The real AI profits won't come from the model makers — they'll flow to the companies that make AI boring enough for procurement departments to buy.
The Summary
- Battery Ventures GP Dharmesh Thakker argues enterprise software with measurable ROI will beat pure AI plays
- Open-source models could crater enterprise AI costs and shift power away from OpenAI and Anthropic
- The winning move: deploy agents that solve specific business problems, not general intelligence
The Signal
Thakker is making the case that matters most for anyone building or buying AI: the companies that win won't be the ones with the smartest models. They'll be the ones that can walk into a CFO's office and show a 3-month payback period.
This is the enterprise software playbook that built Oracle, Salesforce, and ServiceNow. Find a painful, expensive workflow. Automate it with software. Measure the savings. Scale the deployment. Battery Ventures is betting that formula still works, even when the software runs on transformers instead of SQL databases.
"Enterprise software companies that deliver measurable business value will outperform."
The open-source angle is where this gets interesting for the model layer. If Llama 4, Mistral, or whatever Meta ships next can handle 80% of enterprise use cases at 10% of the API cost, why would a CIO lock in with OpenAI? The math changes fast when you're talking about deploying agents across thousands of seats. A $0.002 per call difference compounds into millions when you're automating customer service, legal doc review, or procurement workflows.
Anthropic and OpenAI have been selling on capability. Open-source models are selling on cost and control. Most enterprise AI deployments don't need frontier intelligence. They need reliable, cheap inference that works behind the firewall. That's why we're seeing companies like Databricks, Snowflake, and enterprise clouds bundle open models into their platforms. The infrastructure layer is absorbing the model layer.
The Implication
If you're building in this space, ask yourself: what procurement problem are you solving? "AI-powered" isn't an answer. "Cuts invoice processing time by 60%" is. The companies that figure out how to package agents as boring business tools with clear ROI metrics will take more market share than the ones chasing AGI.
Watch for the gap between model capability and enterprise deployment to keep widening. The models are ready. The companies buying them are just starting to figure out what to do with them. That's where the next $100B in enterprise value gets built.