Skype ran voice calls on users' PCs, scaling to 663M users without phone networks

Skype launched in 2003 from the team behind Kazaa, applying file-sharing's peer-to-peer architecture to voice. The Columbia paper by Baset and Schulzrinne analyzed its traffic and found two node classes: ordinary clients and supernodes, with any client that had a public IP address and spare CPU, memory and bandwidth eligible to become a supernode.

Apart from a login server that authenticates users and keeps names unique, there is no central server. User directories, presence and search queries propagate across the overlay, and clients use a STUN-like technique to work through NATs and firewalls, with wideband codecs keeping calls clear at 32 kb/s.

The architecture turned every user's computer into part of the network, so adding users added capacity instead of load. By May 2011 the Luxembourg-based company had 663 million users and Microsoft agreed to buy it for $8.5 billion, its largest acquisition.

Turn every user's PC into the phone network

User machines supplied the routing and search capacity

NAT traversal let calls reach users behind firewalls

No per-call server cost meant free calling at scale

End-to-end encryption was built in from the start

Infrastructure can be designed as a by-product of use: if every customer contributes capacity, the network scales without the provider buying hardware.

Microsoft paid $8.5bn for Skype in 2011, with 663 million users at the time; the acquisition was Microsoft's largest ever. Skype's peer-to-peer lineage gradually faded as Microsoft moved it onto cloud infrastructure, but the lesson—users as infrastructure—survived in later peer-to-peer products.