case-study

Zoom's Pandemic Scale-Up

Zoom grew from around 10 million to over 300 million daily meeting participants in roughly three months, absorbed by a hybrid architecture and a distributed media routing design.

case-studyzoomelasticitycapacityhybrid-cloud

The event

Zoom reported approximately 10 million daily meeting participants in December 2019 and more than 300 million by April 2020 — a roughly thirty-fold increase in about four months, with no ability to forecast it and no option to fail.

What absorbed it

Hybrid capacity. Zoom operated its own data centres for baseline capacity while using public cloud for elasticity. That combination is what made the surge survivable: owned infrastructure is cheaper for predictable load, and public cloud provides capacity that can be acquired in days rather than the months a hardware order takes. Neither alone would have worked — pure owned capacity cannot be procured that fast; pure cloud at that volume is far more expensive at baseline.

A distributed media routing architecture. Media is relayed through geographically distributed routers rather than through a central point, so participants connect to something near them and capacity is added regionally where demand appears.

Selective forwarding rather than server-side mixing. Video conferencing can either composite all streams on the server (heavy CPU per meeting) or forward streams and let clients compose (much lighter server cost, more client work and more bandwidth). The latter scales far better, which is a design decision made years earlier that determined whether 2020 was survivable.

The uncomfortable part of the case study

The same period exposed security and privacy weaknesses — default settings that permitted meeting intrusion, and misleading claims about end-to-end encryption. Zoom paused feature development for 90 days to address security, which is a rare and instructive act.

The lesson is that scale and trust are separate requirements, and a system can succeed spectacularly at one while failing at the other. Growth increases the consequences of security design choices made when the product was small, and the design that scales is not automatically the design that is safe.

The transferable lessons

Architecture determines what surge you can absorb. No operational heroics substitute for a design where capacity is regional, additive and not bottlenecked centrally.

Hybrid is a legitimate strategy, not a transitional state — owned capacity for the predictable floor, cloud for the unpredictable ceiling.

Defaults are architecture. Most of Zoom's security incidents were about default settings, which determine the behaviour of nearly every user and deserve the scrutiny given to code.