40 Trillion Tokens a Day (Yes, More Than OpenAI) | Lin Qiao, CEO of Fireworks
Gradient Dissent: Conversations on AI
Specialized intelligence represents a critical shift in AI development, moving away from general-purpose, black-box models toward customized, enterprise-owned solutions. Lin Qiao, founder of Fireworks, argues that proprietary data serves as a company’s primary competitive moat, necessitating continuous, specialized model training rather than reliance on a few dominant frontier labs. With Fireworks processing over 40 trillion tokens daily—95% of which involve customized inference—the industry is transitioning from "token maxing" to "value maxing," where cost efficiency and domain-specific accuracy determine long-term viability. By leveraging open-source models and providing a platform for continuous fine-tuning, companies can maintain control over their intellectual property and unit economics. This approach challenges the current industry duopoly, suggesting that the future of AI lies in millions of specialized models tailored to unique business use cases rather than a handful of universal, general-purpose engines.
Sign in to continue reading, translating and more.
Open full episode in Podwise
