Seven months ago, Fireworks AI was a $4 billion startup. Now it's in talks to be worth nearly four times that, per Bloomberg.
The company doesn't even build AI models. It just runs them for other firms.
That work has become the part of the AI boom investors want to own.
Fireworks helps firms put AI models to work. The customer doesn't have to build or run the back-end.
That work is called inference. It's where the bills pile up at scale.
Every time an app pings a model, the meter runs. Most apps ping a lot.
A team of ex-Meta engineers started the company in 2022. CEO Lin Qiao came from Meta too.
The firm runs open-source models like Llama for paying customers. Speed and cost are the main pitch.
Its clients include Cursor, the AI coding tool.
Cursor has been one of the fastest-growing startups of the year. Strong customers help Fireworks win bigger checks.
Index Ventures co-led part of the last round in October. It's set to co-lead this one too.
That round also brought in Lightspeed, Evantic, Sequoia, NVIDIA, and AMD. The new deal hasn't closed yet, so the price could still move.
We break down the AI money trail every weekday in Market Briefs - five minutes a morning, with a free 45-minute investing masterclass when you sign up.
Fireworks isn't the only one cashing in. Baseten hit a $5 billion mark in January.
Features & Labels, known as Fal, hit $4.5 billion in December. Each of those rounds tells you the same thing.
Capital is flowing into the layer that runs models. Not just the layer that trains them.
The reason is simple. Training a model is a one-time cost.
Running it never stops. A $4 billion price tag last fall already looked rich.
A $15 billion mark in May suggests the AI memory and compute layer is now priced like a utility.
One that every AI app pays every time it answers.
Nothing is final until the round closes. The price still tells you where AI money flows.
Not into another foundation model. Into the picks and shovels that run them.
That part of the stack is starting to look like its own AI cooling play gold rush.
The next signs to watch are simple. Does the round close at $15 billion or higher?
Do other inference firms raise at fresh marks too? And do the AI labs themselves start to push back?
OpenAI and Anthropic both run their own inference. The bigger they get, the more it costs.
That cost is what makes a Fireworks valuable. Run the models cheaper, and you take a cut of every query.
Inference is also a growth tax on every other AI product. Each new feature means more queries.
And each new query means more bills for someone to pay.
If you want this kind of read every morning, join 350,000+ investors reading Market Briefs - a 45-minute investing course is thrown in as a bonus when you join.
