Venture capitalist Chamath Palihapitiya said he can’t believe how much his software startup is spending on AI.
“Since November 2025, our costs have more than tripled,” Palihapitiya said on an episode of the “All In Podcast” posted Friday. “If you add up the huge inference costs we pay to AWS, the Cursor costs, and the Anthropic costs, we’re just spending millions of dollars.”
Palihapitiya said 8090, which started with “the goal of replacing/rewriting all the legacy software in the world,” tends to spend $10 million a year on AI costs. His concerns about ballooning bills come as the technology industry continues to understand how AI will upend established fields such as software engineering.
The biggest concern expressed by Palihapitiya is that while AI costs are rising, 8090’s revenue is not increasing at the same pace.
“The problem is that costs are going up three times every three months,” he said. “My income is not like that.”
Palihapitiya told X that the current system is partially subsidized by large venture capital firms, which are the biggest backers of AI companies such as OpenAI and Anthropic. Some tech experts question the sustainability of the current approach, comparing it to the fact that Uber rides start out cheap and then increase in price over time.
“Thank you to the VCs who are funding this all-you-can-eat token consumption through their huge investments,” he wrote.
Palihapitiya cited Cursor, a popular AI coding tool, as responsible for some of the 8090’s biggest AI costs. Anthropic’s “Claude Code” is a much better deal, he said.
“You need to move away from Cursor,” he wrote in X. “The cost is too high compared to Claude Code. The latter is comparable, and the Pro plan eliminates Cursor’s huge charges for token consumption.”
Concerns about soaring AI usage fees continue to arise. OpenCode creator Dax Raad recently said that CFOs are starting to wake up to the cost of AI costs.
“Your CFO is like, what does it mean that each engineer is now paying an extra $2,000 a month in LLM billing,” Raad wrote to X.
Palihapitiya said he suspects some of the 8090 AI claims are due to Ralph Loops, the nickname for a technique used that essentially keeps feeding back the same prompt to an AI model until the problem is resolved. This approach is named after Ralph Wiggum, a character from “The Simpsons.” Ralph Wiggum is a very tenacious and beloved fool.
“Everyone’s obsessed with this thing called the Ralph Wiggum Loop, where you just send it out and figure something out,” Palihapitiya said on the podcast. “A, I don’t know anything, and B, I just get this huge bill from Cursor.”
Palihapitiya said switching between models needs to be more flexible going forward. He said Anthropic’s recent rift with the Department of Defense shows why this is necessary.
“We need more flexibility to swap between models without breaking everything,” he wrote in X. “I think this is a No. 1 cost issue, but also a strategic flexibility issue after what happened between Anthropic and DoW.”
