Enterprise AI Budgets Shift to Proprietary Models Despite Open-Source Hype

The narrative surrounding open-source artificial intelligence often emphasizes democratization and cost savings, but enterprise purchasing patterns tell a markedly different story. According to recent survey data, businesses are consolidating their AI investments around proprietary platforms rather than diversifying toward freely available alternatives.

The Spending Gap Widens

The latest a16z CIO survey polled 100 verified Global 2000 technology executives about their AI deployment strategies. The findings reveal a decisive shift: open-source AI accounted for just 11% of enterprise spending this year, down from 19% the previous year. Closed models captured the remaining 89%, up from 81%.

David Sacks highlighted these figures on the All-In podcast, arguing that metrics like GitHub stars and model downloads obscure the more consequential question of where enterprises actually allocate budgets. The survey data supports his position. As of January 2026, 36% of respondents preferred closed-source models compared to 30% favoring open-source options. Average annual spending on large language models reached $7 million, up from $4.5 million two years prior, with 65% of respondents expecting further increases in 2026.

The beneficiaries of this spending concentration include OpenAI, Anthropic, and Google, according to 247wallst.com, which first reported these details.

Why Lower Token Costs Don't Tell the Full Story

Open-source advocates point to a legitimate advantage: inference costs that run five to 20 times lower than proprietary alternatives. In theory, this price differential should enable open models to process substantially more tokens even while generating less revenue.

The challenge lies in distinguishing between token volume and business value. Enterprises appear to reserve closed models for production systems where reliability and performance matter most—applications involving AI agents, extended conversation histories, large context windows, and custom integrations. These workloads often consume more tokens per task due to repeated interactions and sophisticated reasoning requirements.

Once deployed, these production systems create switching costs that favor incumbents. The operational inertia resembles patterns seen in previous enterprise software cycles, where deeply embedded platforms prove difficult to replace regardless of theoretical alternatives.

Why It Matters

This spending pattern has direct implications for technology investors evaluating the AI landscape. While open-source models will likely remain important for development, research, and cost-sensitive deployments, the highest-value enterprise workloads are concentrating around companies offering frontier performance and enterprise-grade support. The gap between popularity metrics and actual revenue capture is widening, suggesting that attention and profitability are increasingly separate conversations in the AI market.

The data was originally reported by 247wallst.com based on the a16z CIO survey and commentary from David Sacks.