Built by Claude FableDOCN | How the Business Works
DigitalOcean rents cloud infrastructure on a usage basis. Customers pay for what they consume (compute hours, storage, bandwidth, GPU time, inference tokens), billed monthly, with a growing layer of multi-year committed contracts from its largest customers. There is one reporting segment; management splits the business into two engines in its disclosures.
- Core cloud (general purpose, ~79% of ARR). CPU Droplets (virtual machines), managed Kubernetes, App Platform, managed databases, block, object and network file storage, load balancers and VPC networking. Sold mostly self-serve through a product-led funnel of 4M+ active developers, plus a small direct sales and migrations team that moves workloads off the hyperscalers (76 migrations in Q2'25 alone).
- AI / inference cloud (~21% of ARR). The Gradient AI Agentic Cloud has three layers: infrastructure (NVIDIA and AMD GPU Droplets, bare metal), platform (serverless inference APIs, model hosting, evaluations) and agents. Inference services now make up over 70% of AI customer ARR and grew close to 800% YoY in Q2'26; 85% of AI ARR is inference plus core cloud, not bare metal.
The pitch is simplicity and price/performance versus AWS, Azure and GCP, with no lock-in and open source at every layer. That is also the weakness: the primitives are portable, so DOCN is a price taker (see Thematics).
| Metric | Q3'24 | Q4'24 | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 |
|---|---|---|---|---|---|---|---|---|
| Revenue ($M) | 198.5 | 204.9 | 210.7 | 218.7 | 229.6 | 242.4 | 257.9 | 281.2 |
| Revenue YoY | 12.1% | 13.3% | 14.1% | 13.6% | 15.7% | 18.3% | 22.4% | 28.6% |
| ARR ($M) | 798 | 820 | 843 | 875 | 920 | 970 | 1,032 | 1,125 |
| Incremental ARR ($M) | 24 | 26 | ~23 | 32 | 44 | 51 | 62 | 93 |
| AI customer ARR ($M) | n/d | n/d | n/d | n/d | n/d | 120 | 170 | 234 |
| AI customer ARR YoY | ~200% (AI/ML) | ~160% | >160% | n/d | >100% | 150% | 221% | 212% |
| $1M+ customer ARR ($M) | n/d | n/d | n/d | n/d | 110 | 133 | 183 | ~259 |
| Net dollar retention | 97% | 99% | 100% | 99% | 99% | n/d | n/d | n/d |
| GAAP gross margin | 60.2% | 57.1% | 61.4% | 59.9% | 59.6% | 58.7% | 56.1% | 55.0% |
| Adj. EBITDA margin | 44% | 42% | 41% | 41% | 43% | 41% | 41% | 40% |
ARR is quarterly revenue x4 from Q4'24 onward (Q3'24 is the old final-month x12 method). Q1'25 incremental ARR is calculated from the ARR change. Q2'26 $1M+ ARR is 23% of total ARR as disclosed. AI ARR growth before Q4'25 refers to the AI/ML product line, a narrower definition than "AI customer ARR." Management says NDR excludes AI and has become "a less useful measure"; after Q3'25 it disclosed NDR only by cohort (D&E 102%, $1M+ 115% in Q4'25). n/d = not disclosed in the calls.
| Cohort | What it is | Latest disclosure |
|---|---|---|
| Developers / learners | Self-serve, low spend, long tail | ~11% to 12% of revenue (2024 to 2025); lower NDR than D&E |
| Builders, Scalers, Scalers+ | Higher-spend digital natives (170K+ customers) | 88% to 89% of revenue (2024 to 2025) |
| Scalers+ ($100K+ ARR) | Largest digital native accounts | 22% of revenue, +37% YoY (Q4'24); ARR +98% YoY (Q2'26) |
| $500K+ ARR | Fast-scaling D&E and AI natives | ARR +160% YoY (Q2'26) |
| $1M+ ARR | AI natives and largest D&E (Cursor, Higgsfield, Ideogram, Fal) | ARR +214% YoY; 23% of ARR vs 9% a year ago; 115% NDR, zero churn (Q4'25) |
| D&E total | Digital native enterprises | $604M ARR, 62% of total, +30% YoY, 102% NDR (Q4'25) |
Two things are happening at once. The top cohorts, mostly AI-native companies running inference, are compounding at triple-digit rates and signing 8-figure committed contracts (RPO $243M in Q1'26, up ~1,700% YoY). The legacy core underneath is growing roughly 11% (PM review estimate: core ARR ~$891M vs ~$800M a year ago).
DigitalOcean leases data center space and buys or finances the equipment inside it. Growth is now gated by power capacity, so the model has shifted from a capital-light SaaS-like cloud to a capex-heavy infrastructure business.
| Capacity step | Size | Timing | What it drives |
|---|---|---|---|
| Atlanta AI data center | Largest single site | Live 2025 | GPU inference capacity |
| Incremental leases (Q3'25) | ~30 MW | Through 2026 | ~25% exit growth in 2026 |
| Committed for 2026 (Q4'25 framing) | 31 MW, 3 facilities | Online by end of 2026 | FY26 guide built only on this |
| Post equity raise (Q1'26) | +60 MW, 4 locations | Ramps through 2027 | Funded by the $888M raise |
| Q2'26 add | +20 MW | Late 2027 into 2028 | 2027 "50% plus" growth target |
| Total committed | ~155 MW | Majority by end of 2027 |
Management's argument is that software makes each megawatt worth more: inference, storage and core cloud attached to GPUs earn more ARR per MW than bare-metal neoclouds (roughly 70% of $100K+ AI customers also buy core cloud). The cost shows up on the Financials page: GAAP gross margin is down ~490 bps YoY, TTM free cash flow is about $15M, finance leases total $578M, and diluted shares are up 25.8% YoY. The FY26 adjusted FCF margin guide was reset to 11% to 13% (from 18% to 20%) to pull forward 2027 capacity costs.
| Arena | DOCN share | Main competitors | DOCN's edge |
|---|---|---|---|
| Global cloud infrastructure | ~0.2% | AWS ~28%, Azure ~20%, GCP ~15% | Simplicity, predictable pricing, developer brand |
| Developer / SMB alt-cloud | Niche leader | Akamai/Linode, Vultr, Hetzner, OVHcloud | Broadest integrated stack in the niche |
| Neocloud / GPU cloud | ~0.7% | CoreWeave (10x DOCN revenue), Nebius, Lambda, Crusoe | Full stack beyond bare metal |
| Managed inference | Small, fast-growing | Together, Fireworks, Baseten, hyperscaler APIs, CoreWeave | Price/performance (e.g. ~50% lower cost per token for character.ai) |
Bottom line: a good product in two fast-growing markets with no structural pricing power. The investable question is whether AI inference ARR per megawatt earns a return above the cost of the equity and lease capital funding it. See Overview for the full score.