The AI industry spent the last few weeks doing something that matters far more than any product launch: it started turning itself into public companies.
Three of the largest stock market listings in history are expected before the end of 2026. SpaceX filed its public prospectus on May 20, targeting a valuation of around $1.75 trillion and a raise of up to $75 billion. OpenAI is preparing its own confidential filing, targeting a September debut at a valuation between $850 billion and $1 trillion. Anthropic is aiming for October at roughly $900 billion, with Goldman Sachs, JPMorgan, and Morgan Stanley reportedly in discussions as underwriters. The combined valuation of all three is somewhere near $3.6 trillion, roughly the size of France's entire economy.
That is not just a financial story. Once these companies are public, they report quarterly to millions of shareholders. Every decision they make, every price change, every new product, every cost cut, gets scrutinized by people who measure success in ninety-day intervals. That pressure will shape what AI tools cost, how they are built, and how companies that depend on them get treated as customers.
A chip company called Cerebras opened that door. It went public on May 14 and raised $5.55 billion, the largest US tech listing since Uber in 2019. The stock opened nearly double its IPO price of $185, and closed the day up 68% with a market value around $95 billion. The company makes very large chips designed specifically to run AI faster than standard processors, and it has a $24.6 billion backlog of signed contracts. Cerebras has one large risk: two customers in the UAE account for the majority of its 2025 revenue. But the public market's appetite was unambiguous, with demand reportedly more than 20 times the available shares.
The strangest story of this period came from a data center in Memphis, Tennessee. Anthropic, which makes the Claude AI assistant, rented the entire Colossus 1 facility from xAI, Elon Musk's AI company. The full price was buried in SpaceX's IPO filing: $1.25 billion per month through May 2029, totaling over $40 billion. That single contract absorbs roughly half of Anthropic's current annualized revenue. The detail that makes it stranger: xAI built Colossus for its own AI assistant, Grok. Grok's users have been declining, the facility was sitting largely idle, and renting it out to a rival was simply good economics ahead of an IPO. The deal gives Anthropic 220,000 high-performance chips immediately. It gives xAI cash it badly needs.
Then there is the talent side. On May 19, Andrej Karpathy, one of the original founders of OpenAI and the former head of AI at Tesla, announced he joined Anthropic. He will work directly on the core training process that teaches Claude what it knows, with a specific focus on using Claude to help accelerate that same training process. That sentence sounds circular but it describes one of the most consequential research goals in AI: getting an AI system to help improve itself. Karpathy is the person most qualified to attempt it at scale.
This is also part of a broader Anthropic pattern. Since 2025, the CTOs of major companies including Workday, Instagram, and Box have left senior leadership roles to take hands-on research positions at Anthropic. Nobody takes a pay cut and a title reduction unless they believe the place they are going is where the most important work is happening.
What does this mean for a business operator who uses or is considering AI tools? Three things. First, these IPOs will create quarterly earnings pressure on OpenAI, Anthropic, and the companies around them. Prices, service limits, and product decisions will increasingly be shaped by what public shareholders want to see. Second, the scale of compute spending, Anthropic alone committing $15 billion per year just for server access, confirms that AI capabilities will keep improving because the financial pressure to justify that spending is enormous. Third, the companies winning right now are winning on cost and talent simultaneously, not just on the quality of their models. That combination is hard to replicate quickly.