Rebuilding the Database from First Principles
Before Dylan Cui built one of the most-starred open-source databases on the planet, he was untangling one of China's most unwieldy Redis deployments. At Wandoujia - the mobile app company that eventually landed in Alibaba's orbit - Cui and his colleagues ran an infrastructure stack that worked, barely, until it didn't. The database kept getting in the way. Not theoretically. Literally.
That friction became the founding thesis of PingCAP. In April 2015, Cui left Wandoujia alongside Max Liu and Edward Huang to build a distributed SQL database that could scale horizontally without forcing engineers to choose between consistency and speed. They called it TiDB. Within a decade, it had become one of the most widely deployed open-source NewSQL databases in the world, adopted by thousands of enterprises across fintech, gaming, e-commerce, and logistics.
Cui's role at PingCAP has never been purely organizational. He came in as a co-author and core committer of TiDB itself - the kind of founder who ships code, not just strategy decks. His background runs through ads infrastructure at Sogou and Wandoujia, the kind of backend work that teaches you exactly what breaks under pressure and why MySQL alone was never going to cut it at scale.
"We wanted to build a database that would make database scaling invisible to the application - you just add nodes, and it works."
- PingCAP founding missionPingCAP's approach was deliberate: TiDB would be MySQL-compatible from day one. Not because MySQL was perfect - it wasn't - but because the world had already bet on it. Switching databases is expensive. Making your new distributed database speak fluent MySQL means the migration cost drops to near-zero. That single design decision became one of TiDB's most powerful growth levers.
The Redis Origin Story
Most infrastructure companies are born from failure - not of imagination but of scale. Dylan Cui and the other PingCAP co-founders were running one of the world's largest Redis clusters at Wandoujia when the limits of existing database solutions became undeniable. They had tried the workarounds: manual sharding, proxy layers, custom middleware. None of it was sustainable. The real problem wasn't any single tool - it was the architecture. So they went back to first principles. They read the Google Spanner and F1 papers. Then they built TiDB.
Why TiDB Is a Different Kind of Bet
TiDB sits at an unusual intersection. It handles both transactional workloads (OLTP) and analytical workloads (OLAP) in a single system - what the industry calls HTAP, or Hybrid Transactional/Analytical Processing. Most databases force a choice: you either optimize for fast writes or for complex analytics. TiDB, through its TiFlash columnar storage engine, refuses that trade-off.
The underlying storage layer - TiKV - is built on the Raft consensus algorithm and runs as a separate distributed key-value store. In September 2020, TiKV became a CNCF Graduated project, placing it in the same tier as Kubernetes and Prometheus. That was a signal: this wasn't a startup vanity project. The open-source infrastructure community had reviewed it, stress-tested it, and certified it for production at scale.
The practical result: a company running TiDB can run real-time analytics against live transactional data without maintaining a separate data warehouse, without complex ETL pipelines, and without the 24-hour lag that comes from nightly batch jobs. For fintech companies that need both fast writes and instant audit queries, or gaming studios tracking player behavior in real time, the value proposition is clear.
From Three Engineers to a $3 Billion Infrastructure Company
PingCAP's growth from a three-person bet to a 600-person global company happened without the usual playbook. They didn't start with enterprise sales. They started with GitHub. The decision to open-source TiDB from day one was strategic - open-source databases are auditable, trustworthy, and community-validated in ways that proprietary systems can never fully be. For infrastructure software, trust is the product.
The funding story tells a similar arc. A $1.6M seed in 2015. A $5M Series A from Yunqi Partners in 2016. $15M in 2017. $50M in 2018, when PingCAP began seriously expanding into North America. And then Series D: $270 million in November 2020, valuing the company at approximately $3 billion. For a company headquartered in Beijing building infrastructure software, reaching that valuation - before the current wave of database cloud spending - was a signal that the market recognized what Cui, Liu, and Huang had built.
PingCAP's mission statement - "the best and most respected infrastructure company in the world" - reads as ambitious. But the trajectory of TiDB suggests it's less bravado and more north star: a company that decided to compete on quality of engineering rather than on sales process or lock-in.
The Journey
What He Helped Build
TiDB - Open Source Icon
Co-authored a distributed SQL database that reached 33,000+ GitHub stars and thousands of enterprise deployments across 30+ countries.
CNCF Graduated: TiKV
Helped build TiKV to the point where the CNCF - the body behind Kubernetes - certified it as a production-grade cloud-native project in 2020.
$341M+ Capital Raised
Helped build a company that attracted over $341 million in institutional funding and a $3 billion valuation - rare for a database company built outside Silicon Valley.
MySQL Compatibility Play
Co-designed TiDB's MySQL compatibility from the ground up - the feature that turned a technically impressive database into an enterprise product anyone could migrate to.
HTAP Pioneer
Helped pioneer commercial HTAP databases, enabling enterprises to run real-time analytics on live transactional data without separate systems.
600-Person Global Team
Co-built a company from three engineers to 600+ employees with offices in Beijing, Sunnyvale, and beyond - all organized around open-source first principles.
The Bet on Open Infrastructure
There is a specific kind of ambition in infrastructure software: you are not building something users love, you are building something users trust. The distinction matters. Users love apps. Users trust databases. Love can be won by design. Trust is earned by uptime, correctness, and documentation that tells the truth about edge cases.
Dylan Cui's technical fingerprints are all over TiDB's approach to that trust question. The decision to publish TiDB as open-source meant that any organization could read the code, run it on their own hardware, and verify its behavior before committing to it. In the database market - where vendor lock-in has historically been a feature, not a bug - that was a radical position.
TiDB's architecture also reflects a specific reading of the future of data. Dylan Cui and the PingCAP team bet early that the artificial boundary between OLTP and OLAP databases was a historical accident, not a technical necessity. The rise of TiFlash - TiDB's columnar analytics engine running alongside the row-based TiKV storage - proved that bet correct. Companies that once maintained separate MySQL instances for transactions and Hadoop clusters for analytics can now run both workloads on a single TiDB deployment.
The cloud-native trajectory matters too. TiDB runs natively on Kubernetes, comes with Terraform support, and integrates across AWS, Azure, and Google Cloud Platform. PingCAP launched TiDB Cloud as a managed DBaaS product, bringing the operational simplicity of cloud databases to TiDB's enterprise capabilities. For Cui and colleagues, the goal was always portability: your database should work wherever your infrastructure lives, not the other way around.
PingCAP's goal is to become "the best and most respected infrastructure company in the world" - a mission that starts with building a database too good to ignore.
- PingCAP official mission statementThe technology stack Dylan Cui helped build now stretches well beyond the core database. TiCDC handles change data capture for event-driven architectures. TiSpark integrates with Apache Spark for big data processing. The TiDB ecosystem touches Kafka, Flink, Kubernetes operators, vector search for AI workloads, and serverless database deployments. What started as three engineers frustrated with Redis scaling has become a complete data infrastructure platform.