Inside AI’s Surge With Fireworks AI CEO Lin Qiao

The AI increase is not slowing down; it is accelerating.

Lin Qiao, a former Meta engineer who helped construct PyTorch, now runs Fireworks AI, a $4 billion startup processing 15 trillion AI tokens a day, and she or he says demand is just simply getting began.

“That is the 12 months token consumption goes to develop exponentially,” Qiao advised me in a current interview.

Fireworks AI’s inference cloud platform is now processing roughly 15 trillion AI tokens per day, up from 13 trillion just some months in the past and 10 trillion in late 2025. (Fashions break down phrases and different inputs into numerical tokens to make them simpler to course of and perceive. One token is about ¾ of a phrase. They’re additionally used to cost AI mannequin use, through an industry-standard value per million tokens.

Qiao has been right here earlier than. Lengthy earlier than the present generative AI increase, she was inside Meta serving to construct PyTorch, the open-source framework that powered the primary wave of contemporary AI adoption. Again then, there have been no GPUs optimized for AI, no mature tooling, and no clear roadmap.

“We needed to construct every little thing from the bottom up,” Qiao stated.

The size of that progress, Qiao stated, displays how rapidly AI is embedding itself into on a regular basis workflows throughout industries.

Token utilization is not confined to tech groups. Qiao described finance departments utilizing AI to automate forecasting, her personal authorized crew constructing inside AI instruments, and even gig staff creating music on demand with generative AI fashions. Her college-age daughter makes use of a number of AI techniques concurrently — one to generate solutions and others to confirm them.

“That is the world we’re residing in,” Qiao stated. “Actually each single particular person is utilizing these instruments.”

That surge is rippling down the complete know-how stack. GPU provide is tight, prices are rising, and even energy infrastructure is beneath pressure as corporations race to deploy extra AI capability.

“The entire system is saturated,” Qiao stated, describing bottlenecks stretching from semiconductor parts to vitality grids.

Her credibility on these traits stems from her function in constructing PyTorch, which helped democratize AI growth throughout corporations starting from Tesla to Walmart. That early publicity confirmed her how rapidly AI may unfold past Silicon Valley into industries like agriculture and manufacturing.

Now, she sees an analogous, however far sooner, wave unfolding.

Why exist?

Nonetheless, a core query hangs over corporations like Fireworks AI: why do they exist in any respect? If hyperscalers Amazon, Google, Microsoft, and Oracle already lease out GPUs, why not go on to them?

Qiao’s reply is complexity and pace. Enterprises, she stated, wrestle to maintain up with quickly altering fashions and {hardware}, from new Nvidia chips arriving each few months to new AI fashions each few weeks. Fireworks handles that churn — optimizing efficiency, managing infrastructure, and serving to clients migrate rapidly — so they do not need to.

For Qiao, the lesson from each PyTorch and Fireworks is constant: as soon as AI turns into usable, adoption accelerates dramatically. And based mostly on present token volumes, that acceleration is simply getting began.

Join BI’s Tech Memo publication here. Attain out to me through e mail at abarr@businessinsider.com.