Former Google CEO Schmidt Weighs Open Source vs. Closed AI Models: 'I Alternate Every Six Months'
Former Google CEO Eric Schmidt admitted he himself oscillates between which approach will dominate artificial intelligence's future — closed, proprietary systems like those from OpenAI and Anthropic, or open source models that anyone can modify and deploy. "I alternate every six months," Schmidt revealed during a PARC Forum appearance, highlighting a fundamental tension shaping the technology's trajectory.
End of Miles reports that Schmidt's candid admission reflects broader uncertainty among even the most experienced tech leaders about which business model will ultimately prevail in AI development.
The resilience advantage of scale
Schmidt outlined a compelling technical argument favoring large, closed models, noting that when AI systems are reduced in size — a process called distillation — they lose important capabilities.
"When you make a model and you dumb it down, that is you make it smaller, believe it or not, it loses resilience. It becomes more fragile," Schmidt explained. "So strangely, if you're focused on quality, you should try to get the largest model possible." Eric Schmidt, PARC Forum
This technical reality creates a natural advantage for companies with massive computing resources. Yet Schmidt pointed out how the economics might favor a different outcome, as large models remain "too expensive to serve" for many applications.
The regulation question
Beyond technical considerations, Schmidt highlighted how the closed-system approach offers clearer paths for government oversight. With a small number of powerful models housed in identifiable data centers, regulation becomes more straightforward.
"If you believe that the future is a small number of these big models... then we have one set of outcomes. And in that set of outcomes, first, they can be regulated literally. You know where they are. The president can order the troops to take over the data center." Schmidt
The Stanford visiting scholar contrasted this with an open-source future that would be "a very different world" from a governance perspective, potentially undermining attempts at creating AI guardrails.
The open source challenge
Despite these advantages for closed systems, Schmidt noted that powerful open source alternatives continue emerging. He specifically mentioned Deep Seek, though he observed it was "trained by other models" rather than built from scratch.
The tech veteran suggested that these competitive dynamics create an environment where conviction follows employment. "What I found is people who work in the closed source companies believe they're going to win. The people who work in open source companies believe they're going to win," he said.
"But I think a dispassionate view would be to say that we don't know yet." Schmidt
The distillation economics
Schmidt detailed how smaller, specialized models might still find their place through the process of distillation — taking thousands of queries against large models to build focused applications.
"You can build something which is very, very close to the major model for 1/50th of the price," he explained. "It's not as robust, but is it functional for that particular application? And we don't know."
This balanced assessment from someone who has led some of the most consequential technology companies suggests that AI's future architecture remains genuinely uncertain, with profound implications for who controls what may become humanity's most powerful tool.
