Venture capitalist Chamath Palihapitiya said he can’t believe how much his software startup is spending on AI.
Loading audio narration…
“Our costs have more than tripled since November of 25,” Palihapitiya said during an episode of the “All-In Podcast” posted on Friday. “Between the inference cost that we pay AWS, which is ginormous, between our cost with Cursor, between Anthropic, we are just spending millions.”
Palihapitiya said 8090, which he started with “the goal of replacing/rewriting all the legacy software in the world,” is trending toward spending $10 million a year on AI costs. His concern about ballooning bills comes as the tech industry continues to digest how AI is upending well-established fields like software engineering.
The biggest fear Palihapitiya expressed is that while AI costs are rising, 8090’s revenues are not increasing at the same pace.
“The problem is that my costs are going up 3X every three months,” he said. “My revenues are not.”
The current system, Palihapitiya said on X, is partially subsidized by large venture capital firms that remain some of the biggest backers of AI companies like OpenAI and Anthropic. Some in tech have questioned the sustainability of the current approach, comparing it to how Uber rides were first cheap and then increased in cost over time.
“Thank you to the VCs who will fund this all-you-can-eat token consumption through their huge investments,” he wrote.
Palihapitiya singled out Cursor, a popular AI coding tool, as a source of some of 8090’s biggest AI costs. He said that Anthropic’s Claude Code is a much better bargain.
“We need to migrate off of Cursor,” he wrote on X. “Its just too expensive vs Claude Code. The latter is equivalent and if you use the Pro plan, you eliminate huge Cursor bills for token consumption.”
Concerns about rising AI usage bills are still bubbling up. OpenCode creator Dax Raad recently said that CFOs are beginning to wake up to the cost of AI bills.
“Your CFO is like what do you mean each engineer now costs $2000 extra per month in LLM bills,” Raad wrote on X.
Palihapitiya said that he suspects some of 8090’s AI bills might be due to Ralph loops, the nickname for a usage technique that essentially keeps feeding the same prompt back into an AI model until the problem is solved. The approach is named after “The Simpsons” character Ralph Wiggum, a beloved doofus who can be very persistent.
“Everybody has gotten infatuated with what we call these Ralph Wiggum loops, just like send the thing off and it’ll just go figure something out,” Palihapitiya said during the podcast. “A, It never figures anything out. And B, you just get this ginormous bill from Cursor.”
Going forward, Palihapitiya said there needs to be more flexibility to switch between models. He said that Anthropic’s recent falling out with the Pentagon illustrates why this is needed.
“We need to gain more flexibility to swap between models without everything breaking,” he wrote on X. “I think this is both a cost problem per #1 but its also a strategic flexibility issue after the events between Anthropic and DoW.”
