Fireworks Says Companies Waste Money on AI Without a Portfolio of Models

Companies waste money on artificial intelligence if they rely on a single model instead of using a mix of different options. That warning comes from Rob Ferguson, vice president at the managed inference provider Fireworks AI.

Ferguson said in an interview that buying and running dedicated hardware is a poor financial choice if the equipment sits idle. He noted that ten thousand graphics processing units represent one billion dollars in capital, meaning companies cannot afford to leave them unused. While Fireworks supports on-premises artificial intelligence for sensitive workloads through a service called Sovereign, Ferguson stated that most inference work does not require that level of protection.

Instead, the best approach is to use a portfolio of models and route tasks between them. Fireworks operates a platform that automatically evaluates incoming tasks and directs them to the optimal model based on specific strengths. Ferguson explained that different models have different abilities, such as one model performing slightly better at JavaScript than Python.

The company claims that using a combination of closed and open models can cut a customer's bill in half right away. At a company event, Lin Qiao, CEO and founder of Fireworks, noted that unchecked token spending has made costs out of control for businesses trying to scale. During the same event, Nvidia CEO Jensen Huang said that specialized intelligence trained on specific domain knowledge holds most of the world's value. Huang added that companies can create powerful systems by taking an open frontier model and adapting it with their own data and tools.

Ferguson also cautioned that token cost is not the true expense, pointing out that constant artificial intelligence rework without proper caching drives up actual costs per task. He advised companies to track cost per task and consider the value of their data, warning that many providers retain user data for training without compensating the owners. Along with lowering expenses, Ferguson emphasized that fast response times significantly change how users interact with artificial intelligence.

Summary

Companies spend heavily on hardware and software to run artificial intelligence systems. By showing how automated routing across a mix of models can reduce costs and maximize performance, providers like Fireworks highlight new ways for businesses to manage their technology budgets.