Scott & the Quiet JVM

Scott Sellers does not look like a Silicon Valley antagonist. He runs an enterprise software company out of an office park off Moffett Park Drive, gives careful interviews, and has the bearing of an electrical engineer who would rather show you the benchmark than win the room. Then you read the receipts. Azul, the company he co-founded in 2002 and still runs, sells a drop-in replacement for the Java runtime Oracle charges by the processor for. Tens of thousands of enterprise JVMs around the world quietly tick along on Azul's builds instead. That is not a positioning statement. That is a line item on someone else's revenue forecast.

Azul makes two products that most people outside server rooms never see. Zulu is its free, certified OpenJDK distribution, the one downloaded by developers who simply want a Java that works. Platform Prime, formerly Zing, is the paid one, a JVM tuned for low-latency trading floors and high-throughput backends where a 50-millisecond garbage collection pause is the difference between a profitable quarter and a postmortem. Sellers will, given the opening, talk about C4 collectors and Falcon JIT compilers with the patience of a man who has been doing it since the Bush administration. The first one.

To understand why he is still at it, you have to back up to 1994. Sellers, fresh off Silicon Graphics workstations and a stint as a CPU architect at a company called Pellucid, co-founded 3dfx Interactive. 3dfx was the reason a generation of PC gamers learned the word "Voodoo." The Voodoo Graphics card, and its successors, did something previously reserved for $50,000 SGI machines: real-time, hardware-accelerated 3D, on a beige tower in a bedroom. 3dfx went public in 1997. Three years later, it was acquired by a company called NVIDIA. You may have heard of them since.

This is the detail that explains everything that came after. Sellers has now co-founded two category-defining companies in two separate hardware eras, and the through-line is not graphics or Java. It is the willingness to bet a decade on a single engineering thesis: that the right runtime, on the right silicon, can outrun the obvious choice. At 3dfx the thesis was that consumer PCs would render 3D. At Azul the thesis was that managed runtimes would not, in fact, be replaced by Go and Rust and a YAML file. Each thesis took roughly ten years to land. Each was correct.

Azul did not start where it ended up. The first version was a hardware company. Sellers, Gil Tene, and Shyam Pillalamarri set out to build a custom microprocessor specifically engineered for Java. "We felt," Sellers has said of the founding moment, "that we could build a microprocessor that was incredibly efficient at running Java with some hooks built into that hardware that would make the pains of Java go away." The Vega appliances shipped. Banks bought them. Then the laws of commodity x86 economics ate the hardware market alive and Azul made the hardest call a founder ever has to make: keep the customers, drop the products. The company pivoted to software, took the runtime engineering it had done in silicon, and shipped it as a pure JVM. The customer list largely came with it.

If you want a single statistic that captures Sellers' Azul, here is one. The company has raised about $177 million across its life. Oracle's Java SE business is widely estimated to be a multi-billion-dollar annual line. Azul does not need to take all of it. It only needs to give every CFO with a six-figure Oracle quote a credible second number to write down on the same slide.

That, increasingly, is what is happening. In January 2023, Oracle changed its Java SE licensing to a per-employee model. Within months, Azul was the first phone call. By 2024 and 2025 the company was shipping JVM Inventory, a tool whose pitch is roughly "we'll find every Java process running anywhere in your company, including the ones your team forgot exist, so you can stop paying Oracle to license them." It is the kind of product only a Java company that genuinely loves Java would build. Azul launched Intelligence Cloud on top of it: a SaaS layer that turns runtime telemetry into recommendations for dead code, security exposure, and cloud-cost waste. Sellers describes the company's stance with the matter-of-factness of someone who has watched this movie before. "Our product Platform Core is a true drop-in replacement for Oracle. Economically, ours is much more affordable."

Then there is the AI question, the one every CEO of every infrastructure company gets asked twice a week now. Sellers' answer is the unfashionable one. He thinks Java, not Python, is where serious AI infrastructure ends up. The reasoning is unsexy and probably right: most of the data that AI systems will actually operate against lives inside Java applications already, and getting it out is harder than rewriting the model code. He is willing to wait out the consensus. He has done it before.

The biographical detail people repeat about Sellers - because it is too good not to repeat - is that he was the backup quarterback at Princeton behind Jason Garrett, who later coached the Dallas Cowboys. It is offered as trivia. It reads more like a thesis. Sellers is comfortable being the second name on a depth chart for as long as the situation requires, because he is paying attention to the rest of the field while everyone else is watching the starter. Eight patents in high-performance graphics and computing. Two startups taken from whiteboard to outcome. A magna cum laude electrical engineering degree from Princeton with Phi Beta Kappa to match. He does not bring those up. They are simply the receipts.

Azul today has around 530 employees, customers across financial services, telecommunications, retail, government, and the long boring tail of enterprise IT, and a strategic recapitalization completed in November 2025 that gave the company fresh capital to keep eating the margins out of a market Oracle treats as an annuity. The competitor that mattered when Sellers started was Sun Microsystems, which no longer exists. The competitor that matters now is Oracle, which would prefer that Sellers also no longer exists. He is still there. Patience, again, doing its quiet work.

There is a temptation, profiling a CEO of a mature infrastructure company, to wrap up with a forward-looking quote about the next chapter. Sellers has supplied them, dutifully, in many publications. The more interesting line is the one he gave to interviewers about Azul's founding pitch, which still describes the operating system of his career: the willingness to look at a problem most people consider settled, decide it is not, and stay in the chair long enough to prove it. Voodoo cards became NVIDIA. JVMs became Azul. Whatever is next, it is probably already in his backlog and the rest of the industry has not noticed yet.