DigitalOcean Holdings, Inc. (DOCN) | The Buildout — AI Infrastructure
The Verdict
DigitalOcean buys or leases data center capacity, installs GPUs and cloud infrastructure, and sells an integrated inference-and-agent platform to AI-native companies. It sits above the power and silicon layers and below the applications, combining open-model serving, managed data services, agent tooling, and core cloud resources. Its role is to turn raw compute into production AI workloads rather than to rent bare-metal GPU capacity.
| Market Cap | — |
| Revenue (TTM) | $949M |
| Revenue Growth | +17.6% |
| EBITDA Margin (TTM) | 38.0% |
| Net Debt | $158M |
| Earnings Beats | 7 of 7 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- AI customer ARR reached $234M in Q2, up 212% YoY; inference services are over 70% of that and grew ~800% YoY.
- The Inference Engine launched in late April 2026 and had over 6,000 customers by the Q2 call, with token volume up 30x over 60 days.
- RPO jumped from $243M in Q1 to $894M in Q2, up more than 12x YoY, with a 3.7-year average life.
- Committed capacity nearly doubled from ~75 MW to ~155 MW, while demand pipeline is 3-4x capacity.
- Balance sheet de-risked: $888M equity raised, $500M Term Loan A repaid, $472M of 2030 notes equitized, pro forma net leverage 0.7x.
What We’re Watching
- 50%+ 2027 growth is tied to data center implementation timing; the CFO says there is still a fair bit of moving parts for next year.
- FY2027 adjusted FCF margin was guided to high teens, down from a prior implied 20%+ expectation.
- Gross margin fell from 61% in Q1 2025 to 56% in Q1 2026 as capacity costs arrive before revenue; management is refocusing on operating margin.
- Top 25 customers represented 20% of ARR in Q2; management described this as broad diversification.
The thesis is strengthening: management has twice raised FY2026 growth guidance, capacity is arriving ahead of schedule, and AI customer ARR and RPO are scaling quickly. The main reservations are not demand but execution, specifically capacity timing and a lower FY2027 adjusted FCF margin path. The open question is whether full-stack inference and agentic workloads can produce durable cash conversion as the mix shifts.
Earnings Beat
Q2 revenue was $281 million, up 29% year-over-year and above the high end of guidance. Incremental ARR reached a record $93 million, nearly triple the year-ago quarter. AI customer ARR reached $234 million, up 212% year-over-year, while adjusted EBITDA was $114 million, a 40% margin.
| Metric | Q1 FY2026 | Q4 FY2025 | Q1 FY2025 | YoY |
|---|---|---|---|---|
| Revenue | $258M | $242M | $211M | +22.4% |
| Gross margin | 56.1% | 58.7% | 61.4% | -530bps |
| EBITDA | $82M | $130M | $67M | +22.6% |
| EPS | $0.14 | $0.24 | $0.37 | −62.2% |
| AI customer ARR | $234M | $170M | n/a | +212% YoY |
| RPO | $894M | $243M | n/a | >12x YoY |
This flexibility enabled us to increase list prices on numerous GPU fleets recently by approximately 30%.— Paddy Srinivasan, CEO, August 4, 2026
Management tone: Management shifted from a Q1 platform-launch narrative to a Q2 message that strategy is becoming results and results are building momentum. The tone on the Q2 call was more bullish than Q1, with management twice raising FY2026 growth guidance and expressing increased confidence in 2027 growth. Management also reframed away from gross margin and net dollar retention toward operating margin and ARR per megawatt.
Management Guidance
For Q3 2026, management guided revenue of $304M-$307M at 32-34% YoY growth, adjusted EBITDA margin of 38-39%, and non-GAAP diluted EPS of $0.28-$0.30. Full-year 2026 guidance is now $1.17B-$1.18B in revenue at approximately 30.5% growth, Q4 exit growth of 35%+, adjusted EBITDA margin of about 39%, adjusted FCF margin of 11-13%, and non-GAAP diluted EPS of $1.35-$1.40. No formal 2027 guidance was given, but management said it is even more confident in 50%+ revenue growth for 2027.
Trajectory
Revenue growth is accelerating from 22% YoY in Q1 2026 to 29% in Q2 2026, with Q3 guided to 32-34% and Q4 exit guided to 35%+. The acceleration is demand-led: supply is the binding constraint, and the newest 2026 data center contributed less than $500,000 of Q1 revenue. Gross margin is compressing as capacity costs land ahead of revenue, while adjusted EBITDA margin held around 40%.
The Model
The model projects FY+1 revenue of $1,160 million and EBITDA of $348 million, a 30.0% EBITDA margin, then FY+2 revenue of $1,800 million and EBITDA of $583 million, a 32.4% margin. The FY+1 revenue projection sits below management's FY2026 revenue guide of $1.17B-$1.18B, while the FY+2 revenue projection is above management's directional 2027 view of revenue exceeding $1.7 billion, with capacity delivery as the central driver.
| Metric | FY2025 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $901M | $1.2B | $1.8B |
| YoY Growth | — | +28.7% | +55.2% |
| EBITDA | $345M | $348M | $583M |
| EBITDA Margin | 38.3% | 30.0% | 32.4% |
Projections are the median of 5 independent model runs. The model’s revenue sits 2.9% above analyst consensus.
For Q3 2026, management guided revenue of $304M-$307M at 32-34% YoY growth, adjusted EBITDA margin of 38-39%, and non-GAAP diluted EPS of $0.28-$0.30. Full-year 2026 guidance is now $1.17B-$1.18B in revenue at approximately 30.5% growth, Q4 exit growth of 35%+, adjusted EBITDA margin of about 39%, adjusted FCF margin of 11-13%, and non-GAAP diluted EPS of $1.35-$1.40. No formal 2027 guidance was given, but management said it is even more confident in 50%+ revenue growth for 2027.
What Could Go Right — and Wrong
- Capacity arrives ahead of schedule again, pulling 2027 revenue forward along the committed ~155 MW path.
- Formal 2027 guidance comes in above the directional 50%+ revenue growth view; the CFO signaled potential additional upside.
- Open-weight token share holds near 75% and day-0 launches repeat Kimi K3's 400+ net new customers in the first week.
- Inference Engine monetization expands beyond tokens through Model Synthesis, the inference router, and managed agents, lifting ARR per megawatt.
- Another 9-figure annual revenue commitment lands, adding to RPO visibility.
- Memphis, the 60 MW additions, or the 20 MW addition slip, compressing the 2027 revenue ramp.
- Open-weight model momentum reverses and token share falls back from close to 75%.
- Hyperscalers and other large cloud providers intensify competition for constrained GPU supply, pressuring capacity availability.
- Pricing normalizes faster than expected as supply catches up, reversing the ~30% list price increases.
- Customer concentration rises sharply as large RPO commitments ramp, raising single-account risk.
Looking Ahead
The next twelve months revolve around capacity delivery and converting early inference traction into higher-layer platform adoption. Memphis is expected in the second half of 2026, with 60 MW ramping through 2027 and 20 MW online late 2027 into 2028. Management has not yet provided formal 2027 guidance, currently directional at 50%+ revenue growth.
- Second-half 2026Memphis data center launch — The remaining 2026 facility at 15 MW; tests on-time delivery and first revenue.
- Later in 2026Formal 2027 guidance — Management will quantify 2027 revenue, EBITDA, and FCF outlook beyond 50%+ growth.
- 20262026 convertible notes maturity — Management intended to retire the $312M 2026 convertible notes at maturity; exact date not disclosed.
- 202760 MW capacity ramp — Four locations begin ramping revenue throughout 2027; tests capacity-led growth.
- Late 2027 into 2028Additional 20 MW online — Targeted to come online over the last part of 2027 into 2028.
Financials
Annual Summary
| Metric | FY2024 | FY2025 | TTM | YoY |
|---|---|---|---|---|
| Revenue | $781M | $901M | $949M | +15.5% |
| Gross Margin | 58.9% | 59.9% | 58.5% | +98bps |
| EBITDA | $221M | $345M | $988M | +56.2% |
| EBITDA Margin | 28.3% | 38.3% | 38.0% | +998bps |
| Net Income | $84M | $259M | $237M | +207.2% |
| Free Cash Flow | $99M | $8M | $261M | — |
| Net Cash | — | — | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)58.5%
- EBITDA Margin (TTM)38.0%
- Net Margin (TTM)25.0%
- ROIC11.8%
- FCF Conversion2.9%
- SBC / Revenue8.8%
The Company
DigitalOcean operates an AI-native cloud purpose-built for inference and agentic workloads. Its stack spans foundational GPU and CPU infrastructure, an Inference Engine for serverless and dedicated endpoints, data and learning services, a managed agents platform, and core cloud attach such as Droplets, Kubernetes, and managed databases. The business matters because AI workloads are shifting from training toward inference and agentic execution, and DigitalOcean sells that layer rather than raw GPU rental.
DigitalOcean is an operator, not a power producer or silicon designer: it leases data center space, installs GPUs and supporting infrastructure, and sells services. The 10-K discloses leased data centers in the New York, San Francisco, and Atlanta metropolitan areas, as well as Australia, Canada, Germany, India, the Netherlands, Singapore, and the United Kingdom, with 2026 openings planned in Memphis, Richmond, and Kansas City. Management has cited 20 global data centers. It uses NVIDIA Blackwell Ultra GPUs and relies on a limited number of suppliers for several components, including GPUs; power, colocation, and cooling supplier names in the intel file are inferred and unverified. It also finances a material portion of equipment.
Business Segments
Competitive Landscape
The 10-K names large cloud providers Amazon AWS, Microsoft Azure, Google GCP, IBM Cloud, Alibaba Cloud, and Oracle Cloud, alongside smaller cloud providers OVHcloud, Akamai/Linode, Hetzner, Vultr, and Contabo, AI/ML infrastructure providers CoreWeave and Lambda Labs, and managed hosts Kinsta and WP Engine. Management frames DigitalOcean as a full-stack AI-native inference cloud, not a GPU rental business, and distinguishes it from Neoclouds through inference and agentic focus plus an integrated open-source stack.
- Amazon AWSNamed in the 10-K as a large cloud competitor; not discussed in detail in the source.
- Microsoft AzureNamed in the 10-K as a large cloud competitor; neighbor evidence describes Azure as constrained through at least 2026.
- Google GCPNamed in the 10-K as a large cloud competitor; neighbor evidence says Google is compute constrained in the near term.
- CoreWeaveNamed in the 10-K as an AI/ML infrastructure competitor; management distinguishes DOCN as inference and agentic focused rather than training-first.
- Lambda LabsNamed in the 10-K as an AI/ML infrastructure competitor; no detailed discussion in the source.
Supply Chain
DigitalOcean sits between data center landlords and equipment suppliers on one side, and AI-native application builders on the other. It buys or leases capacity and sells inference, data, agent, and core cloud services. No neighbor transcript in the provided intel file named DigitalOcean directly.
More on DOCN: Earnings recap