The first Runpod data centers had laundry rooms nearby. They had household circuit breakers, residential internet, and two spouses who had been persuaded that roughly $50,000 of specialized computers made sense as a hobby. In late 2021, Zhen Lu and Pardeep Singh were running Ethereum-mining rigs in their separate New Jersey basements. The machines made some cryptocurrency. They did not repay the investment. Worse, Lu found the work boring.
A dull hobby can be expensive. It can also be clarifying. Ethereum was moving away from the kind of mining that used their graphics processors, and selling the hardware would have ended the experiment at a loss. Lu and Singh had been exploring machine learning, where the same GPUs could do far more interesting work. They converted the rigs into AI servers and found a better problem: the hardware was useful, but the software around it was miserable.
Environment setup broke. Networking hid behind fog. One machine was awkward; several machines behaved like a committee with no chair. Lu later compressed the experience into two words fit for a product brief written after midnight: “hot garbage.” Runpod began as an attempt to remove that friction. The company’s mythology has GPUs in it, but the founding irritation was software.
A scientist changes experiments
This was not Lu’s first reinvention. He studied chemistry at the University of Pittsburgh, then pursued a PhD in computational chemistry at Temple University from 2009 to 2015. His research included the electronic structure of DNA base pairs, work at the intersection of chemistry, mathematics, physics, and biology. The training suited his appetite for difficult questions. The pace did not.
After graduate school, Lu became an assistant professor of chemistry at the University of Pittsburgh at Johnstown. He enjoyed the subject, but came to believe teaching moved too slowly for the kind of immediate effect he wanted. Two years in, he quit. He spent a summer teaching himself to code and began interviewing for software engineering jobs. It was an unceremonious reset: a doctorate, a faculty title, and then the beginner’s task of proving he could build.
“I’m somebody that really likes challenges, and I’m not afraid to change my life around to give myself those challenges.”Zhen Lu
Singh took the chance on him at Comcast. They worked together for about six years, building distributed systems and helping a team grow from roughly eight people to nearly 100. The pairing acquired something more valuable than a clever founder-meets-founder story: a long record of seeing how the other person behaved when systems failed, people needed direction, and the neat plan met the untidy day.
Lu’s route into technology also explains a recurring instinct in his hiring. He has said he looks for trajectory over pedigree, a preference with autobiographical evidence behind it. At Comcast, he personally hired almost half of the expanded team. At Runpod, he continued reviewing every role before it was posted, even after the company had enough runway to be less careful. His phrase about compensation was warmer and slightly mischievous: “All of my people are priceless.” Every finance department deserves a sentence that makes the spreadsheet blink.
The go-to-market plan was a Reddit post
The two engineers spent about three months making a simple GPU cloud. Neither claimed any gift for marketing. There was no practiced founder patter and no grand launch choreography. Lu posted in AI-oriented subreddits and offered free compute time to anyone willing to try the product and report back. Early users arrived for the free GPUs. Their complaints, experiments, and odd use cases gave Runpod its first map.
Those beta testers became paying customers. Within nine months, the company had passed $1 million in revenue. The number matters, but the method says more. Runpod did not try to educate a theoretical market from a conference stage. It went to the people already wrestling with the problem, handed them a tool, and waited to hear where it hurt.
Reported developer community growth
Success immediately made the basements look less charming. Business customers wanted to run real workloads, but not on anonymous machines in private homes. Lu and Singh still did not reach first for venture capital. They formed revenue-sharing partnerships with data centers, expanding capacity without taking on debt. For almost two years, Runpod bootstrapped. There was no broad free tier because the service had to pay for itself.
This arrangement made capacity a daily discipline. If users opened the console and found no GPUs, affection could migrate elsewhere in one click. The company had to stay ahead of demand while learning what demand actually meant. Reddit and Discord served as both distribution and early-warning system. Support was product research conducted in public, occasionally with capital letters.
The chore that stopped fitting
Lu kept his Comcast job while Runpod grew through nights, weekends, and stray minutes. He answered support tickets from his phone, including at his son’s birthday. His wife told him to put it down and be present. He later admitted she was right. The queue was rising every week, and faster replies would not fix the mismatch. Neither would the fantasy that one heroic hire could absorb the whole machine.
“Sometimes it is a chore that quietly stops fitting inside your life.”Zhen Lu, on recognizing traction
Not long afterward, he left his full-time job. It is a more useful founder threshold than the ceremonial versions. A company becomes real when its recurring work no longer fits inside the founders’ old lives. The support ticket, humble and annoying, can tell the truth before a press release does.
The transition also asked Lu to surrender a comfortable identity. He had come late to software and spent years, as he put it, trying to earn the word engineer. He cared about clean abstractions and code he would not be embarrassed to have read line by line. Running infrastructure changed the test. “Good enough” became a question of customer risk: if this fails, how bad a day will somebody else have?
That distinction now shapes how he talks about speed. Work deep in the stack, where a failure can take down a customer’s production system, gets the slow treatment. Experiments elsewhere can move quickly. The slogan “move fast” becomes much less romantic when the thing moving fast is somebody else’s outage.
From useful cloud to depended-upon cloud
Institutional money eventually found Runpod. A venture investor discovered the company through its Reddit presence and began a conversation before Lu knew how to pitch. In 2024, after the platform had reached about 100,000 developers, Dell Technologies Capital and Intel Capital co-led a $20 million seed round. By January 2026, Runpod reported 500,000 developers, 31 regions, and a $120 million annual revenue run rate.
In June 2026, the company announced a $100 million Series A led by Summit Partners at a reported $1 billion valuation. It also said more than one million developers had used the platform and its serverless product had handled over ten billion requests. Runpod now spans development and training Pods, serverless inference, and clusters for multi-node work. The old mining rigs have become an origin prop. The harder problem is reliability at a scale where a customer can say, without metaphor, that if Runpod goes down, their company goes down.
Listen closely → find repeated pain → build the fix → teach it back to the community. Lu has described early relationships with builders as a flywheel: concentrated learning from ambitious users becomes a product that more people can use.
Lu’s product ambition is deliberately larger than renting graphics processors. He argues that AI developers should be able to move from experiment to training, fine-tuning, inference, and scale without stitching together a small republic of clouds. The fewer infrastructure choices a builder must make, the more attention remains for the model and the product. “Our goal is to be what this next generation of software developers grows up on,” he has said.
There is an elegant tension in that goal. Lu likes hard problems enough to reorganize his life around them. His company succeeds when millions of other developers do not have to notice those same problems. The scientist wants to understand the machinery. The CEO wants the machinery to disappear.
Still willing to begin again
The headline numbers make the story look inevitable in reverse: professor becomes engineer, mining rig becomes AI server, Reddit tester becomes enterprise customer. Lived forward, none of those arrows was guaranteed. Lu’s first year as a software engineer could have remained an interesting second act. The basement hardware could have been sold. Early users could have taken the free hours and vanished.
What endured was a habit of changing the experiment without abandoning the question. In chemistry, Lu chased difficult systems across disciplinary borders. At Comcast, he learned how software and teams scale together. At Runpod, he and Singh kept the feedback loop close enough that a support conversation could alter the product. Even the company’s largest financing arrived after years of customers paying it into existence.
A basement is a poor data center, but a fine place to learn whether anybody cares. Lu and Singh discovered that people cared very much. Then they faced the less cinematic work: making the circuit breakers, servers, partnerships, hiring decisions, and promises hold. The cloud left the basement. The experimental temperament did not.