Back in September 2020, one of the guests at the East Asian round of the Topcoder Open (TCO) competition was Lou Tiancheng, now best known as the co-founder and CTO of autonomous driving firm Pony.ai. Lou is regarded as a legendary figure among TCO observers, having dominated the Topcoder leaderboard for a decade. For the 2020 event, he was invited to share his experiences with more than 100 younger contestants.
Half an hour after his talk ended, the contest began, and Lou’s name suddenly appeared on the competitors list. Without telling anyone, he jumped in himself, taking the stage not as a mentor but as a rival.
That spontaneous moment of competitive fervor captures Lou’s character and, in many ways, the core DNA of Pony.ai. To him, whether in programming or entrepreneurship, life itself is a competition, and the only mission is to win.
That relentless drive for excellence has been embedded into Pony.ai’s culture. Over the past nine years, the company he co-founded with CEO James Peng has remained one of the most visible players in the autonomous driving race. From the outset, the team refused to take shortcuts.
When the industry entered a deep freeze in 2021 and many firms pivoted from Level 4 autonomous driving to simpler assisted driving products, Pony.ai chose the harder route, confronting the challenge head-on and undergoing a difficult internal reorganization. Perfection was the only acceptable standard.
Lou told 36Kr in an interview that he realized early on that in top-tier competition, everyone at the summit has talent, luck, and skill. “The only things truly in your control are focus and hard work,” he said.
In November 2024, Pony.ai went public on Nasdaq, becoming the world’s first listed robotaxi company. Less than a year later, it debuted on the Hong Kong Stock Exchange (HKEX), marking not just a dual primary listing across the US and Hong Kong but what Lou called the company’s coming-of-age moment.
Pony.ai’s nine-year story is one of a startup navigating uncharted territory, and a saga of persistence, pain, and conviction.
From one line of code to a fleet of cars
Pony.ai was founded in late 2016 in a modest Silicon Valley office. CEO Peng had spent seven years at Google and served as chief architect at Baidu’s US research center. CTO Lou, already well-known in competitive programming circles, had worked on the self-driving project that would become Waymo. With such a powerhouse founding duo, the startup quickly assembled what insiders called a dream team.
Rumor had it that whenever the two appeared in a Silicon Valley cafe, another top engineer would soon resign to join them. The early team, composed mostly of Tsinghua University alumni and programming contest winners, was united by an audacious goal: achieving Level 4 autonomous driving.
By 2017, that goal remained a distant lighthouse. To test their ideas in reality, the team needed a breakthrough.
In the second half of 2017, Pony.ai shifted operations to China, setting up an office in Nansha, Guangzhou. Facing local authorities, the team made a bold promise: to launch China’s first public robotaxi service before the Lunar New Year.
The timeline was almost impossible. With only months to prepare and everything to build from scratch, the team launched an all-out technical mobilization spanning China and the US. “It was like we’d been injected with adrenaline,” Peng later told 36Kr. Overnight debugging sessions became routine. Lou was said to work roughly 100 hours a week, driving his engineers with the same competitive intensity.
The showdown came just before the holiday. On February 2, 2018, a fleet of modified Lincoln MKZ robotaxis hit the public roads of Nansha, giving local residents their first autonomous rides. Each car drove itself out of the garage, stopped precisely for passengers, and delivered them smoothly to their destinations.
RELATED ARTICLE
Surviving the grind
If the early days were fueled by talent and adrenaline, the years between 2019–2022 were the hardest and most defining.
“My perception of smart driving changed drastically around 2019,” Lou told 36Kr years later. Until then, he had believed that collecting enough real-world driving data would naturally lead to smarter autonomous systems. But as development progressed, he hit an invisible wall: data alone could no longer deliver breakthroughs.
“For Level 2 assisted driving, matching human performance is enough, so data accumulation works. But Level 4 needs to outperform human cognition by ten times,” Lou said. The problem, he added, is that most human-generated driving data is mediocre, or simply bad. “If you learn chess from weak players, you’ll never surpass them.”
That insight led him to a radical conclusion: the better you get at Level 2 autonomous driving, the further you drift from Level 4.
Lou made a bold decision: to scrap three years of accumulated technology and rebuild from zero around a new framework he called the “world model.”
Instead of passively imitating human drivers, the new system used high-fidelity virtual environments that allowed artificial intelligence to practice endlessly in simulation, evolving toward optimal performance far beyond human ability.
“Don’t expect visible progress for two years,” Lou warned his team. It was a massive gamble. From the outside, Pony.ai appeared stagnant, and morale inside began to waver.
Meanwhile, as skepticism about Level 4 technology mounted, competitors pivoted toward cash-generating Level 2 products. Pony.ai, however, held its ground. The company pulled back from side projects and funneled all resources into the development of Level 4 robotaxis and robotrucks, staking its future in uncharted territory.
That conviction stemmed from deep technical faith. “The world model is the most important thing, period,” Lou said. He described it as a near-omniscient simulation engine capable of recreating complex traffic scenarios, from hidden pedestrians darting across the street to runners decelerating with near-gravitational force. Acting as an all-seeing referee, it graded the system’s performance with precision and objectivity.
The payoff was tangible. Pony.ai’s architecture enabled 500,000 hours of fully unmanned operation across rain, night, and complex intersections. The model reportedly generated more than ten billion kilometers of simulated data weekly, dramatically improving training efficiency while reducing real-world risk. Using mass-production hardware, system safety surpassed human driving by a factor of ten.
Leaving the lab
By late 2022, Pony.ai began to see light as Beijing and Guangzhou granted driverless operation permits.
Commercialization gears started turning, but profitability required crossing one key threshold: the 1,000-car mark.
“In megacities like Beijing, Shanghai, Guangzhou, and Shenzhen, robotaxi operations break even only after about 1,000 vehicles,” said vice president Zhang Ning.
Beyond that point, network effects kick in, creating a self-reinforcing loop. Peng elaborated: “Autonomous driving scales non-linearly. Managing 10,000 cars isn’t ten times harder than 1,000, it’s maybe three times.”
To reach that scale, Pony.ai launched “Kunlun” in 2023, its robotaxi mass production program. The seventh-generation system achieved three milestones: a 70% cost reduction from the prior generation, fully automotive-grade components rated for 600,000 kilometers, and a modular design adaptable across vehicle types.
That shift turned robotaxis from costly lab experiments into replicable commercial products.
In 2024, Pony.ai announced manufacturing partnerships with BAIC Group, GAC Aion, and Toyota. Vehicles equipped with the seventh-generation system entered mass production. Today, its robotaxi fleet exceeds 720 vehicles, operating around the clock in Guangzhou and Shenzhen with more than 500,000 hours of unmanned driving logged.
“With technology no longer the bottleneck,” Lou said, “the remaining challenges are production scale, cost, and operations.” For commercialization, he set a bold milestone: “We’ll reach per-car profitability next year.”
In an industry still burning cash, that statement was a jolt of adrenaline. It suggested not just technical maturity but a viable business model.
From the Middle East to the global stage
The trajectory of any deep tech firm depends on steady capital flow. Pony.ai’s funding path mirrors the evolution of the entire sector, starting with early-stage backing by top-tier US venture capital to strategic investments from automakers such as Toyota, and later, support from sovereign funds like the Neom Investment Fund.
These investments signal confidence in Pony.ai’s persistence. With dual listings on Nasdaq and HKEX, the company has built a global financing network. In June this year, it became the first Level 4 autonomous driving company included in the Nasdaq Golden Dragon China Index, underscoring that mainstream investors now recognize Chinese autonomous driving technology. Ark Invest’s Cathie Wood has repeatedly increased holdings, while Fidelity Investments and Baillie Gifford have continued to buy in, signaling long-term faith in Pony.ai’s technical and commercial potential.
Armed with fresh capital, Pony.ai has gone global, expanding into the Middle East and Europe and partnering with governments and enterprises in the UAE, Qatar, and Singapore to deploy robotaxi services.
Lou once said that as a challenger, the key is to stay close enough to the leader to strike when it stumbles. “Waymo’s indifference toward the Chinese market was one of its biggest mistakes,” he said. Pony.ai seized that opening, refining its systems in China’s chaotic yet dynamic traffic environments—a crucible few competitors dared to enter.
Now, equipped with experience proven in China, Pony.ai is stepping onto the global stage. From a single line of code in Silicon Valley to fleets operating in multiple cities worldwide, it is helping write a new chapter in Chinese technology globalization.
“The automotive sector requires a long game,” Peng told 36Kr. “As long as you keep building good products, resources will flow to the leaders. The goal is to stay at the top.”
IDG Capital partner Guo Yihong described Pony.ai’s growth as “consistently solid, pushing forward year after year, avoiding major mistakes, staying united, and quietly pulling ahead.” That steady rhythm, reminiscent of a marathon runner, may be its greatest strength.
Even under commercial pressure, the company has never abandoned its pursuit of Level 4 autonomy. When Pony.ai rolled out its “Level 2++” system as a bridge between assisted and fully autonomous driving, Peng emphasized that it would not compete in the crowded Level 2 space but focus on technology that shares nearly 70% of its stack with robotaxis, edging closer to Level 3 and 4 capabilities.
Lou once mused to friends that the world itself is a simulation and each of us a fleeting memory cell. That impermanence, he said, makes creating value all the more meaningful.
“Some people are given talent because there’s something they are meant to do,” he said. “Whether or not you achieve it isn’t the point. But misuse it, and heaven may take it back.”
Lou’s hunger for victory, obsession with technology, and faith in hard work all converge on one goal: to honor that gift by solving a problem worthy of it.
For Lou and his team, the race for autonomous driving is far from over. As AI reshapes the world, the large-scale commercialization of robotaxis is only just beginning.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Xi for 36Kr.