Zerodha co-founder Nikhil Kamath and Coinbase CEO Brian Armstrong have warned that the sky-high valuations of premium AI companies like OpenAI and Anthropic face a massive structural threat even as there is a growing investor skepticism surrounding the artificial intelligence (AI) boom.During an interview, the two prominent business leaders drew direct parallels between the current AI frenzy, the 2000s Dot-Com crash and standard crypto market bubbles. Both agreed that the primary concern is expensive, proprietary AI models that are losing their competitive advantages to cheap open-source alternatives and localized, domestic tech.“Like me, the stock trader investor, I’m starting to feel at this point that if I were to take every private company in AI and short their stock today, in five years, I might make money,” Kamath stated, adding, “It feels a bit like… the ‘Internet bubble’.
‘Every country will have its own model’
Kamath argues that the AI industry will shift from a globalised market dominated by a few American companies to a fragmented, regional economy. He predicts that through reverse-engineering, copying, and rapid development, individual nations will choose self-reliance over expensive imports.“India will have its own copy of the model. Another country will have its own copy. The tokens, the energy, all of that will sit domestically within our country,” Kamath noted. While these domestic variations might not sit at the absolute cutting edge, they will be entirely functional for everyday use. “If the world goes in that direction, I don’t see the reason to pay the multiples that these private companies have today,” he added.
Coinbase CEO says there is ‘99% cheaper’ threat
Coinbase CEO Brian Armstrong agreed with Kamath’s market assessment, pointing out that while top-tier labs spend billions to build the next breakthrough, open-source alternatives trailing just six months behind are hitting the market at a mere fraction of the price.“The open-source models are really like six months behind and they’re like 99% cheaper or more sometimes for inference. So I think it’s entirely possible that a larger percentage of the workload goes to these models that are 99% cheaper,” Armstrong explained.According to Armstrong, while elite frontier models will remain valuable for highly specialized tasks like discovering new physics, average consumers and businesses will become heavily price-sensitive. He said that once standard models become efficient enough to run on cheap, everyday commodity hardware, the corporate defenses protecting high-value AI companies could completely dissolve.“It makes me a little nervous when I see these valuations growing this fast as well. Like I’ve seen things like this happen before in crypto. They correct, and then there’s real value under it, so then they grow later,” Armstrong concluded.