On June 12, 2025, a significant Google Cloud outage reminded the tech world of a fundamental truth: even the most robust cloud providers can experience unexpected failures. What began as a routine policy update at approximately 10:45 AM PDT quickly cascaded into a widespread cloud service disruption, affecting millions of users and businesses worldwide.
The Outage
According to Google's Root Cause Analysis (RCA), the issue stemmed from a critical bug in dormant code that was inadvertently triggered during a policy update to regional Spanner tables. Those conditions arrived on June 12, 2025, at approximately 10:45 AM PDT, when a new policy update was inserted into Google Cloud's regional Spanner databases that Service Control depends on. You can read the full details in Google's official RCA.
Impact on Deepgram's Customers
For Deepgram, this outage presented both a challenge and an opportunity to demonstrate the value of our resilience-first architecture. Several of our customers whoalso leverage Google’s speech to text APIs, experienced significant service interruptions from Google speech-to-text during the outage window.
As Google's speech recognition APIs became unavailable, these customers found themselves unable to process those critical audio workloads. Customer support teams couldn't transcribe calls, media companies couldn't generate captions, and developers' applications that relied on real-time speech recognition ground to a halt.
However, customers who had implemented Deepgram as a backup solution to Google speech to text, or had recently completely migrated to Deepgram experienced minimal disruption. Many were able to seamlessly failover to Deepgram's APIs, maintaining business continuity while Google worked to restore their services.
Deepgram's Architecture: Built for Resilience
This incident highlighted the critical importance of the architectural decisions we made early in Deepgram's development. Unlike many cloud-native services that rely heavily on a single public cloud provider, we have architectured for resilience from the ground up.
Multi-Cloud by Design: Our api.deepgram.com infrastructure does not rely on any single public cloud provider. This deliberate architectural choice means that when one cloud provider experiences issues, our services continue to operate normally. We maintain infrastructure across multiple independent infrastructure providers and regions, ensuring that our customers' workloads remain stable even during major cloud outages.
Redundancy at Every Layer: We've built redundancy into every layer of our stack, from our API gateway and load balancing infrastructure to our model serving and storage layers. This isn't just about having backup servers—it's about having entirely separate infrastructure paths that can handle full production loads independently.
The Customer Migration Story
In the days following the Google outage, we saw a notable increase in customers evaluating Deepgram as an alternative to Google Cloud's speech services. Many had experienced firsthand the business impact of depending on a single cloud provider for critical functionality.
The migration process for these customers was straightforward. Deepgram's API is designed to be developer-friendly, with clear documentation and SDK support across multiple programming languages. Customers who had been using Google's Speech-to-Text API found they could switch to Deepgram with minimal code changes, often completing the integration in a matter of hours rather than days or weeks.
More importantly, these customers discovered that beyond reliability, Deepgram often provided superior accuracy and performance for their specific use cases, along with more flexible pricing models and better customer support.
Lessons for the Industry
The June 12 outage serves as a reminder that cloud resilience isn't just about choosing a "reliable" provider—it's about architectural decisions that account for the reality that any system can fail. As businesses become increasingly dependent on cloud services for critical operations, the importance of multi-cloud strategies and provider diversity becomes paramount.
For Deepgram, this event reinforced our commitment to building infrastructure our customers can depend on, regardless of what's happening in the broader cloud ecosystem. We'll continue investing in our multi-cloud architecture and redundancy systems, ensuring that when the next major cloud outage occurs—and there will be a next time—our customers can continue operating without interruption.
The cloud landscape is more robust—but also more complex—than ever. By choosing providers who’ve built resilience into their core architecture rather than bolting it on as an afterthought, businesses can ensure they're prepared for whatever challenges the cloud may throw their way.