Facebook October 2021: 6 hours, $100M+ revenue, $60B market cap

Timeline

What happened, hour by hour

Timeline from Facebook's official post-incident write-up and contemporaneous reporting from Cloudflare.

Root Cause

The BGP withdrawal and the recovery cascade

During a routine maintenance operation, an engineer issued a command intended to assess global backbone capacity. A bug in the audit tooling that should have validated the command failed to catch a misconfigured variant of it. The command propagated and withdrew all of Facebook's BGP route advertisements from the global routing table. With no routes advertised, the global internet could not reach any of Facebook's DNS servers, which meant DNS resolution for facebook.com, instagram.com, whatsapp.com, and all other Facebook-controlled domains failed worldwide.

The recovery would have been straightforward in principle: re-advertise the routes. In practice, Facebook's own internal communication and operational tooling ran on the same network as the customer-facing services, so engineers could not reach the affected routers remotely. The badge access system that controlled physical entry to the data center buildings, also running on the same internal network, was reportedly affected, complicating physical access for the on-site response. Recovery required physically reaching the affected routers and restoring the route advertisements manually, which took several hours longer than the underlying technical problem would have required.

The cascade is the canonical example of why critical operational tooling should not run on the same infrastructure it manages. Modern operational guidance from Google's SRE community and others has long warned against this pattern. The Facebook incident is the most-cited illustration of why.

Economic Impact

Direct revenue, market cap, and competitive growth

The direct revenue figure derives from Facebook's ad-revenue run rate at the time (approximately $28 billion in Q3 2021, which annualises to ~$112B or ~$13M per hour averaged across all hours). Peak hours run higher; the affected window included US morning and European afternoon, both peak ad-served minutes. Estimates of direct ad-revenue loss above $100 million for the ~6 hour window are conservative.

The share-price impact is harder to attribute cleanly because the outage coincided with the Wall Street Journal's "Facebook Files" whistleblower disclosures, which were dominating financial-press coverage of Meta during the same week. The ~5% one-day drop and ~$60B market cap loss on 4 October 2021 was almost certainly amplified by the existing negative news flow. Most of the share-price drop reversed within several weeks, but the broader 2021-2022 share-price weakness in Meta partly traces back to this period of compounding negative narrative.

Competitive Impact

Telegram, Signal, and the migration moment

Telegram CEO Pavel Durov publicly disclosed that the company added 70 million new users on 4 October 2021, an unusually large single-day signup figure attributed primarily to WhatsApp being unavailable globally. Signal also reported a temporary signup spike, though smaller. Most of these new users remained active to some degree, with Telegram's headline MAU figure stepping up materially in the months following.

The competitive-migration cost is a line that does not appear in standard outage cost models. For a customer-facing service operating in a competitive market, an extended outage is a discrete moment when users actively seek and try alternatives. Most return after the original service is restored, but a meaningful fraction (typically 5 to 15% of those who tried an alternative) become long-term users of the alternative. For WhatsApp, with a customer base in the low billions, even a 1% defection to Telegram or Signal represents tens of millions of users. The lifetime-value loss from this defection materialises slowly and is rarely calculated, but it is real.

Lessons

What other operators learned

Frequently Asked

Common Questions