How Instagram moved the service's more than 20 billion photos from AWS to Facebook servers
How Facebook Moved 20 Billion Instagram Photos Without You Noticing — From left to right, Instagram and Facebook engineers Pedro Canahuati, Patrick Bozeman, Rick Branson, Nick Shortway …
Context & Ripple Effects
Facebook paid $1 billion for Instagram in April 2012 only after Twitter's secret offer forced its hand, and the deal came with an awkward dependency: the photo service it bought ran entirely on Amazon's cloud. Two years later that dependency is gone — Instagram's engineering team, including Pedro Canahuati and Patrick Bozeman, has finished shifting more than 20 billion photos onto Facebook's own servers with no visible disruption to users.
The migration is the clearest sign yet that Instagram is being absorbed into Facebook's infrastructure rather than run as a standalone cost center — and the story traveled unusually far for a plumbing tale, picked up by VentureBeat, ZDNet, Business Insider, Data Center Knowledge and others alongside Wired's own report.
First-order effects
- Instagram's photo-serving costs move from metered AWS rent onto Facebook's already-built data centers, converting a per-photo variable expense into marginal cost on capacity Facebook owns.
- AWS loses one of its most famous startup customers — the poster child for building fast on Amazon's cloud — at a moment when Instagram's storage needs were among the largest any single tenant could claim.
Second-order effects
- For startups weighing an exit, the playbook now reads: build on AWS to ship quickly, but expect an acquirer to pull you onto its own fleet once scale makes owned hardware cheaper — weakening AWS's grip on post-acquisition workloads.
- Competing hyperscalers and Facebook's own Parse platform gain an argument that at true consumer scale, self-operated infrastructure beats public-cloud economics, pressuring Amazon to defend marquee tenants with pricing.
Third-order effects
- If the pattern holds across future acquisitions, large consumer-internet buyers will treat public cloud as a staging layer rather than a destination, structurally capping how much of the fastest-growing photo/video workload stays on AWS.
- Infrastructure ownership becomes part of acquisition math itself: a buyer's data-center footprint directly lowers the operating cost of what it buys, tilting M&A toward companies that can absorb workloads cheaply.
The trend: Consumer internet giants are pulling acquired workloads off public clouds and onto their own data centers once scale flips the economics from renting to owning.