Developers Switch to Open-Weight Models to Cut AI Costs
Rising expenses for proprietary AI tools are pushing software firms toward open-weight alternatives, despite higher infrastructure demands and privacy concerns.
Key points
- Software firms are switching to open-weight AI models to lower costs and maintain data control, according to a Bloomberg report.
- AT&T cut AI task costs by 56% with minimal performance loss, aiming to raise open-source usage from 40% to 70%.
- Open-weight adoption faces trade-offs including high infrastructure costs, specialized talent needs, and privacy concerns with Chinese models.
Software companies are increasingly abandoning expensive proprietary AI models in favor of open-weight alternatives. This shift is driven by the need to control costs and retain data sovereignty, according to a recent report by Bloomberg.
As reported by PYMNTS.com, the move reflects a broader industry trend where firms seek to reduce reliance on external vendors. By using open models, developers can customize tools with their own data, avoiding the risk of outsourcing critical technology to third-party providers.
Higher upfront costs challenge the strategy
However, the transition is not without significant hurdles. Building a custom model from open weights requires substantial upfront investment in specialized talent, computer infrastructure, and data management. For companies lacking sufficient proprietary data, this approach often proves less effective than expected.
There is also a notable catch regarding source origin. When firms utilize Chinese open-weight models, some clients express concern over data privacy. This creates a complex trade-off where cost savings must be weighed against potential security and compliance risks associated with foreign-developed software.
AT&T achieves significant cost reductions
Despite these challenges, some organizations are seeing tangible benefits. AT&T reported reducing costs for coding and advanced AI tasks by up to 56% by routing queries to cheaper models when appropriate. The company noted that this strategy resulted in only a 2% decline in performance quality.
The telecommunications giant plans to increase the share of employee queries powered by open-source models from the current 40% to between 60% and 70% in coming years. This indicates a long-term commitment to hybrid strategies that balance capability with efficiency.
Billing models drive the financial shift
The rising costs of proprietary models are largely attributed to the shift from simple chatbots to complex agents that consume more computing power. Additionally, AI labs have moved from flat subscriptions to token-based billing, which can lead to unpredictable expenses for high-volume users.
Chinese laboratories are able to charge less than their U.S. counterparts due to more efficient model architectures and lower energy costs in the region. This economic advantage makes open-weight options from these labs particularly attractive to cost-conscious software developers seeking to mitigate financial pressure.
