Venture capitalist Chamath Palihapitiya has warned that rising artificial intelligence expenses are becoming unsustainable for software startups. During a recent episode of the All-In Podcast, Palihapitiya disclosed that his company, 8090, is spending millions of dollars annually on AI inference and tool usage, with costs having more than tripled since November. He projects annual AI expenditures could reach $10 million, driven primarily by bills from Amazon Web Services, the coding assistant Cursor, and Anthropic. Palihapitiya highlighted a critical economic imbalance where operational costs are increasing at three times the rate of revenue growth. The current business model appears to rely heavily on venture capital subsidies that cover what Palihapitiya described as unlimited token consumption. He expressed concern that once venture funding dries up, the economics of building software with AI will become unviable, drawing parallels to early ride-sharing subsidies that eventually disappeared. Specific tools are drawing scrutiny. Palihapitiya singled out Cursor as a major source of excessive costs compared to alternatives. He announced plans to migrate away from Cursor toward Anthropic's Claude Code, arguing that the latter offers equivalent performance at a lower price point, particularly on its Pro plan. He also identified a common inefficiency in AI usage known as the Ralph loop, a term derived from The Simpsons character Ralph Wiggum. This practice involves repeatedly feeding the same prompt into an AI model in hopes of resolving an issue, which Palihapitiya noted rarely succeeds but significantly inflates bills. These concerns reflect a broader industry trend. Dax Raad, creator of OpenCode, recently noted that financial officers are finally waking up to the fact that engineers now cost an additional $2,000 per month in large language model bills. Palihapitiya emphasized that the solution requires greater flexibility to switch between AI models without disrupting operations. He cited the recent fallout between Anthropic and the U.S. Department of Defense as evidence of why reliance on a single provider is a strategic risk. The situation underscores a growing tension between the rapid adoption of AI tools and the need for economically viable deployment strategies in the tech sector.