PlanetScale.

Here is a fact that databases people know and everyone else finds baffling: the scariest thing an engineer does all week is often change a column. Not add a feature, not deploy code - alter a table. For decades, modifying the shape of a production database meant locking it, holding your breath, and hoping the migration finished before the site fell over. PlanetScale's entire business is the proposition that this fear is a bug, not a law of nature, and that it can be engineered away.

The company was founded in 2018 by Jiten Vaidya and Sugu Sougoumarane, who are not database tourists. They were among the creators of Vitess, a system built inside YouTube starting around 2010 to stop MySQL from buckling under video-scale traffic. Vitess did the unglamorous work of sharding - slicing one enormous database into many smaller ones while pretending, to the application, that it was still just one database. It worked well enough that Slack, GitHub, Square, Etsy and Pinterest eventually ran on it, and well enough that it became a graduated project at the Cloud Native Computing Foundation.

So the founders had a piece of open-source infrastructure that some of the largest sites on the internet depended on. What they did next is the interesting business decision. Rather than keep Vitess as a thing you had to be a Google-caliber team to operate, they built a company to run it for you. The logic is the logic of every cloud database: the hard part was never the SQL, it was the operations - the failovers, the backups, the connection limits, the 3 a.m. migrations. If you can sell the operations as a service, you can sell it to everyone who has the problem but not the platform team.

Databases That Work Like Git

The feature that made PlanetScale famous with developers is branching. You take your production database and you branch it, the way you branch a Git repository. You make your schema changes on the branch. You open something called a deploy request - the database equivalent of a pull request - and the change gets reviewed, tested against the real schema, and then merged back without locking the table or taking the application down. If it goes wrong, you roll it back.

Described plainly it sounds like developer-tool theater, a gimmick that maps a familiar workflow onto an unfamiliar object. It stops sounding like theater the first time a bad migration gets caught in a review instead of in production at peak traffic. The pitch is not really "branching is cool." The pitch is "the thing you were afraid of is now reviewable, testable, and reversible." That is a different product than a faster database. It is insurance sold as a workflow.

None of this would matter if the underlying thing were slow, which brings us to the part of the story where PlanetScale started talking about hardware in a way most database companies avoid.

The Metal Argument

In March 2025 the company launched PlanetScale Metal, a class of database nodes that run on locally-attached NVMe SSD drives instead of the network-attached storage that most cloud databases quietly rely on. The distinction is technical but the consequence is not: network storage meters your input/output and charges for it, and it sits a network hop away from your compute. Local NVMe sits on the same machine. PlanetScale's marketing word for the result is "unlimited I/O," which is the kind of phrase that makes a careful reader reach for a footnote.

The numbers they published are the actual argument. Metal, they said, had been running in production for three months before launch, serving over 5 trillion queries across 5 petabytes of storage, with some workloads seeing p99 query latency drop as much as 65% and costs falling 53% compared with Amazon Aurora. You should read benchmarks a company publishes about itself the way you read a restaurant's description of its own food - directionally, not literally. But the direction is the point: PlanetScale is arguing that a large chunk of what you pay a cloud database is a tax on architecture everyone else assumed was fixed.

Current pricing follows the same instinct. A single-node Postgres instance starts around $5 a month; a production, highly-available three-node cluster starts in the tens of dollars and climbs from there into serious enterprise territory. This is not a free-forever consumer product. In 2024 the company removed its free Hobby tier and took public criticism for it, the way companies always do when they stop subsidizing hobbyists. It kept growing anyway, which is its own kind of data point.

The MySQL Company Ships Postgres

For most of its life PlanetScale was, proudly, a MySQL company. Vitess is MySQL. The team loved MySQL loudly and on the record. So the most revealing decision of 2025 was shipping Postgres. By the company's own telling, the demand showed up on the day they launched Metal - customers who wanted the NVMe economics but ran Postgres, and who would not stop asking. By the end of that launch day, the founders have said, they knew they had to build it. PlanetScale for Postgres was announced in mid-2025 and reached general availability the same year.

There is a tidy lesson here for anyone who has ever drawn a firm line around what their product will never do. When your own customers are begging for the thing you swore off, that is not noise to be managed. That is the roadmap arriving early. The next chapter is already named: Neki, a horizontal-sharding solution for Postgres - which is to say, the same brutal scaling problem the founders solved once with Vitess, now attempted a second time on a different database. Solving it twice would not be luck. It would be a team that knows exactly where the bodies are buried.

Running the company through all of this is Sam Lambert, the CEO, who came from GitHub, where he ran engineering. His public argument is almost aggressively unexciting: the database should be the most boring part of your infrastructure. Excitement in a database, in his framing, usually means an outage. It is a good slogan for a company whose best-case outcome is that you never think about it - that the migration ships, the failover happens, the query returns, and you go back to building the thing that actually makes you money. PlanetScale is selling the absence of a bad night. That is a harder thing to market than speed, and, if they are right about the fear, a much larger thing to sell.