Google Cloud Outage History: 50 of 86 High-Severity Incidents Began With a Google Change

Google Cloud outage history, rebuilt from Google’s own feed: 1,472 public incidents from 2013 to October 2026. Of the 86 high-severity incidents since 2018 where Google published a cause, 50 (58%) began with a Google configuration change, code rollout or maintenance step, including the June 12, 2025 Google Cloud outage that hit 55 products.

Google’s public incident file, status.cloud.google.com/incidents.json, only holds a rolling window of recent incidents, so no single download shows the history. We combined 91 readable monthly Internet Archive captures of that file since 2015 with the live copy from October 29, 2026, keeping the latest version of each incident. Counts for 2022 to 2025 match Google’s own per-product history pages exactly. We then read the root-cause text of every high-severity incident since 2018. It is the same approach behind our AWS outage history and the companion Azure outage casebook.

The June 12, 2025 Google Cloud outage

In Google’s words, “This occurred globally given each regional deployment.” Afterwards it froze all changes to the Service Control stack and manual policy pushes. The pattern is the one that dominates the record below: a change that passed a staged rollout, then a second change, here a data change, that hit every region at once.

Are Google Cloud’s big outages config and rollout driven?

More often than not, yes. Google rated 140 incidents high severity from 2018 to 2026. We set aside 12 duplicate records (several products filed separately for one event), 3 where Google’s statement does not determine a cause, and 39 with no root-cause statement at all, mostly before 2021. That leaves 86 incidents with a cause Google stated. 31 began with a configuration or operational change, such as a config push, a maintenance task, a migration or a cleanup job, and 18 with a code or binary rollout. Physical causes were 16: 6 power, cooling or facility events and 10 hardware, fiber or network link failures.

Coded by The Uptime Report from Google’s incident reports. “Own change” includes one fiber failure that Google says began during its own scheduled maintenance (August 20, 2026).

The share moves between periods but never drops below half: 16 of 23 in 2018 to 2020, 16 of 32 in 2021 to 2023, and 18 of 31 since 2024. Among high-severity incidents that hit six or more products, or Google’s shared infrastructure, 16 of 26 began with a change. Cloudflare’s post-mortems point the same way, as our Cloudflare outage analysis found.

Which Google Cloud products go down most?

Networking leads by a distance. Since 2018, Google Cloud Networking is named in 212 incidents and Compute Engine in 143, followed by Google BigQuery (104) and Google App Engine (94). An incident that lists several products counts once for each.

2018 to October 29, 2026. Before mid-2020 each record named one product; since then Google lists every affected product.

Incidents per product, per year

Bars run 2018 to 2026; red marks each product’s peak year. 7 of the 12 products peak in 2022 to 2024, and all 12 fall to 7 or fewer incidents a year from 2025, which tracks the publishing change described below.

How wide is the blast radius?

Most Google Cloud incidents stay inside one product. Since 2021, when the feed began listing every affected product, 565 of 900 incidents (63%) named one product and 13 named 21 or more. High-severity incidents reach further: 30% named six or more products. us-central1 (Iowa) appears in the location list of 316 of the 900 incidents, more than any other region.

The twelve incidents since 2021 that named the most products. Duration runs from Google’s begin to end time on the public record.

Google Cloud outage history by year and severity

Read the yearly totals with care, because Google changed what it posts twice. In 2022 it began posting informational notices, with status “Service information” and severity low; the first appears on May 06, 2022, and they made up 120 of 2022’s 210 incidents and 279 of 2023’s 351. Then Google narrowed the page itself. Its dashboard documentation, last updated September 5, 2024, listed “Major incidents” among the criteria; the version dated October 1, 2024 says “For most Google Cloud incidents, impacted customers receive incident communications directly through Personalized Service Health.” Public posts kept their pace for a few months (52 from October 2024 through January 2025), then fell to 5 in February 2025 and 2 in March. Nothing was posted from August 2025 through January 2026.

*2026 is January 1 to October 29 (8 incidents, against 22 in the same period of 2025). The last column counts unique incidents on Google’s per-product history pages, which keep five years; it matches our rebuild for every full year shown.

High-severity posts fell too, from 28 in 2024 to 4 in 2025, so the drop is not only the informational notices going away. Google now defines a “Broad Severe incident” as one with “global impact or affecting a significant percentage of customer projects across multiple regions.” By that rule, the public dashboard is a record of the largest incidents only.

Is Google Cloud down? How to check

• Personalized Service Health first. Google says limited-scope incidents, “such as those impacting a single location, zone, or a smaller subset of projects”, go to customers through Personalized Service Health in the console. Set its alerts to email, Pub/Sub or a webhook into your PSA.
• Then status.cloud.google.com. The public dashboard is for broad, severe incidents and for times when Personalized Service Health itself is unavailable. On June 12, 2025 it lagged the crashes by about an hour.
• Poll the feed, not the page. Google publishes the same data as status.cloud.google.com/incidents.json and an RSS feed, so your monitoring can check it on a schedule.
• Check per-product history. Each product page keeps up to five years of incidents, useful for a client’s renewal or SLA review.

What this means for MSPs

Treat Google’s change calendar as a risk calendar. 58% of high-severity incidents with a stated cause began with something Google changed. The June 2025 outage got through a region-by-region rollout because the failing path needed a later data change to fire. Staged rollout on the provider side does not remove the need for your own fallbacks.

Single-region designs cover most incidents, not the worst ones. 63% of incidents since 2021 named one product. The incidents that reach dozens of products, such as June 2025 (API control plane, every region) and February 2024 in us-west1 (a regional metadata store), are the ones that hurt. For workloads that must stay up, a second region, or a second provider for DNS and status pages, is the main protection against a shared control plane. Weigh that against SLA credit terms with our 99.9% uptime calculator.

Your clients will not see most incidents on the public page. Since late 2024, most incident notices go only to affected projects. If you manage client projects, wire Personalized Service Health into your ticketing per organization, or you will hear about incidents from users first.

Do not benchmark providers on public incident counts. Google’s 22 public incidents in 2025 and Microsoft’s Azure reviews follow different rules, and both narrowed theirs. Compare root causes and blast radius instead, as in the GitHub outages scorecard.

Frequently asked questions

Is there a Google Cloud outage right now?

This page is a history, not a live monitor. Check status.cloud.google.com for broad incidents and Personalized Service Health in the Google Cloud console for anything affecting your own projects. Google says incidents with a limited scope, such as a single zone, are communicated only through Personalized Service Health.

What caused the Google Cloud outage on June 12, 2025?

Google’s incident report says a policy change with unintended blank fields reached Service Control, its API policy and quota system, and triggered a null pointer in code released on May 29, 2025 without a feature flag. The binaries crashed in every region. Google’s public record lists 55 affected products and stayed open 7 h 27 min.

Has Google Cloud ever had a major outage?

Yes. Google’s public feed holds 140 high-severity incidents from 2018 to 2026. The widest by product count were June 12, 2025 (55 products) and February 14, 2024 in us-west1 (53 products).

What causes Google Cloud outages?

Of the 86 high-severity incidents since 2018 where Google published a cause, 50 (58%) began with Google’s own configuration change, code rollout or maintenance work. Physical causes such as power, cooling, fiber and hardware account for 16.

Why did Google Cloud report so few incidents in 2025?

Google’s public dashboard listed 164 incidents in 2024 and 22 in 2025. Google’s documentation changed in late 2024 to say most incidents go only to affected customers through Personalized Service Health, and the public page now carries broad, severe incidents. Google’s own product history pages show the same 22, so the drop is a publishing change, not evidence of fewer failures.

Which Google Cloud products have the most outages?

Since 2018, Google Cloud Networking is named in 212 public incidents and Compute Engine in 143, more than any other products. Since 2021, 63% of incidents named a single product.

Methodology

References

1. Google Cloud Service Health, incidents.json. Accessed October 2026.
2. Internet Archive, captures of status.cloud.google.com/incidents.json, 2015 to 2026. Accessed October 2026.
3. Google Cloud Service Health, history of incidents reported by product. Accessed October 2026.
4. Google Cloud, incident report, Multiple GCP products, June 12, 2025. Accessed October 2026.
5. Google Cloud, Use Google Cloud Service Health as a fallback. Accessed October 2026.
6. Google Cloud, Google Cloud incident communication. Accessed October 2026.
7. Google Cloud, Service Health dashboard FAQ, version last updated September 5, 2024 (Internet Archive). Accessed October 2026.
8. Google Cloud, Service Health dashboard FAQ, version last updated October 1, 2024 (Internet Archive). Accessed October 2026.