OpenAI, Anthropic Speed Toward IPOs Amid Growing Scrutiny of Token Payments

America’s leading AI labs are set to spend the second half of this year preparing for initial public offerings that will vault them into instant megacap status. Anthropic, valued at $965 billion in May, is slated to debut as early as October, and OpenAI, valued at $852 billion in March, is likely to follow in 2027.

As listings loom, questions about the nature of the companies’ token-payment business model are getting louder. Palantir CEO Alex Karp blasted the model last week, telling CNBC that “something has gone completely wrong.” Enterprise customers have begun to question the burdensome cost of pay-per-use token consumption and look to cheaper, less sophisticated open-weight models. So can the world’s two great AI labs keep up the pace?

Karp About It

The aggressive adoption of agentic AI in some workplaces this year led to the creation of the slang term tokenmaxxing. That’s what happens when engineers, under pressure to demonstrate they are integrating the new technology but with no clear guidelines, use AI models to excess. Executives have quickly realized this isn’t the most efficient way of doing business. As a result, companies including Uber, Microsoft, Salesforce and Meta have taken steps to ration their employees’ use of advanced AI because the token payment structure preferred by Anthropic and OpenAI has proven more expensive than it’s worth.

Speaking to TBPN, Palantir’s Karp said the excessive use of AI without regard for whether it creates value is “kind of like a porn addiction.” During his CNBC appearance, he said the US AI industry should not dismiss the potential for cheaper open-weight models, especially those in development in China, to close the gap:

• Beijing startup Z.ai’s GLM-5.2 model is now ranked among the top 10 large language models by Artificial Analysis, and is ranked as the second-best model for web development by AI evaluation platform Code Arena, placing it alongside Anthropic, OpenAI and Google. The open-weight model is also four to six times cheaper than frontier AI.
• Some US and international enterprise customers have already reported switching to cheaper Chinese models like DeepSeek and cutting back on payments to OpenAI and Anthropic. 

Raising the Stakes: The Financial Times reported last week that OpenAI has held talks with the Trump administration about giving the US government a 5% stake. But the paper said its proposal hinges on other US AI labs, like Anthropic, agreeing to do the same. Experts, meanwhile, warn that recent export controls on advanced US AI models may accelerate the international adoption of models developed by Chinese firms. Two respected Silicon Valley financiers who have the president’s ear understand this: Former Trump advisor David Sacks and current Trump advisor Marc Andreessen have both noted GLM-5.2’s power in recent weeks.