Companies Turn to Open-Weight AI as Spending Overruns Drive Cost Scrutiny

A Capgemini survey indicates that large companies are reassessing how they deploy and pay for artificial intelligence. About 60% of surveyed large companies plan to expand their use of open-weight AI models. At the same time, 59% reported AI spending overruns, averaging 24%. Together, these findings suggest that the next stage of enterprise AI adoption will be shaped not only by model capability, but also by budget control and the ability to match model costs to specific tasks.

The results are relevant to government procurement teams and federal contractors as a market signal, not evidence of a government purchasing decision or a change in federal policy. Buyers may increasingly ask vendors to substantiate the cost and performance of proposed AI solutions, including whether lower-cost open-weight models can meet requirements for particular workloads. For contractors, that can mean more detailed scrutiny of licensing, hosting, inference, integration, and ongoing operating expenses—alongside the technical merits of the model itself.

The findings do not imply that organizations will replace every closed model with an open-weight alternative. Instead, buyers may adopt a portfolio approach: using lower-priced models where they provide adequate performance, while retaining higher-cost closed models for tasks where their capabilities justify the additional expense. In procurements, this places a premium on clearly defining task requirements and demonstrating how model selection affects outcomes, cost, security, and operational support. Proposals that assume one model type is the best fit for every use case may be less persuasive than those that explain practical alternatives.

For companies selling AI platforms, models, or integration services—including SAP, Anthropic, DeepSeek, Mistral, and OpenRouter, which are named in the source data—the survey points to a more price-conscious buying environment. It does not establish that any of these vendors participated in the survey or hold a government contract. Their relevance to GovCon is the broader competitive pressure: suppliers may need to explain model-choice options, disclose cost drivers, and show where a premium model delivers task-specific value. Contractors preparing bids should be ready to compare alternatives transparently and describe how their architecture can accommodate buyer preferences without overstating savings or capabilities.

• About 60% of large companies plan to expand their use of open-weight AI models, indicating growing interest in lower-cost options.
• 59% reported AI spending overruns, with overruns averaging 24%—a signal of heightened budget scrutiny.
• The survey reflects private-sector findings; it does not report a government buying decision, solicitation, award, or policy change.
• Government buyers may assess AI proposals against defined tasks and total cost rather than treating one model category as suitable for all workloads.
• Contractors should explain pricing assumptions and identify potential model-substitution options, including any effects on performance and operating requirements.
• Higher-cost closed models may remain appropriate when their capabilities provide task-specific value; proposals should make that value explicit.
• AI suppliers may face increased price pressure as buyers consider open-weight alternatives and compare models more closely.
• Flexible architectures and clear cost comparisons may help contractors address buyer questions, but savings and suitability should be supported by workload-specific evidence.
• Companies named in the source data include SAP, Anthropic, DeepSeek, Mistral, and OpenRouter; the source does not identify them as government awardees or survey participants.