Open Models Change The Economics of AI

Morgan says enterprise interest is shifting toward open models, driven by coding agents and assistants that handle longer tasks. Cost is the most immediate benefit, while greater control and customization are longer-term goals. He describes token use growing 150 times on Ollama Cloud since the start of the year.

Morgan says faster model releases make custom training harder, though improving tools help teams keep up. For a day-zero launch, Ollama coordinates inference support, useful harnesses, cloud capacity, and hardware so models run reliably and quickly. He compares the work to an operating system that connects applications with hardware and drivers.

Morgan expects most business tokens to pass through open models, while frontier models handle the hardest tasks. He estimates open models could account for 80 to 90 percent of tokens but only 10 to 20 percent of spending. He also expects local models to handle simpler work, with cloud models available when a task needs more capability.

Morgan recounts how he and his co-founder moved from developer security projects toward making open models easier to run. They shipped the first Ollama version in two weeks, around the release of Llama 2, after earlier ideas had not found a strong customer problem. He says the team later saw coding agents create demand for cloud access to open models, while stressing the need to stay close to customers.