The AI boom is not slowing down. It’s accelerating.
Lin Qiao, a former Meta engineer who helped build PyTorch, now runs Fireworks AI, a $4 billion startup that processes 15 trillion AI tokens per day, and she says demand is just getting started.
“This is the year that token consumption will grow exponentially,” Qiao told me in a recent interview.
Fireworks AI’s inference cloud platform currently processes approximately 15 trillion AI tokens per day, increasing from 13 trillion just a few months ago to 10 trillion by late 2025. (Models break down words and other input into numerical tokens to make them easier to process and understand. One token is approximately 3/4 of a word.) It is also used to price the use of AI models through an industry-standard cost per million tokens.
Mr. Qiao has been here before. Long before the current generative AI boom, she was in-house at Meta, helping build PyTorch, the open source framework that powered the first wave of modern AI adoption. At the time, there were no AI-optimized GPUs, mature tools, or a clear roadmap.
“We had to build everything from scratch,” Qiao said.
The scale of that growth reflects how quickly AI is being incorporated into daily workflows across industries, Qiao said.
The use of the token is not limited to technical teams. Qiao discussed how finance departments are using AI to automate forecasting, her own legal team is building in-house AI tools, and even gig workers are using generative AI models to create music on demand. Her college-age daughter uses multiple AI systems simultaneously, one generating answers and one validating them.
“That’s the world we live in,” Chao said. “Literally everyone is using these tools.”
That surge is rippled across the technology stack. As companies rush to deploy AI capabilities, GPU supplies are tight, prices are rising, and even power infrastructure is being strained.
“The whole system is saturated,” Qiao said, describing bottlenecks ranging from semiconductor components to the energy grid.
Her credibility with these trends stems from her role in building PyTorch, which helped democratize AI development across companies from Tesla to Walmart. This early experience showed her how quickly AI could spread beyond Silicon Valley and into industries like agriculture and manufacturing.
Now, she’s seeing a similar, but much faster, wave unfolding.
Why does it exist?
Still, companies like Fireworks AI beg the core question of why they exist in the first place. If hyperscalers Amazon, Google, Microsoft, and Oracle already rent GPUs, why not go directly to them?
Qiao’s answer is complexity and speed. He said companies are struggling to keep up with rapidly changing models and hardware, from new Nvidia chips coming out every few months to new AI models coming out every few weeks. Fireworks addresses such churn by optimizing performance, managing infrastructure, and helping customers migrate quickly. So customers don’t have to.
For Qiao, the lessons learned from both PyTorch and Fireworks are consistent. Once AI is available, adoption will dramatically accelerate. And based on current token volumes, that acceleration is only just beginning.
Sign up for BI’s Tech Memo newsletter here. Please contact us by email. abarr@businessinsider.com.