What is Datadog data-center strategy through 2027?

Direct Answer

Datadog runs zero data centers of its own — the entire platform sits on AWS, Azure, and GCP, and that asset-light posture is non-negotiable through 2027. The strategy is three-pronged: (1) deepen regional coverage on the big-three hyperscalers wherever a Fortune-500 logo lands, (2) stand up sovereign-cloud instances in EU, UK, KSA, India, and AU to clear data-residency procurement gates, and (3) lean on named GPU partnerships — Anthropic, OpenAI, and AWS Trainium — for Bits AI and LLM Observability inference rather than buying H100s. Every region decision is a cost-vs-compliance calculation: each new region adds 50-150bps of gross-margin drag from duplicated control-plane and idle storage capacity. Management has been explicit on earnings calls — they would rather slow regional expansion than build a single rack.

The Footprint Today (May 2026)

AWS (primary, ~70% of workload)

• US1 (us-east-1, N. Virginia) — flagship, largest tenant base
• US3 (us-west-2, Oregon) — DR + west-coast latency tier
• US5 (us-east-2, Ohio) — overflow + FedRAMP boundary
• EU1 (eu-west-1, Ireland) — primary EU commercial region
• AP1 (ap-northeast-1, Tokyo) — Japan + APAC commercial
• AP2 (ap-southeast-2, Sydney) — Australia data-residency tier

Azure (~20% of workload, growing)

• US1-FED (Azure Gov) — FedRAMP High candidate region
• EU2 (West Europe, Netherlands) — Microsoft-shop EU customers
• AP1-AZ (Australia East) — co-located with AU sovereign tier

GCP (~10%, strategic)

• US1-GCP (us-central1, Iowa) — Google-shop logos + BigQuery integration tier
• EU1-GCP (europe-west3, Frankfurt) — German data-residency option

Sovereign / regulated

• US1-FED (FedRAMP Moderate, FedRAMP High in audit)
• EU1 + DE option for GDPR + Schrems II
• KSA (Riyadh, AWS Middle East) — Vision 2030 procurement gate
• India (ap-south-1, Mumbai) — DPDP Act compliance, live Q2 FY26
• AU (Sydney) — IRAP Protected target FY27

What Drives Expansion 2026-28

• Logo-pull from Fortune-500 sales cycles — a single eight-figure ACV deal that needs in-country residency justifies a region
• Regulatory deadlines — EU Data Act (Sept 2025), DPDP India (live), Saudi PDPL, UK Data Protection and Digital Information Bill
• FedRAMP High milestone — unlocks DoD IL4/IL5 pipeline, est. $200M+ TAM
• Hyperscaler region launches — Datadog typically lights up 6-12 months after AWS/Azure opens a new region in a target geo
• AI/LLM Obs co-location — Bits AI inference latency demands regional GPU-adjacent presence
• Competitive pressure — Dynatrace and New Relic regional parity in EMEA forces Datadog to match or lose RFPs

The Sovereign Cloud Strategy

• EU sovereign tier — separate control plane, EU-citizen-only operators, no US-employee admin access (Schrems II hardening)
• UK — leveraging AWS London, separate billing entity for UK Crown procurement
• KSA — partnered with AWS Middle East (Bahrain + Riyadh local zones), sovereign operator agreements with STC/Mobily fronting
• India — DPDP Act forces in-country storage; ap-south-1 region went GA late FY26
• Australia — IRAP Protected accreditation in flight, target FY27 close, unlocks federal + state government
• Pattern: never build sovereign infrastructure standalone — always ride the hyperscaler's existing sovereign offering (AWS GovCloud-equivalent, Azure Sovereign Landing Zones)

The GPU + AI Inference Strategy

• Anthropic partnership — Bits AI assistant calls Claude via Anthropic's hosted inference, no local GPU needed
• OpenAI partnership — secondary LLM Obs traces + redaction pipeline run on Azure OpenAI for tenants who require it
• AWS Trainium / Inferentia adoption — moving cost-sensitive inference (anomaly detection, log clustering) to Trainium2 to compress unit economics by 40-60% vs H100
• No GPU CapEx — Datadog has explicitly told the street they will not buy GPU clusters; partner inference is the model through 2027
• Co-location latency — inference endpoints deployed in same hyperscaler region as customer's Datadog tenant to avoid cross-region egress charges

The Cost Discipline

• Each new region = 50-150bps gross-margin drag in year-1 from duplicated control plane + idle storage headroom
• Reserved instance / Savings Plans coverage — Datadog discloses ~80% RI coverage on AWS, the lever they pull when GMs compress
• Egress fees are the silent killer — multi-region replication and cross-cloud telemetry routing eat ~3-5% of COGS, actively renegotiated in master agreements
• Storage tiering — hot 15 days on NVMe, warm 90 days on S3 Standard, cold archive on S3 Glacier Deep — the tiering algo is a quiet margin lever
• Customer-attributable allocation — every region is P&L-tracked; regions that don't hit ACV thresholds within 18 months get capacity-frozen

What's NOT In The Strategy

• No owned data centers — zero plans to build, lease colo cages, or operate a private cloud
• No aggressive China expansion — geopolitical + data-export risk too high; China remains served via overseas regions for MNC subsidiaries only
• No GPU cluster purchases — partner inference instead of buying H100/B200 capacity
• No multi-cloud abstraction layer for customers — Datadog stays cloud-agnostic in its own backend but does not sell a 'cloud-arbitrage' SKU
• No edge / CDN-tier presence — last-mile telemetry collection stays on customer agents, not Datadog edge nodes

Region x Cloud Matrix

Strategy Flow

Bottom Line

Datadog's data-center strategy through 2027 is to never have one. They ride AWS, Azure, and GCP regional buildout, layer sovereign control planes only where procurement forces it, and partner with Anthropic/OpenAI/Trainium for the GPU bill they refuse to pay themselves. The discipline is brutal — every region is P&L-tracked and capacity-frozen if ACV doesn't materialize in 18 months. Expect 4-6 new regions through FY27, all hyperscaler-hosted, with FedRAMP High and IRAP Protected as the headline accreditations. Cross-link [q1674 — Datadog AWS dependency risk](/lab/q1674.html), [q1686 — Datadog FedRAMP roadmap](/lab/q1686.html), [q1687 — Datadog sovereign cloud playbook](/lab/q1687.html).