Y CombinatorOpen Models Change The Economics of AI
At a glance
WHAT IT’S REALLY ABOUT
Open AI models drive cheaper tokens, faster tools, enterprise adoption
- Ollama is seeing a pronounced enterprise shift toward open models, driven first by coding agents and then by long-running “co-worker” agents that dramatically increase per-user token consumption.
- Cost is the immediate adoption catalyst, but enterprises’ longer-term goal is control: the ability to customize, govern, and run models in secure environments aligned with internal requirements.
- Open-model usage is bifurcating by environment: cloud usage on Ollama skews heavily toward Chinese-origin models, while local usage is a more even mix of US/European/Chinese models.
- The operational challenge is no longer just model quality; it’s day-zero launch readiness across inference engines, harnesses/tool-calling, capacity, and hardware optimization—an “OS-layer” integration problem Ollama aims to standardize.
- The likely steady state is hybrid: most tokens inside companies flow through open models (80–90%), while a smaller share of spend remains with frontier closed models reserved for the hardest tasks and orchestration.
IDEAS WORTH REMEMBERING
5 ideasOpen models are winning enterprise volume primarily on cost, then control.
Jeffrey Morgan argues cost is the biggest near-term pain open models solve, but the “North Star” for businesses is control—customization, governance, and deployment choices that closed labs can’t always provide.
Agents—not chat—are the main driver of the token demand surge.
Ollama’s per-developer token usage spiked first with coding agents and later with “co-worker” agent workflows (e.g., OpenClaw/Hermes) that run longer, use tools, and exploit larger context windows.
The open-model release cadence makes fine-tuning harder, but tooling is catching up.
Rapid iteration (e.g., multiple DeepSeek Flash releases in a summer) can “stomp” bespoke fine-tunes, yet improving post-training tooling is making it feasible for teams that truly need specialization.
Open models can outperform closed models in security testing because they refuse less.
For pen-testing and security research, closed models may block requests; open and specialized “security researcher” variants can be more usable, creating a clear niche where openness is a functional advantage.
The biggest platform value is integration across many hidden layers.
Running models well requires coordination across inference engines, harnesses/SDKs, tool-calling mechanics, benchmarks, capacity planning, and hardware drivers—an “OS-like” combinatorial integration problem Ollama targets with a standardized runtime.
WORDS WORTH SAVING
5 quotesCost is by far the largest pain point that open models can jump in and solve. But, you know, every business has a vision of getting better control over AI and customizing it for their business, and that's really their North Star.
— Jeffrey Morgan
This is on an individual user ba- user basis, how many tokens are they using a week?
— Jeffrey Morgan
As a whole through Ollama's cloud, we saw 150x since the start of the year.
— Jeffrey Morgan
The super majority of tokens, and this is our take, it will be open models within a business. Call it 80, 90%. That doesn't mean 80, 90% of the, the budget will go to open models.
— Jeffrey Morgan
And there's this concept that if you're a layer on top of something else, that you're in kind of a vulnerable position as a startup Which is absolutely not true in the AI world.
— Jeffrey Morgan
High quality AI-generated summary created from speaker-labeled transcript.
