1Blueprints and Curiosity
Lin Qiao grew up in China, spending hours with her father—a senior mechanical engineer—learning to decipher ship blueprints. Those early lessons in complex systems would shape her entire career. She learned that big things are built from understanding small components.
She studied at Fudan University in Shanghai, then moved to the United States for a PhD in distributed systems and database management at UC Santa Barbara, graduating in 2005. Her thesis work focused on making complex systems work at scale.
"Growing up, I spent hours with my father learning ship blueprints. That taught me how complex systems work—layer by layer, component by component. It's exactly how I think about AI infrastructure today."
— Lin Qiao
2Riding Every Wave
After her PhD, Qiao joined IBM as a software engineer. Then LinkedIn. Then Facebook (now Meta). At each stop, she witnessed technology cycles firsthand—new paradigms emerging, transforming everything, creating new opportunities for those paying attention.
At Meta, she witnessed the most seismic shift yet: the transition from CPUs to GPUs for AI workloads. It was a fundamental change in how computers think. And she was right in the middle of it.
3Building PyTorch
At Meta, Qiao became head of PyTorch—the deep learning framework that now powers most of modern AI research and production. What started as what she thought would be a six-month project turned into five years of rebuilding the entire AI stack from scratch.
Her team grew from 5 people to 300. By the time she left, PyTorch was sustaining more than 5 trillion inferences per day across Meta's entire AI workload. She had built the infrastructure that made modern AI possible.
5 trillion inferences per day. That's 5,000,000,000,000 AI predictions running through the system Lin built—every single day. When researchers and companies build AI today, most of them are building on her work.
4Seeing the Next Wave
In October 2022—just before ChatGPT's launch—Qiao recognized the next opportunity. Companies wanted to prioritize AI but lacked the infrastructure, resources, and talent to deploy it. She'd spent five years solving that problem at Meta. Now she could solve it for everyone else.
Inspired by PyTorch's flame logo, she founded Fireworks AI with a mission: make it simple, fast, and cost-effective for enterprises to build and scale generative AI products. She was building the infrastructure layer for the AI revolution.
The Problem:
Enterprises want AI but can't build the infrastructure. It's too hard, too expensive, requires too much talent.
The Solution:
Fireworks handles the infrastructure so companies can focus on their AI products. One API, infinite scale.
5Powering the Revolution
Fireworks AI now processes 13 trillion tokens per day. Their customers include Uber, Verizon, and DoorDash—companies that need AI at massive scale. The company raised over $300 million and hit a $4 billion valuation.
From February 2024, the platform's user base grew from 12,000 developers to over 23,000. Lin had bet that the AI revolution would need infrastructure— and she was right.
6Key Lessons for Founders
1. Build the picks and shovels
During a gold rush, sell infrastructure. Fireworks doesn't build AI apps— it powers them. The infrastructure layer captures value from the entire ecosystem.
2. Ride multiple waves
IBM, LinkedIn, Meta, Fireworks—Qiao didn't stay in one place. Each jump positioned her for the next big shift.
3. Your big company experience is your edge
Five years building PyTorch at Meta gave Qiao knowledge no startup founder has. Don't dismiss corporate experience—extract its lessons.
4. Time your exit
Qiao left Meta in October 2022, right before ChatGPT launched. She saw the wave forming and positioned herself to ride it.
5. Startups aren't about incremental changes
"Startups aren't about incremental changes," Qiao says. "They're about ten-times leaps." Go big or go home.