The CEO of a major AI infrastructure company has stated that the consumption of AI tokens is set to grow exponentially this year, as the technology embeds itself into everyday workflows across all industries. Lin Qiao, founder and CEO of the $4 billion startup Fireworks AI, revealed her company's platform is now processing roughly 15 trillion AI tokens daily, a significant increase from 10 trillion in late 2025.
Qiao, a former Meta engineer who helped build the foundational AI framework PyTorch, made the comments in a recent interview. She described a world where "literally every single person is using these tools," from finance departments and legal teams to gig workers and college students.
Unprecedented Scale and Saturation
The surge in demand is creating bottlenecks across the entire technology stack. Qiao stated that GPU supply is tight, prices are rising, and power infrastructure is under strain as companies race to deploy more AI capacity. "The whole system is saturated," she said, describing constraints from semiconductor components to energy grids.
Fireworks AI's own metrics illustrate this acceleration. The company's inference cloud platform processed 13 trillion tokens daily just a few months ago, up from 10 trillion in late 2025. Tokens are numerical units used by AI models to process language; one token is roughly equivalent to three-quarters of a word.
From PyTorch to the Current Boom
Qiao's perspective is informed by her prior role at Meta, where she helped develop PyTorch, the open-source framework that powered the first wave of modern AI adoption at companies like Tesla and Walmart. "We had to build everything from the ground up," she recalled of that earlier era, which lacked optimized GPUs and mature tooling.
She sees the current generative AI wave unfolding in a similar pattern but at a far faster pace. Her experience showed how quickly AI could spread beyond Silicon Valley into sectors like agriculture and manufacturing, a trend she observes accelerating now.
The Enterprise Challenge and Fireworks AI's Role
A core question for infrastructure providers like Fireworks AI is their necessity alongside cloud hyperscalers like Amazon, Google, and Microsoft. Qiao argues that enterprises struggle with the rapid churn of models and hardware, citing new Nvidia chips every few months and new AI models every few weeks.
Her company's value proposition, she says, is managing this complexity—optimising performance, handling infrastructure, and facilitating quick migration—so clients can focus on application rather than operational overhead.
For Qiao, the consistent lesson from both the PyTorch era and today is clear: once AI becomes genuinely usable, adoption accelerates dramatically. With current token volumes continuing to climb, all evidence suggests that this acceleration is only in its initial stages.