Artificial intelligence has spent the last two years celebrating the rise of open models, with many businesses viewing them as a lower-cost alternative to proprietary systems from OpenAI, Anthropic and Alphabet Inc‘s (NASDAQ:GOOGL) (NASDAQ:GOOG) Google. Nvidia Corp (NASDAQ:NVDA) CEO Jensen Huang thinks that assumption is wrong.
Huang argues that companies often underestimate the hidden costs of deploying open models, going as far as to say that proprietary alternatives can actually be the cheaper option.
Speaking in an interview with Bloomberg Television during a visit to South Korea, Huang challenged one of the most common beliefs in enterprise AI. “Frankly, I think closed models are cheaper.”
While companies often focus on licensing costs, they overlook everything required to operate an open model themselves, he explained.
“If you don’t have to build yourself, if you don’t have the training yourself cost a lot of expertise to fine tune and maintain and guardrail and keep it safe and evaluated and, and of course, even build computers to host it,” Huang said, describing the engineering and infrastructure burden that comes with self-hosting AI models.
For many enterprises, that means the total cost of ownership can exceed simply paying for access to a hosted proprietary model.
OpenAI Is About Control, Not Cost
While Huang pushed back on the idea that open models are inherently cheaper, he made clear he still sees an important role for them.
“The reason why you need open models is because you need to have control, because you need to adapt something for your own very specialized use cases,” he said.
According to Huang, companies should build their own AI only when they have compelling reasons, such as protecting proprietary knowledge, meeting regulatory requirements, or maintaining technological sovereignty. Otherwise, he suggested, using commercial models available through the cloud is often the simpler approach.
Earlier in the interview, Huang said he encourages businesses—including Nvidia itself—to use services such as OpenAI, Claude, Cursor, Cognition, Perplexity whenever practical because “it’s just easier,” adding that companies should “build only what you must.”
Challenging One Of AI’s Biggest Assumptions
Huang concluded that much of the debate around open versus closed AI has become overly simplistic.
“I think there’s a lot of misunderstanding about closed versus open,” he said.
His comments highlight a shift in how enterprise AI decisions may increasingly be made. Rather than choosing between open and closed models based solely on licensing costs, businesses may need to weigh the full expense of training, hosting, securing and maintaining AI systems over time—a calculation that, according to Nvidia’s CEO, can make proprietary models the more economical choice.
Image via Shutterstock
Recent Comments