Understanding Inference and the Economics of Enterprise AI (Vista Equity Partners)

Summary

• Third-party research estimates a $3 trillion opportunity for software as AI agents take on tasks previously performed by people.p.2
• The most advanced models charge between $1 and $75 per one million tokens, depending on the model and whether tokens are input or output.p.4
• One analysis finds inference costs average 23% of revenue at AI-native B2B companies.p.6
• Bain & Company projects $5 to $7 trillion in cumulative value flowing to AI-enabled software by 2030.p.6
• Global data center demand is projected to grow from 82.3 GW in 2025 to 219 GW by 2030.p.5
• Open-source models can reduce costs by 40 to 90% with proper tuning and guardrails.p.7

Key insights

1. Inference is the variable cost incurred every time a model is used, paid by whoever runs the model.p.3
2. A traditional SaaS company has gross margins of 75 to 80%.p.5
3. Inference workload demand grows 35% annually from 20.9 GW to 93.3 GW.p.5
4. Models vary in cost per token by 10x or more.p.7
5. Vista draws its three key cost levers from agents in production across over 50 portfolio companies.p.7
6. Left unmanaged, new revenue from agentic products is captured by model providers and hyperscalers rather than the software company.p.6