In “The Dark Knight,” there is a scene in which the Joker burns a huge pile of money. Today, deciding to develop your own frontier AI model could be a similar exercise to burning cash.
Taylor called the Training New AI model “a good way to burn millions of dollars” because he worked for three companies that trained LLMS, including Google, Facebook and Openai.
In a recent episode of The Minus One podcast, Taylor advised AI founders to build services and use cases, but there was no complete new frontier model.
“Unless you’re working for Openai, Anthropic, Google or Meta, you probably haven’t built one of them,” said Taylor, who also co-founded Sierra AI. “It requires so much capital, so it tends to go towards integration.”
That high capital has stopped the formation of an “indie data center market,” Taylor said.
Taylor advised that the founder would work with AI Juggernauts instead. This is noteworthy that AI Giants like Openai, who is the chairman, will benefit directly. Openai sells “tokens” and accesses the API. APIs can be embedded by developers into applications and programs.
The US LLM market remains largely integrated, but international players are testing Taylor’s theory. In January, Deepseek released an inference model for R1 and a corresponding chatbot. Deepseek used fewer and less advanced chips to build LLM to minimize capital costs.
The Chinese AI app was shot No. 1 on the App Store chart, surpassing ChatGpt, sparking discussions on Tech and Wall Street, and is about whether the tech giants are spending it on AI models.
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
In his podcast interview, Taylor laid out other paths that entrepreneurs could take in the AI market rather than training new models. One was “AI Tools Market.”
“This is the collarbone of the Gold Rush proverb,” Taylor said. “I think there are a lot of things that foundation model providers are scratching today’s itch that they might do tomorrow, so it’s a dangerous space.”
Entrepreneurs could also try to build what Taylor called an “applied AI company.”
“The 2010 SaaS application will in my opinion be the agent company of 2030,” Taylor said.
But building a model from scratch is a surefire way to “destroy your capital,” Taylor said. He called the handmade model “fast assets,” and cost millions of dollars builders rather than cheap ones.
“There are things you can lease, there are things that are open source,” Taylor said. “Please don’t do it.”
