Nearly 60% of senior executives at large companies plan to ramp up use of open-weight AI models over the next year to rein in ballooning AI budgets, according to a Capgemini survey released Thursday. The shift follows reports of AI spending overshooting expectations by 24% on average in the past year.
TL;DR
- 59% of senior executives at large firms plan to increase open-weight AI adoption to control costs, per Capgemini survey
- Open-weight models are on average 87% cheaper than closed alternatives, yet handled 72% of traffic on OpenRouter in September
- Half of respondents will still use closed models for complex tasks despite cost concerns
What happened
A Capgemini survey of 500 executives at companies with over $1 billion in annual revenue found 59% plan to increase their use of open-weight AI models in the next year to manage rising costs. The shift comes as 59% of respondents reported AI spending exceeded budgets by an average of 24% over the past year.
Open-weight models—like those from DeepSeek or Mistral—can be downloaded and run locally, while closed models such as Claude, ChatGPT, or Gemini are cloud-only. Open-weight models handled 72% of traffic on OpenRouter in September, though they accounted for just 27% of total AI spending, according to Capgemini.
The cost gap is stark: open-weight tokens are on average 87% cheaper than closed ones, making them a clear target for budget-conscious enterprises. Half of the surveyed executives said they would still retain closed models for tasks where complexity justified the higher expense.
The trend mirrors moves by SAP, whose finance chief announced last month the company was replacing some costly AI models with cheaper alternatives where performance remained comparable.
Why it matters
For AI startups and developers, the shift signals a growing preference for open-weight models, which could drive demand for efficient, deployable AI systems that run on-premise or in hybrid environments. This may pressure closed-model providers to justify their pricing or risk losing enterprise customers to more cost-effective alternatives.
Investors should note that while open-weight models offer significant cost savings, their adoption may be uneven—half of enterprises still see value in closed models for complex, high-stakes tasks. This dual approach could create opportunities for startups offering specialized open-weight solutions or hybrid deployment tools.
The survey underscores a broader industry reckoning with AI cost management, where budget overruns are forcing companies to rethink their AI strategies. Startups that can demonstrate clear ROI or cost advantages in open-weight deployments will likely gain traction in enterprise sales cycles.
The findings also highlight the role of multi-AI providers like OpenRouter in shaping spending patterns. As open-weight models capture more traffic but less spending share, providers may need to adjust pricing or packaging to align with enterprise priorities.
Key facts
- 59% of senior executives at large firms plan to increase open-weight AI adoption to control costs, per Capgemini survey
- AI spending exceeded budgets by an average of 24% over the past year, according to 59% of respondents
- Open-weight models handled 72% of traffic on OpenRouter in September but accounted for just 27% of total AI spending
- Open-weight tokens are on average 87% cheaper than closed models, per Capgemini
- Half of respondents said they would retain closed models for complex tasks where performance justified the cost
- Capgemini surveyed 500 executives at companies with over $1 billion in annual revenue across 10 countries
- SAP's finance chief announced last month the company was replacing some costly AI models with cheaper alternatives
- Anthropic warned in its IPO prospectus that customers may shift to cheaper options, including open-weight models
Context
The push toward open-weight models reflects a broader industry shift as enterprises grapple with the high costs of AI adoption. Closed models, while convenient, often come with premium pricing tied to cloud compute and proprietary APIs, which can inflate budgets quickly.
Open-weight models offer greater flexibility and control, allowing companies to run AI locally or in private cloud environments. This aligns with growing enterprise concerns about data privacy, latency, and vendor lock-in, particularly in regulated industries like finance and healthcare.
The trend also mirrors the rise of open-source AI ecosystems, where models like Mistral's Mixtral or DeepSeek's V3 are gaining traction for their performance and cost efficiency. Startups and incumbents alike are increasingly building on these open-weight foundations to differentiate their offerings.
