Meet the Red-Hot Artificial Intelligence (AI) Infrastructure Stock That Has Crushed Oracle With 115% Gains in a Year. It Is Still Worth Buying Hand Over Fist.

DigitalOcean reported 2025 revenue of $901M (+15% YoY) and AI ARR rose 150% YoY to $120M; management now expects ~21% revenue growth in 2026 and ~30% in 2027 while adding 31MW of capacity. Shares have surged ~115% over the past year and trade at 8.4x sales; the article projects $1.78B revenue in three years implying a ~$14.2B market cap at 8x sales (nearly 2x current market cap). By contrast, Oracle shows a $553B RPO (+325% YoY) but investor concern over large hyperscaler contracts and OpenAI exposure has shifted sentiment toward DigitalOcean, which targets SMB/developer customers and expects to keep unlevered adj. FCF margins at 18–20% despite capex. Risks include near-term margin pressure from capacity builds and execution on forecasted stronger growth.

Analysis

The cloud AI market is bifurcating: hyperscalers will continue to command training workloads while a growing cohort of developer-focused clouds is capturing inference and lightweight model hosting. That shift creates a two-speed demand signal for hardware — sustained, high-margin demand at the top end (training) and an expanding, more price-sensitive segment for inference that favors lower-cost CPU/GPU nodes and predictable OPEX pricing. Expect margin expansion for providers that successfully monetize software and managed services on top of commodity compute, with a measurable inflection once software attach reaches a mid-teens percent of ARR.

Second-order supply effects matter. As small clouds scale their own capex, OEM procurement channels and the secondary market for accelerators will deepen, muting blunt GPU spot-price shocks but prolonging a two-tier market for new-generation accelerators versus legacy cards. This benefits vendors that sell into both tiers: high-end accelerator makers keep pricing power for training, while diversified silicon suppliers and system OEMs can win share on cost-sensitive inference racks.

Key risks are execution and rapid efficiency improvements in model serving that compress hardware intensity per inference. In the near-term (quarters), earnings/ARR beats will keep sentiment elevated; in the medium term (12–36 months), margin outcomes, utilization rates and the pace of model quantization/pruning will determine whether capacity additions are accretive or destructive to returns. Watch leading indicators — ARR mix shift to managed AI services, utilization rate per MW, and spot rental rates for accelerators — as triggers that validate or reverse the current momentum.

From a positioning perspective, this is not a one-way trade: buy-side allocations should separate exposure to (A) high-end accelerator demand (structurally bullish for companies with leading GPU ecosystems) and (B) low-end, developer-first clouds where software monetization and predictable pricing drive durable FCF. Hedging between those buckets — and using options to express convexity around hardware cycles and ARR read-throughs — will materially improve risk-adjusted returns.

Key Decisions for Investors

• Long DOCN equity (2–3% NAV), 12–36 month horizon. Rationale: asymmetric upside from software attach and developer-driven recurring revenue; set tactical stop-loss at -30% and reduce position if quarterly ARR growth falls below guidance or unlevered adj. FCF margin contracts by >300bps. Target: 80–120% upside if software mix and utilization metrics trend positively.
• Long NVDA Jan-2027 call spread (buy 18–24 month calls, sell higher strike calls to fund ~50% cost), size 1–2% NAV. Rationale: protected exposure to sustained high-end accelerator pricing and inventory cycles; expected payoff if training demand remains tight. Risk: 100% premium loss if AI capex disappoints; reward skew 2:1+ on modeled scenarios.
• Pairs trade: Long DOCN / Short MSFT (equal notional, 12–24 months) sized 1–2% NAV. Rationale: capture relative outperformance of developer-first, margin-accretive clouds versus hyperscaler exposure to capital-intensive training workloads. Hedge rules: unwind or rebalance if MSFT cloud gross margin expands >200bps quarter-over-quarter or DOCN misses acceleration thresholds.
• Speculative long INTC Jan-2027 calls (small allocation 0.5–1% NAV). Rationale: contrarian edge if cost-sensitive inference nodes migrate to x86/DPUs and Intel demonstrates share recovery; maximal asymmetry for limited spend. Cut if Intel fails to show sustained server design wins in two consecutive quarters.