We realized that to scale our models, we needed to transition from pure research to a business model that could sustain the investment required for large AI models.

Our decision to pivot from CAD to coding was influenced by our personal interests and the realization that the existing science for 3D modeling wasn't ready for effective AI applications.

The best companies are those that continuously reinvent themselves, as seen with Nvidia's shift from gaming to AI, which has driven its current success.

We decided to stop working on our initial business model and pivot to something else, which involved a lot of difficult decisions, including letting go of our team.

The TPU program was initiated to meet the demand for high compute power for inference and training in AI applications.

Our business evolved to include AI-based applications and agentic workflows for enterprises and government customers, moving beyond just data production.

We realized that the existing AI models were not sufficient for our needs, prompting us to develop our own models to enhance product quality.

In the early 2010s, many software businesses transitioned to the cloud, creating numerous opportunities for founders to replace outdated software. We are currently experiencing a similar moment with AI, where existing software can be rebuilt from scratch to integrate AI effectively.

At Fireworks, we manage the complexity of AI application scaling by utilizing a virtual cloud infrastructure that ensures high quality and reliability in GPU selection.