Portland data centre loses power twice in five months
Mar 2024Published 4 Oct 2026Updated 13 Oct 2026Responsible editor: Chief Editor
Get the next one by email
Free alerts on data centre incidents in the United States, and an email when EIR-1989 is updated. No account needed; unsubscribe in one click.
Confirm your email and every data centre incident on the register arrives at once as a spreadsheet (CSV), free.
Get a free sample brief Three complete verified records and the incidents-per-GW table, by email within a minute.
We will also email you about the register's paid records and plans, at most once a week and three times in all. You can stop these at any time. By signing up you agree to the privacy policy.
Companies named in the reporting of this entry (being named is not a finding of fault): Flexential
What happened
On 26 March 2024 at 14:58 UTC, the PDX01 data centre in Portland, Oregon, lost all power, less than five months after an earlier outage at the same facility. Cloudflare says the loss followed a reportedly simultaneous failure of four Flexential-owned and operated switchboards serving all of Cloudflare's cages, which deactivated both the primary and redundant power paths across the environment. Cloudflare's systems reacted automatically. By 15:05 UTC its APIs and dashboards were operating normally without human intervention, although a few specific services needed manual recovery.
The data plane, which carries customer traffic, was not affected. The Analytics platform, which depends on PDX01, was impacted and was not fully restored until later that day. The facility's cold start took roughly 10 hours, compared with about 72 hours after the November 2023 outage.
Contributing factors
Cloudflare states that an initial assessment of the root cause of the switchboard failures points to incorrectly set breaker coordination settings within the four switchboards as one contributing factor. When one or more breakers tripped, the remaining active boards failed in a cascade, causing a total loss of power to Cloudflare's cage and others on the shared infrastructure. Cloudflare was told that the facility team found the incorrect trip settings, reset the boards and adjusted them to expected values, which allowed servers to be powered up in a staged way. Cloudflare did not know when the settings were established. The Analytics platform was dependent on a single facility.
Lessons learned
- Verify breaker coordination settings against downstream provisioned loads during commissioning and after any change, and keep the supporting study on record.
- Test automatic failover regularly, including full cut tests, so that recovery is proven rather than assumed.
- Remove single-facility dependencies from analytics and logging pipelines.
- Plan and rehearse staged cold restart procedures so that power is restored to large loads in a controlled order.
More lessons on electrical protection and isolation, worker safety and safe systems of work, from every incident on the register.
Source
Published 2026-10-04. Written from public reporting. Descriptive, not investigative. See the data accuracy disclaimer.