr: BofA $103→$107, Oppenheimer $100→$115. May 6: Morgan Stanley $75→$175. May 7: Goldman $78→$179. Jun: KeyBanc initiates OW $200. Jul: Canaccord $200, Stifel upgrades to Buy $160, Baird initiates Outperform $165. Post-Q2: Citi $185→$190; Barclays $160→$161; UBS $155→$140 — the only cut. Consensus buy, mean target ~$172–177, ~30% above spot.

Note the shape: targets tripled in ninety days, chasing the price rather than leading it. A sell-side mean 30% above the tape after a 295% run is a momentum consensus, not an independent valuation.

Insiders: zero open-market purchases in twelve months, and heavy selling clustered at the top. CFO Steinfort sold 20,000 at $55.40 (Mar 3), then 25,000 at $152.50 (May 15) and 10,000 at $170.07 (Jun 2) — roughly $6.6M, with the large blocks at the peak. Director Jenson exercised 20,000 options at $19.47 and sold the same day at $147.62. Director Schneider 4,338 at $156.38. Director Adelman 4,200 at $124.01 on Aug 7, post-earnings. Officer Barrett 22,000 at $54.77. CEO Srinivasan: no sales found — only RSU grants. Separately, Access Industries sold 3.69M shares in May at ~$148–162, about $557M — a sponsor exit. ⚠️ Form 4 feeds do not tag 10b5-1 status; Steinfort's earlier even-sized tranches look programmatic, but the May/June blocks are larger and price-clustered. Verify the footnotes before inferring intent.

Management. CEO Paddy Srinivasan since 2024-02-12 (ex-GoTo) — the AI-native repositioning is entirely his, and it has worked. CFO Steinfort unchanged since 2022. Real churn at product: CPTO Bratin Saha out; Vinay Kumar, a founding member of Oracle Cloud Infrastructure, appointed CPTO 2026-01-20 — a hire that matches the pivot precisely.

Capital structure, current. $625M of 0.00% converts due 2030 at a $39.17 conversion price were issued Aug 2025. In July 2026 the company repurchased ~$472M of them — deep in the money at ~$117 — funded by a registered direct offering of ~12.5M shares at $117.54, so ~96% of the new shares offset shares the converts would have created and ~4% was the premium. ~$153M of principal remains (~3.9M shares). A further ~$312M of 2026 notes matures 2026-12-01, fully covered by $767M of cash. Pro-forma net leverage ~0.7×. A $100M buyback authorization runs to 2027-07-31 and is being used to mop up incremental shares. The convert overhang is largely cleared — this was competent liability management and deserves saying.

7. Valuation

Company type: high-growth infrastructure, no dividend, GAAP earnings distorted, capital intensity partly off-balance-sheet. DDM and DYT are N/A. Graham is nearly meaningless here and I will say why rather than quote it as a target. Weight EV/revenue against growth, EV/adjusted EBITDA, and a reverse-DCF.

Multiples at $134.07 (116.9M shares, EV ≈ $15.8B):

Metric	Value
EV / FY26 revenue ($1.175B)	13.4×
EV / FY26 adj. EBITDA ($458M)	34.5×
EV / FY26 adj. FCF ($141M mid)	~112× — a 0.9% FCF yield
P/E on FY26 non-GAAP EPS ($1.375)	~97×
P/E trailing, as reported	60.9×
P/E trailing, tax-normalised	~129×
P/B	15.1×
Rule of 40 (30% growth + 39% EBITDA margin)	69

Graham IV — √(22.5 × 2.20 × 8.86) = $20.95, and on the tax-normalised $1.04 EPS it is $14.40. The price is 6–9× either figure. Graham is the wrong instrument for a 30%-growing asset-light-turned-asset-heavy cloud, and quoting it as a target would be false precision. What it correctly establishes is that there is no asset or earnings floor anywhere within sight of this price — the entire value is future growth.

The Rule-of-40 score of 69 is genuinely exceptional and is the strongest number in the bull case. The caveat is that it uses adjusted EBITDA, which sits above the lease and depreciation lines through which this business now funds itself — for a company adding ~$1B of PP&E a year, EBITDA is a particularly forgiving denominator.

Reverse-DCF — what must happen to justify $134? At a $15.8B enterprise value and a 10% discount rate, you need roughly $1.2–1.6B of steady-state free cash flow. On a 20% mature-state FCF margin that implies $6–8B of revenue — five to seven times today's — with capital intensity normalising at the same time. Reaching $6B by 2033 requires ~26% compound growth for seven straight years in a commodity market against Hetzner, Cloudflare and three hyperscalers. Not impossible. Nowhere near assured.

Bogle expected return: 0% dividend + revenue growth 30% (guided ≥50% for FY27) − ~2%/yr dilution ± multiple change. The multiple is the whole game at 13.4× revenue.

Fair value range: $70 – $165, centre ~$105. - Bear $70 — FY27 growth decelerates to 20–25%, the AI cohort concentration reverses on one whale, EV/revenue compresses to 6× on ~$1.45B. - Base $105 — FY27 revenue ~$1.65B (+40%, short of the ≥50% guide), EV/revenue 7.5×; inference keeps compounding but gross margin keeps sliding and true capital intensity is eventually marked. - Bull $165 — the ≥50% FY27 guide is delivered at ~$1.76B, EV/revenue 11×, RPO converts, and the inference-plus-attach flywheel proves durable.

The range is deliberately wide, because a 30–50% grower with a weak moat genuinely has that distribution. At $134.07 the stock sits between base and bull — roughly 28% above the centre and inside the top quartile of the range. It is not absurd. It is fully priced for success.

8. Tensions and the debate round

Fundamentals vs. Fundamentals — the sharpest internal conflict in this file. The income statement shows revenue accelerating from 14% to 29%, adjusted EBITDA at a 40% margin, and a Rule-of-40 of 69. The balance sheet shows gross PP&E up $958M in twelve months and finance leases up 122%. Resolution: both are true, and the second one is the one nobody is looking at. DigitalOcean is a genuinely accelerating business that is buying that acceleration with leased capital, and reporting a positive FCF margin by defining the leased capital out of the metric. The right way to hold both facts is: the growth is real, the free cash flow is not yet.

Moat vs. Sentiment. Moat rates this 4/10 — no cost advantage, Hetzner 3–5× cheaper, the SEO funnel devalued by the same agentic wave driving the revenue. Sentiment counters that the sell-side mean is $172–177 and every house raised. Resolution: Moat wins. Targets that tripled in ninety days while chasing price are momentum, not analysis, and UBS's single cut is the only independent act in the set. Note too that the moat verdict and the revenue verdict are not in conflict — a weak-moat company can grow 30% for years in an expanding market. It just should not be capitalised at 13.4× revenue while doing it.

Valuation vs. the growth guide. Valuation says 112× adjusted FCF and a reverse-DCF requiring $6–8B of eventual revenue. The bull replies that FY27 is guided to ≥50% growth, RPO is up 10× to $894M, and the first nine-figure commitment is signed — at 50% growth the multiple halves in two years without the stock falling. Resolution: this is the honest crux and it does not resolve. If ≥50% lands, $134 will look cheap in hindsight. If it lands at 30%, the multiple compresses and the stock halves. The position sizing, not the analysis, is where that uncertainty belongs.

The one thing that tips it. Rising customer concentration ($1M+ cohort from 9% to 23% of ARR, top 25 at 20%) combined with management retiring net dollar retention as a headline metric in the same quarter is the pattern that should make a holder cautious. It is not evidence of anything wrong. It is the removal of the instrument that would show you if something went wrong.

9. Risks

Valuation is the primary risk. 13.4× forward revenue, ~112× adjusted FCF. A deceleration from 30% to 20% would likely halve the multiple.
True capital intensity is off the capex line. ~$958M of PP&E added in twelve months against ~$400M of OCF, funded by finance leases up 122%. Adjusted FCF excludes exactly this.
155MW of committed capacity against $1.18B of revenue — a large fixed forward obligation, with no disclosed utilisation figure to verify the pre-allocation claim.
Customer concentration rising fast — $1M+ cohort 9% → 23% of ARR; top 25 = 20%. One nine-figure customer now matters more than 500,000 droplets.
NDR retired as a headline metric at the moment concentration doubled — a disclosure retreat.
Gross margin compressing ~500bps as AI mix grows; the growth arrives at lower margin.
ROIC ~9.8% against an 11–12% WACC, with invested capital growing 46% YoY.
No cost advantage — Hetzner is 3–5× cheaper on equivalent compute.
The agentic-provisioning threat to the acquisition funnel — the same AI wave driving revenue erodes the tutorial/SEO moat.
Dilution, not retirement — shares 91.9M → 116.9M in eight months; buybacks effectively over.
Insider selling clustered at the highs with zero open-market buys, plus a $557M sponsor exit.
Two of the five legs of the +295% move were mechanical (Russell 1000 inclusion, a 16%-of-float short base). Passive flows reverse.

A portfolio-specific passage was removed from the public build.

11. Verdict

TRIM — conviction 4.5 / 10. Fair value $70–165, centre $105. Entry $75–90.

Replacing the dead sell rule — the actual deliverable. The old Trim $70+ has been inoperative for months. It is replaced with:

Trim $150 — deliberately in dollar form, and re-derive it at every /analyze. $150 ≈ 12× EV/NTM revenue (NTM ≈ $1.45B bridging the FY26 guide and the FY27 ≥50% target). A Nx fwd earnings multiple is void on this name and the reason is mechanical: the site derives the dollar from Yahoo's forward P/E, whose implied EPS of $1.85 sits ~34% above the company's own $1.35–1.40 non-GAAP guide, and the trailing EPS it would otherwise key off carries a ~$120M tax-valuation-allowance release. Both denominators are broken; revenue is the only clean one. As revenue compounds, 12× NTM revenue rises on its own — at the guided FY27 revenue this level moves to roughly $180, which is the point of keying it to a multiple rather than freezing a dollar.

The framework's first principle decides this. Is this a good business? Better than it was — genuinely accelerating, a real and quantified AI business that is 85% inference rather than GPU-landlord, competent liability management, and a CEO who called the pivot correctly. But the moat is 4/10, ROIC is below cost of capital, gross margin is compressing, and the cost position is structurally worse than a German competitor charging a quarter as much. Has the market priced it? At 13.4× forward revenue and ~112× its own adjusted free cash flow — a metric that excludes the finance leases through which nearly $1B of assets arrived last year — the market has priced the bull case in full and is being asked to pay for the base case twice.

Hold the stub rather than churning a $107 position, do not add, and let the $150 rule work. If FY27 comes in at ≥50% this call will look early, and the sell rule is set above spot precisely so that outcome is not foreclosed.

What would change the rating: FY27 ≥50% growth reaffirmed with an adjusted-FCF definition that includes lease-financed equipment → upgrade toward 6.0 · first disclosure of GPU utilisation, if healthy → upgrade · finance-lease growth slowing while revenue accelerates → upgrade · NDR restored as a disclosed metric → upgrade · any insider open-market purchase → meaningful upgrade · a whale customer non-renewal, or the $1M+ cohort's ARR share stalling → downgrade sharply · gross margin below 52% → downgrade.

Next check: Q3 2026 print, early November.