The transition follows a significant financial strain for legal startup Harvey, which saw its gross margins drop from 50% to -50% in the first half of 2024 after a spike in customer usage.
This shift highlights a critical turning point in AI infrastructure as startups prioritize financial sustainability over brand-name partnerships.
While proprietary models like GPT-4 offer high performance, their expensive licensing fees can make scaling difficult for software providers.
By integrating cheaper alternatives, including increasingly capable models from Chinese developers, these startups aim to improve their profit margins while reducing their dependence on a small number of dominant US technology firms.
The use of open-weight models also provides these businesses with greater technical flexibility, as they can customize the AI using their own proprietary data.
This mechanism allows specialized firms, such as those in the legal or financial sectors, to maintain high-quality service while controlling their own capital expenditure.
As the market for generative AI matures, the ability to fine-tune and host independent models is becoming a primary strategy for startups looking to survive the high costs of the current AI boom.