Companies with significant datasets can reduce their AI costs by up to 80% by transitioning to open-weight models, according to Andrew Dudum, CEO of Hims & Hers. In an interview with CNBC Squawkbox, Dudum emphasized that organizations producing large amounts of data should move away from traditional big AI models in favor of these more flexible and cost-effective solutions.
Dudum highlighted his company’s extensive closed-loop dataset of patient interactions as a valuable asset for training AI models. He stated, “I think for companies that have the resources and scale, if they have an independent dataset, that is a path that they will go, no question. The cost can be upward of 70, 80% less.” This significant reduction in expenses can enhance operational efficiency for data-rich companies.
Open-weight models allow users to access and modify a model’s trained parameters, making them cheaper to operate at scale. Dudum noted that models tailored to a company’s specific use cases tend to yield better performance. He remarked, “And so what we launched, our first version, immediately outsurpassed what we could get in market. But the trajectory of what’s possible in just six months is transformative because it’s continuously learning on every new patient that comes into Hims & Hers.” This adaptability can result in accelerated advancements in AI capabilities.
Dudum’s insights align with a broader trend in the tech industry, where cost-saving measures related to AI are becoming increasingly important. Many companies are moving away from practices like tokenmaxxing, where employees were encouraged to maximize their use of AI tokens, and are instead focusing on optimizing the value of their AI expenditures. The strategy of model routing is also gaining traction, as firms match AI models to tasks based on their complexity. Additionally, companies like Coinbase are exploring the use of alternative models, such as the Chinese Kimi K2.7 and GLM 5.2, as standard options.
Data has emerged as a critical commodity in the AI landscape, essential for effective model training. Recently, Google invested $10 million to acquire internal data from the defunct Spirit Airlines, which included a wealth of internal documents and communications. In another example, AI training firm Handshake AI has offered $6 per page for high-quality work documents, up to a maximum of $30,000, as part of its strategy to access more valuable data resources.
Featured image credit
