The AI model race may not be as concentrated as it looks. DigitalOcean Holdings, Inc. Common Stock (NYSE:DOCN) CEO Paddy Srinivasan told Benzinga in an exclusive email interview that frontier models are typically needed for only about 25% of the workloads his customers run, with open-weight models handling much of the rest.
“We believe in: right model, right cost, for every task,” Srinivasan said.
He described a typical 25/75 split between closed and open-weight models, suggesting that AI companies don’t necessarily need to send every task to the most powerful — and most expensive — models from companies such as OpenAI and Anthropic.
That could have major implications for how the AI market develops.
Frontier Models May Only Need 25% of the Job
Frontier models are generally reserved for “the hardest reasoning or specialized use cases,” Srinivasan said. That’s where their additional capabilities justify the higher cost. But those represent only about a quarter of the workload his company sees.
For the remaining 75%, cheaper open-weight models can often provide sufficient performance at a lower cost.
Instead of choosing one model provider and sending everything through it, AI builders can increasingly decide which model makes the most sense for each individual task.
“Most AI Native companies today are already multi-model,” Srinivasan said. “They all have a mixture of models and route specific prompts to the right model.”
AI Builders Want the Best Intelligence Per Dollar
That model-shopping approach is becoming possible as the number of capable AI systems expands.
Srinivasan said DigitalOcean’s inference engine can dynamically select models based on “performance, latency, cost, and preference,” allowing customers to match different models to different workloads.
The goal, he said, is to deliver the “best intelligence per dollar.” That phrase gets to the heart of the shift.
The AI industry’s first phase was largely about building increasingly powerful models. The next phase could be about figuring out which model should handle which task — and at what price.
For OpenAI and Anthropic, that doesn’t necessarily mean losing 75% of the market. Frontier models could remain essential for the hardest and most valuable workloads, while open-weight models expand overall AI usage.
But it does challenge the idea that every AI application needs the most expensive model available.
And that could make the infrastructure layer connecting businesses to multiple models increasingly important.
For investors, the bigger takeaway is that AI may not become a winner-take-all model market. It could become a routing market, where the winners are the companies that help customers get the right intelligence at the right price.
Image courtesy of DigitalOcean Holdings Inc
