An initial public offering, or IPO, is essentially the moment a private company opens its ownership to the public for the first time. Instead of shares being held by founders, employees and private investors, ordinary investors can buy a stake through a stock exchange. For the company, it is a way of raising potentially enormous amounts of capital, while for investors it is the first opportunity to put a market price on a business that was previously valued largely by private investors.
That distinction matters right now because some of the biggest companies preparing to enter the public markets are also some of the most difficult businesses to value. SpaceX's IPO in June was a perfect example. It launched at roughly a $1.77 trillion valuation, making it the largest IPO ever, and briefly pushed its market capitalisation above $2 trillion. There was certainly a genuine business underneath the hype. SpaceX has revolutionised rocket launches, built a huge satellite communications operation through Starlink and created a business with significant long-term potential. But the valuation also reflected something considerably more speculative: what SpaceX could become if AI data centres eventually moved into space, alongside the company's connection to Elon Musk's wider AI ambitions. The idea of orbital data centres sounds futuristic because, well, it is. Yet investors were already being asked to put a value on that future before it had become a meaningful source of revenue.
That is what makes the current wave of AI IPOs so interesting. OpenAI, Anthropic and Databricks are not companies with vague AI ambitions. AI is their business. The question is whether the extraordinary valuations being attached to them are based on the economics of businesses that exist today, or on what investors believe AI could become several years from now.
OpenAI is probably the clearest example. The company behind ChatGPT has become one of the most recognisable technology businesses in the world, with hundreds of millions of people interacting with its products and businesses increasingly paying for access to its models and tools. Its latest private valuation is around $852 billion, and it has been preparing for a potential public listing. On the surface, that looks easier to justify than SpaceX's more speculative AI ambitions because OpenAI already has enormous real-world usage. But there is a slightly awkward problem. Making AI useful is extremely expensive. OpenAI's second-quarter 2026 revenue reached $6.7 billion, yet its operating loss widened to $12.3 billion. Revenue is growing, but so is the amount being spent to generate it.
This creates an unusual situation for investors. A traditional software company can sell another copy of its product at almost no additional cost. An AI company, by comparison, has to pay for enormous quantities of computing power every time customers use increasingly sophisticated models. The better the product becomes, the more people may use it, but that does not automatically mean the economics become better. If usage grows faster than margins, enormous revenue numbers can coexist with enormous losses.
That does not necessarily make OpenAI a bad investment. It could simply mean that the company is in the infrastructure-heavy phase of building something much larger. If AI eventually becomes embedded into almost every business and consumers are willing to pay substantially more for increasingly capable agents, today's spending could look tiny compared with future cash generation. The problem is that investors buying at an $850 billion-plus valuation are already paying for a substantial amount of that future.
Anthropic presents an even more interesting comparison. Founded by former OpenAI employees, Anthropic develops the Claude family of AI models and has increasingly positioned itself around enterprise use rather than simply consumer chat. Its valuation recently reached around $965 billion, putting it above OpenAI on the private market. More remarkably, Anthropic's second-quarter revenue reportedly reached $11.6 billion, more than OpenAI's $6.7 billion, while the company also recorded a small adjusted profit. That makes Anthropic's valuation easier to understand, but not necessarily easy to justify. The company is growing at an extraordinary rate, particularly through Claude Code and other enterprise products. If businesses are genuinely using AI to replace hours of programming, analysis and administrative work, there is a credible economic argument behind that growth.
But once again, the numbers need context. Anthropic is simultaneously forecasting an enormous expansion in future revenue, reportedly targeting around $190-200 billion in 2028 revenue. That means investors are not simply buying today's business. They are effectively betting that the current explosion in AI spending continues for years, that Anthropic maintains a significant share of the market and that the company eventually converts that revenue into substantial profits.
And this is where the AI hype argument becomes difficult to dismiss. The technology is clearly real. People really are using Claude and ChatGPT. Businesses really are spending billions on AI. Productivity is improving in certain areas. But the financial question is different: how much is that productivity actually worth, and how much of that value will end up with the companies building the models?
There is also the possibility that AI companies are trapped in an arms race. OpenAI needs more computing power to build better models, Anthropic needs more computing power to compete with OpenAI, and both need enormous amounts of capital to fund that expansion. If competitors keep improving while model prices fall, the companies may be forced to spend increasingly large amounts simply to maintain their position.
That brings us to Databricks, which arguably offers a more conventional AI investment story. Databricks is not primarily selling consumers an AI chatbot. Instead, it provides businesses with the infrastructure used to store, process and analyse data, while increasingly giving them the tools to build and deploy AI applications on top of that data. In simple terms, if OpenAI and Anthropic are building some of the most advanced AI engines, Databricks is closer to the software infrastructure that helps businesses actually use those engines. This distinction is important because Databricks already had a substantial enterprise software business before the latest AI boom. Its AI opportunity is therefore an extension of an existing business rather than the entire justification for its existence. Recent reports have put its potential valuation somewhere around $165-175 billion, depending on the financing terms, with revenue growth remaining strong.
There is arguably a much stronger fundamental case here. Companies need somewhere to store their data, analyse it and connect it to AI systems regardless of whether the eventual winner is OpenAI, Anthropic, Google or another model provider. Databricks can potentially benefit from the expansion of AI without having to win the entire model war itself. But even Databricks is not immune to the valuation question. At hundreds of billions of dollars, investors still need to believe that its existing software business and AI expansion can generate enough future cash flow to justify an enormous premium. The company may have a more established economic foundation than a frontier AI laboratory, but that does not mean every dollar added to its valuation is automatically justified.
This is ultimately why the upcoming AI IPOs are so interesting. They provide three slightly different ways of betting on the same technological revolution. OpenAI represents the consumer and frontier-model opportunity. Anthropic represents the increasingly lucrative enterprise AI market. Databricks represents the infrastructure and data layer underneath it all.
The common factor is AI, and that may also be the biggest risk. Investors have spent years rewarding almost anything associated with artificial intelligence with extraordinary valuations. SpaceX demonstrated just how powerful that narrative can become when a company with an already successful business suddenly acquires an additional AI story and investors begin pricing in possibilities that have barely reached reality.
The question is therefore not whether AI is overhyped. Saying that would be too easy, and probably wrong. AI is already producing real economic value. The more interesting question is whether the companies at the centre of the boom will capture enough of that value to justify the prices investors are currently willing to pay.
OpenAI and Anthropic may eventually become enormously profitable companies, but they first have to prove that revenue can grow faster than the cost of producing intelligence. Databricks has arguably solved a more conventional problem, but its valuation still assumes years of rapid AI-driven expansion. Perhaps that is the real lesson from the SpaceX IPO. A company does not have to be fraudulent, unprofitable or fundamentally bad for its valuation to be driven partly by hype. It simply has to have a believable future that investors become willing to pay for today.
The upcoming AI IPOs will put that theory to the test. If OpenAI, Anthropic and Databricks can turn extraordinary usage and revenue growth into equally extraordinary profits, their valuations may eventually look entirely reasonable. If they cannot, investors may discover that they were not buying the AI economy of today at all. They were buying a very expensive version of the economy they hoped would exist tomorrow.