The hottest AI company may not be the one building the next celebrity chatbot.

It may be the company charging everyone else to run theirs.

Fireworks AI, the California-based infrastructure startup backed by Nvidia, is reportedly in early discussions to raise fresh capital at a valuation between $30 billion and $40 billion. The talks remain preliminary, the target may change, and no proposed round size has been disclosed. Fireworks AI declined to comment on the reported discussions. [

At the top of the range, the company would be worth ten times its $4 billion valuation from October 2025. More strikingly, it would be valued at more than twice the $17.5 billion price attached to its $1.505 billion Series D in July 2026.

That is not normal startup acceleration.

It is a stress test for the AI infrastructure boom.

The business behind the models

Founded in 2022 by engineers with backgrounds at Meta and Google, Fireworks AI concentrates on the less visible layer of artificial intelligence: helping companies run, orchestrate and train open models.

Its own website pitches a platform for businesses that want to use open models, route work between them and train systems on proprietary information. The company says its platform is trusted by more than 10,000 companies globally and operates across more than 30 regions. Those are company-published claims, rather than independently audited figures.

The investment argument is straightforward. Businesses may not want to build foundational models from scratch, but they still need infrastructure to customize and operate AI efficiently. That creates an opening for specialist providers sitting between model developers, expensive computing hardware and enterprise applications.

In July, Fireworks AI said it had passed $1 billion in annualized revenue. It also reported processing more than 40 trillion tokens per day, up from 15 trillion around its previous financing, with roughly 95 percent coming from models customized using customer data.

If those figures hold, the company is not selling only a vision. It is selling an enormous volume of computation.

But a valuation of $30 billion to $40 billion would still place a formidable price on that growth. Based on the last disclosed $1 billion annualized-revenue figure, the reported range would equal approximately 30 to 40 times that revenue run rate. The true multiple could be lower if revenue has increased substantially since July, but Fireworks AI has not publicly disclosed a newer figure in the reviewed sources.

Why investors keep paying more

The market surrounding Fireworks AI is expanding at extraordinary speed.

IDC reported that global AI infrastructure spending reached $89.7 billion in the first quarter of 2026, an increase of 33.1 percent from the previous year. It raised its forecast for full-year 2026 spending to $497 billion and projects the market will reach $1.08 trillion in 2029.

That spending wave is creating opportunities well beyond chips and data centers. Companies also need inference platforms, orchestration tools, data pipelines and systems that can control the cost of running models at scale.

This is where the valuation story becomes provocative.

Investors are not merely betting that enterprises will use more AI. They are betting that the layer managing and serving those models can become one of the most valuable positions in the technology stack.

The uncomfortable counterweight

The reported financing talks follow an October security incident that exposed API keys and tokens in some customer accounts, according to Tech Funding News.

That detail matters because Fireworks AI is selling infrastructure to companies that may connect models to sensitive data, internal systems and business workflows. In that market, security is not a secondary feature. It is part of the product.

The episode does not erase the company's reported growth, but it sharpens the central question surrounding a possible $40 billion valuation:

Is the market pricing a durable infrastructure champion, or racing ahead of the operational risks that come with scaling this quickly?

The latest valuation remains a target under discussion, not a completed financing.

Still, the message from the market is becoming clear. The next AI fortune may not belong only to whoever builds the smartest model.

It may belong to whoever runs everybody else’s models fastest, cheapest and most reliably.

‍