TL;DR — Key Takeaways
– Enterprise AI spending continues to grow: 82% of enterprises expect to increase AI spending, while 24% now allocate at least 20% of their IT budgets to AI applications.
– AI investments are delivering measurable returns: 52% of companies report positive ROI from AI expenditures, although most gains remain relatively modest.
– Cost management is becoming a priority: 59% of enterprises use token consumption caps to manage inference costs, while only 13% operate without active cost-control strategies.
Enterprise commitment to artificial intelligence (AI) shows no signs of abating, with tech budgets expanding, financial returns ticking upward, and organizations increasingly taking model development into their own hands, according to new research from ETR.
A startling 82% of enterprises expect to increase their AI spending, holding steady with March levels. Nearly a quarter of organizations (24%) now dedicate at least 20% of their total IT budgets specifically to AI applications.
Capital allocation is paying off: 52% of companies report positive return on investment from their AI expenditures, up 4% from March, while the share of firms seeing zero return dropped to 24%. Most financial gains remain moderate, with 40% of companies reporting returns between 1% and 25%.
As deployment scales, enterprises are tightening cost controls.
Token consumption caps have emerged as the primary mechanism to manage inference costs, adopted by 59% of respondents (up 9% since March). Model routing followed at 34%, while techniques like caching, quantization, and deployment of smaller models each saw adoption rates between 31% and 35%. Only 13% of companies now operate without active cost-management strategies.
Internal tools dominate production environments.
Employee support applications lead all deployment categories at 46%, followed closely by enterprise search (42%) and data analytics (40%). External applications lag, led by customer engagement chatbots at 38%.
Internal maturity is restructuring the vendor ecosystem. Enterprise reliance on third-party consultants dropped 8% to 28% as organizations build internal capacity. Seventy percent of firms are upskilling existing developers, and 52% are hiring specialized talent, though 49% still cite skill shortages as a major hurdle.
A multi-model strategy has rapidly become the enterprise baseline. Six in 10 companies now run multiple model providers in production, while single-provider setups dropped to just 10%.
In the battle for developer mindshare, Anthropic’s Claude has surged past OpenAI’s GPT for application development. Claude usage jumped to 69% (up from 48% in March), outpacing GPT’s 62%. The preference for Claude is particularly pronounced among technical practitioners — analysts, architects, and engineers — where Claude holds a 14-point advantage over GPT, double the gap seen at the executive level.
In model selection, open-weight options are gaining traction, with 26% considering them an important decision factor. Google’s Gemma leads the open-weight category with 16% adoption, ahead of Meta Platforms Inc.’s Llama, Microsoft Corp.’s Phi, and Mistral at 9% each.
Microsoft continues to anchor enterprise infrastructure, controlling top spots across development platforms, cloud hosting (64% vs. AWS at 54%), vector databases (36%), identity management via Entra (47%), and observability via Azure Monitor (44%).
Risk management frameworks are also evolving to meet new technical challenges. While data privacy (75%) and security vulnerabilities (67%) remain the top concerns, unauthorized access by autonomous AI agents registered the largest surge, with 52% of enterprises now formally monitoring agent risk, up 9% since March.
