Software Firms Shift to Open-Weight AI Models to Cut Costs

Cost pressures reshape AI model adoption

Software companies are increasingly adopting open-weight artificial intelligence models as an alternative to expensive proprietary systems from major U.S. AI labs, according to a Bloomberg report published Monday. The shift reflects growing concern over the financial burden of closed AI platforms, particularly as usage-based pricing models replace flat subscription fees.

Open-weight models—which allow companies to access model parameters and build customized solutions with their own data—offer two primary advantages: lower operational costs and reduced dependency on external vendors. By maintaining control over their AI infrastructure, firms can minimize the risk of outsourcing critical technology to third parties.

Why it matters

The economics of enterprise AI are forcing a strategic choice between convenience and cost control. As AI capabilities become essential to software products, the decision between proprietary and open models will shape competitive positioning, vendor relationships, and data governance practices across the industry. Companies that successfully navigate the infrastructure complexity of open models may gain significant cost advantages over competitors locked into expensive proprietary platforms.

Implementation challenges surface

Despite the appeal of open-weight models, implementation proves difficult for many organizations. Building proprietary models requires substantial upfront investment in computing infrastructure, specialized talent, and data resources. Companies lacking sufficient proprietary data find the economics particularly challenging.

Chinese open-weight models have emerged as cost-effective options, benefiting from more efficient architectures and lower energy costs in China. However, client concerns about data privacy have limited adoption of these alternatives in some markets.

Notably, even firms that develop their own models continue relying on major AI labs when tasks demand cutting-edge capabilities, suggesting a hybrid approach may be optimal.

Real-world cost reductions

AT&T provides a concrete example of successful cost management through strategic model selection. The telecommunications company reduced expenses for coding and advanced AI tasks by up to 56% by routing employee queries to less expensive models when appropriate. Performance quality declined by only 2% under this approach.

The company currently powers 40% of employee queries with open-source models and plans to increase that share to 60-70% in coming years, demonstrating confidence in the viability of cost-optimized AI strategies.

Pricing dynamics drive change

Several factors have accelerated cost pressures. The evolution from simple chatbots to autonomous agents has increased computing requirements substantially. Simultaneously, major AI labs have shifted from predictable flat subscriptions to token-based billing, making costs less predictable and potentially higher for intensive users.

For middle-market CFOs, the calculation involves weighing the savings and control of open models against the operational burden of self-hosting, which requires computing capacity, storage, cybersecurity controls, monitoring tools, and skilled personnel—responsibilities typically bundled into proprietary platform pricing.

These details were first reported by Bloomberg.