Most databases have a time problem. Not in the dramatic, science-fiction sense, but in a very boring, very expensive sense. When you run a database across multiple data centers in different parts of the world, the clocks on those machines don’t perfectly agree. You can sync them, but you can’t make them agree precisely enough to be certain which transaction happened first when two writes land in different regions at nearly the same moment.
The result is what the industry politely calls “eventual consistency.” Your application has to handle conflicts when they arise. In a social feed, that’s fine. In a payments ledger, it’s a compliance problem waiting to happen.
Google’s answer was to put atomic clocks and GPS receivers in every data center it operates. Not as a research project. As production infrastructure. This is what Cloud Spanner runs on, and it’s why Spanner can make consistency guarantees that no other cloud database can match.
The Physics Lesson You Didn’t Know You Needed
Google calls this infrastructure TrueTime. The key insight is simple even if the engineering behind it isn’t: instead of pretending clocks agree perfectly, TrueTime measures exactly how uncertain each timestamp is. When Spanner records a transaction, it knows the real time falls within a specific, very small window. So it waits for that window to close before committing, which means every committed transaction has a timestamp that is provably correct relative to every other transaction, anywhere in the world.
The practical result is a database that behaves like a single machine even when it’s spread across dozens of data centers on multiple continents. No duplicate charges from conflicting writes. Inventory oversells caused by two regions disagreeing on stock levels stop being a risk. No application code to resolve conflicts that shouldn’t have existed in the first place. The database just handles it, the way databases were always supposed to.
It’s worth pausing on how unusual this is. Most “globally distributed” databases are really just replication with a marketing budget. TrueTime is a fundamentally different approach, and the only reason Google can offer it is because they spent years and enormous capital building the physical infrastructure to make it work.
Why Nobody Else Can Offer This
CockroachDB is a genuinely good distributed SQL database and probably Spanner’s most credible competitor. CockroachDB’s own documentation acknowledges it offers “serializable” consistency, which is the industry’s gold standard. Spanner offers “external” consistency, which is stricter. The gap matters in practice: without hardware-backed time bounds, you can’t make the same ordering guarantees across regions. CockroachDB runs on commodity hardware, which is a real advantage if you need multi-cloud flexibility. But you can’t replicate TrueTime by buying servers at a hardware store.
AWS improved DynamoDB Global Tables with multi-region strong consistency in early 2026, which is a meaningful step forward. DynamoDB is still a NoSQL key-value store though, not a relational database, so if your workload involves complex transactional SQL it’s not really a comparison. Azure SQL Hyperscale is impressive at scale within a single region but isn’t designed for active-active multi-region at all. There’s no Microsoft equivalent to TrueTime, and there probably won’t be one soon, because it requires physical infrastructure that takes years to build and operates at a scale only a few companies in the world can sustain.
What This Means if You’re Building Software
For ISVs on GCP, TrueTime’s value shows up in two places. The first is internal: if you’re running a global payments platform, a multi-currency ledger, or any kind of inventory system across regions, eventual consistency isn’t an abstract concern. It’s a bug category. Using Spanner as your transaction store eliminates that entire class of problem at the infrastructure level, which means your engineers spend time building features instead of debugging race conditions at 2 AM.
The second is product. Spanner’s data residency controls let you build a single multi-tenant database that physically stores EU customer data in EU regions and US customer data in US regions, all within one logical instance. For anyone selling into regulated industries, that’s a data sovereignty story you can tell customers without maintaining separate database clusters per jurisdiction. That’s months of operational complexity converted into a product checkbox, which is the best kind of operational complexity.
Spanner also now handles relational, key-value, graph, and vector workloads against the same dataset, with native Vertex AI integration. For ISVs building AI features on top of operational data, that means your AI retrieval is working with current data rather than whatever was in the sync queue when the last ETL job ran. Fresher context, better responses, one less pipeline to babysit.
The One Real Tradeoff
Spanner runs on Google Cloud. That’s it. If your customers require multi-cloud deployment or self-hosted infrastructure, that’s a hard constraint and it’s worth being honest about that. But if your product genuinely needs global transactional consistency, there’s nothing else on the market that delivers what TrueTime makes possible. The moat isn’t a feature roadmap or a pricing strategy. It’s atomic clocks in buildings. That’s not something a competitor ships in a software update.