We've taken it as understood that if you use these products, you give up your data privacy... What that does is that creates all these artificial barriers to entry for all of these potential users of AI.
While many see DeepSeek as cause for celebration, some are seeing another side of the story. Misconceptions around AI, particularly regarding market demand, infinite computing growth, and sacrificing privacy have all risen to the surface to be reexamined.
We spoke with Nick Reese, COO and Co-founder of Frontier Foundry, about the revelations and misunderstandings that DeepSeek has brought to light:
Market assumptions: Reese referenced Mark Twain's famous quote: "It ain't what you don't know that gets you into trouble. It's what you know for sure that just ain't so," to underscore the risks of entrenched market assumptions.
"When the market holds certain assumptions, that's where trouble starts." Reese explains. "The entire AI industrial complex has been built on the belief that there will be an unending demand for Nvidia chips, compute power, and the infrastructure to support them. DeepSeek has challenged that assumption."
Market fallout: Reese elaborated on how this shift in perspective impacted the stock market:
"That's exactly why we saw a market dip. Investors started questioning, 'Didn’t we just commit $500 billion based on the need for these chips?' But then DeepSeek comes along saying, 'I didn’t even use the top-tier chips. I had the janky ones and still built this.' That realization shook the underlying belief in constant, exponential market growth."
When the market holds certain assumptions, that's where trouble starts. The entire AI industrial complex has been built on the belief that there will be an unending demand for Nvidia chips and the infrastructure to support them. DeepSeek has challenged that assumption.
Privacy limitations: Pointing to biases in DeepSeek and marked concerns about privacy, Reese highlights that most people using AI accept some degree of privacy loss considering that the tools at hand were trained on "entirety of digitized human knowledge."
But Reese sees another path, one where it's possible for privacy to be built into AI.
"We've taken it as understood that if you use these products, you give up your data privacy. We just assume that. What that does is that creates all these artificial barriers to entry for all of these potential users of AI."
Citing regulated entities or companies with privacy compliance laws such as healthcare or law firms in this context, he adds that organizations simply assume they’ll never be able to use AI. "But it doesn't have to be like that," Reese explains. "You can actually build AI again that you can run locally trained just on your data."
Specialists vs. generalists: Reese believes the future of privacy-focused AI lies in specialization rather than generalization:
"There won’t be one monolithic algorithm to rule them all with trillions of parameters. Instead, we employ an ensemble of algorithms, each excelling at specific tasks."
Frontier Foundry, a multimodal agentic AI company, is at the forefront of this shift. They are developing industry-specific AI tools that can operate locally, even without an internet connection:
"We create AI products for organizations with strict data regulations, like those under HIPAA. Our models stay within your environment, never calling out or using your data to train external systems. The result is a truly proprietary algorithm, specialized to meet your unique needs."
