Eight GPUs and a Face-Swap Campaign That Ran Across 80 Countries
In early 2023, Coca-Cola was looking for an AI vendor. They benchmarked every competitor they could find. Then they picked Jeff Lu's company - not because it had the biggest team or the most funding, but because its technology was, in their assessment, significantly better. The resulting campaign: a League of Legends promotional face-swap running across 80-plus countries, generating millions of swaps. That one client changed everything.
What made it remarkable wasn't just the result. It was the math behind it. AKOOL - the company Lu had been quietly building since 2020 - ran its entire AI model training operation on eight graphics processing units. Not eight hundred. Eight. His team of around ten people had built specialized "human models" in weeks - AI that generates realistic digital avatars with a control of the full pipeline from algorithm to data collection to training that larger companies, moving slower with bigger budgets, hadn't matched.
Lu, who goes by Jeff but whose given name is Jiajun, spent his career working at the exact intersection where this kind of technology lives. He interned at Microsoft, earned his PhD at the University of Illinois at Urbana-Champaign working in AI and computer vision, did time at Stanford's Computer Graphics Lab and at Nreal (the augmented reality startup), then landed at Apple in 2018 where he worked on Face ID - the same technology that lets your iPhone recognize your face in under a second. After Apple came Google Cloud, where he focused on video processing. And that's where things got interesting.
"I had to quit Google," Lu said. "At the time, they were very conservative about generative AI." The specific frustration: "They didn't want us working on any products where they couldn't control the outcomes." For someone who had spent a decade watching AI's capability curve, watching a company pump the brakes at exactly the wrong moment was its own kind of education. He left in 2020.
He didn't immediately raise money or announce a company. He started building AKOOL as a side project. By the end of its second year, it had generated just over $100,000 in revenue - a number that, in retrospect, has a certain charm. In 2025, that number became $40 million. The Inc. 5000 gave it the #1 slot with 37,364% growth over three years - a growth rate so absurd it sounds like a typo until you check the math twice.
The platform AKOOL built has two products that pull most of the weight. One is streaming avatars - customizable AI-generated people capable of holding real-time conversations. Roughly half of B2B sales run through this product. The other is a video translation tool that converts a video from one language to another while syncing the speaker's mouth movements to the translated audio in more than 150 languages. That accounts for around 30% of revenue. Together they answer a question brands have been asking since global marketing became standard: how do you make content that feels local in every market?
AKOOL's answer is to not just translate the words but to move the face. Multilingual lip sync at scale, running in real-time. Amazon, Google, and Nvidia have used AKOOL to build technology demonstrations for conferences. The company has generated more than 300 million AI assets for Fortune 500 clients. And all of this runs out of a Palo Alto office - specifically the building where Facebook was originally founded in 2005. The address has a particular kind of Silicon Valley resonance.
Beyond the product, Lu has been building in public as a thinker. He published a book called Enhanced Human, exploring where technology and humanity collide. He was named to both the Top 10 Pioneering CEOs and Top 50 CEOs of AI Companies lists. He speaks English, Chinese, and Korean - a detail that isn't incidental given the platform's focus on multilingual communication. He has led multiple product teams from research to production across Apple, Google, and now AKOOL, but it's the third act that will define his career.
Where Lu is looking next: proprietary video foundation models and voice-generation capabilities that push avatar realism further. The competitive bar he's setting isn't against other startups. It's against Google, OpenAI, and Adobe - companies with armies of engineers and billions in resources. Lu's argument is that live video streaming with minimal latency is where AKOOL already surpasses them. He intends to stay ahead.
The conventional story of a technical founder would put the PhD at the beginning as proof of intelligence, the big-company jobs as credentialing, and the startup as the brave leap. Lu's story runs a little differently. The PhD was the foundation, yes, but every corporate stop taught him something precise - Face ID at Apple, video pipeline architecture at Google Cloud, what happens when you give a corporation a clear view of what's coming and it still slows down. The side project built on 8 GPUs was never really a bet. It was an inevitability waiting for the right moment.
Coca-Cola saw the technology first. The Inc. 5000 saw the growth rate. The rest of the industry is starting to catch up to what Jeff Lu already knew in 2020.