A $3.6 Trillion Pipeline and What It Actually Contains
The IPO pipeline in AI and adjacent technology now totals approximately $3.6 trillion in combined private market valuation across its major candidates. SpaceX, OpenAI, Anthropic, xAI, and a cluster of high-profile companies represent capital waiting for a public market exit. The number is real. It is also doing a lot of work that the underlying business metrics do not yet support.
Private valuations are negotiated between founders, their existing investors, and the participants in the most recent funding round. They are not independently verified. They are not derived from earnings multiples or discounted cash flow models grounded in current performance. They reflect what the last round was willing to pay for anticipated future value. When these companies transition to public markets, that anticipation faces a different standard of evidence.
Public market investors price companies on earnings or on credible, visible paths to earnings. The transition from private valuation to public pricing is where the math for most of these companies gets complicated, and nowhere more so than at OpenAI.
OpenAI's Specific Problem: The Valuation Gap
OpenAI's last private valuation was $300 billion. To justify that number as a publicly traded company at a conventional technology price-to-earnings multiple in the range of 20 to 30 times earnings, OpenAI would need to generate $10 to $15 billion in annual net income. Not revenue. Net income.
Current estimated revenue for 2024 runs around $3.7 billion. Current spending exceeds that revenue. OpenAI is not yet profitable. The gap between where the business is and where it needs to be to justify its private valuation is not a rounding error. It is roughly a 3x to 4x revenue growth requirement, simultaneously with achieving profitability.
That scenario is not impossible. Amazon operated at losses for years while its valuation grew, because the market could see the trajectory. Tesla ran its valuation well ahead of earnings across nearly a decade. Public markets can and do price companies on expected future performance rather than current results. The question is whether the trajectory for OpenAI is as legible as it was for those cases.
Amazon's path to profitability ran through AWS, a genuinely new business line with structural cost advantages that produced reliable recurring margin. OpenAI's path runs through AI adoption continuing at its current pace, through API pricing stabilizing at levels that generate margin, and through costs declining faster than competition erodes pricing power. Each of those conditions is uncertain, and they all need to hold simultaneously.
The Cost Structure and the Path to Margin
OpenAI's cost base has three major components: compute, talent, and research infrastructure. All three are large and none of them compress easily without affecting product quality or competitive position.
Compute is the most visible cost. Training and serving frontier models at scale requires GPU clusters that cost hundreds of millions to build and tens of millions per month to operate. As the user base grows, serving costs grow with it. Efficiency gains and hardware improvements reduce the per-query cost over time, but the absolute cost base keeps rising as the product scales. Microsoft's infrastructure investment provides some cushion here, but the economics of that partnership at scale are not fully public.
Talent is the other constraint that compounds over time. The researchers capable of maintaining a frontier model position command compensation reflecting both their scarcity and the alternatives available to them. Anthropic, Google DeepMind, Meta, and Microsoft's MAI team are all competing for the same population of people, and that competition is not easing. Retention at the frontier is expensive in ways that do not appear cleanly in standard cost metrics.
The path to profitability requires revenue to grow faster than costs, which means either charging more per query, serving more queries per dollar of compute, or both at once. OpenAI's API pricing has trended down, not up, because competition from open-source models and other providers requires it. The margin improvement has to come primarily from efficiency, and efficiency gains in frontier model training and inference have been real but not unlimited.
What Investors Are Actually Betting On
The bull case for an OpenAI IPO is not about current earnings. It is about the agentic internet thesis: the idea that AI agents will become the primary interface for software tasks, that those agents will run through API infrastructure, and that OpenAI's API will be the dominant layer in that infrastructure stack.
If that thesis is correct, the per-transaction economics become compelling at scale. Every software workflow that moves through an AI agent generates API calls. At true internet scale, that is a metered business with recurring revenue tied to software usage broadly, not just to people paying subscription fees for a chat product. The analogy to AWS's infrastructure economics is imprecise but directionally useful: AWS's value was not in selling cloud storage. It was in becoming the substrate for internet software generally.
Bloomberg Intelligence offers a different read on the IPO timing. The wave could be primarily a liquidity event for early investors rather than a signal of underlying business maturity. Venture funds have capital locked in these companies. Fund lifetimes have limits. Partners who invested in 2019 or 2020 are looking at a 7-year hold with no exit in sight if these companies stay private. An IPO serves the existing investors regardless of whether it is optimal timing for the business itself.
Both framings can be true simultaneously. An IPO can be a genuine long-term bet on the infrastructure thesis and a necessary liquidity mechanism for early investors. These motivations do not cancel each other out. But they do suggest that IPO timing reflects investor dynamics as much as business readiness, which is relevant context for anyone evaluating the public offering when it comes.
The Anthropic Variable and What It Signals
Anthropic has explicitly stated it may never pursue a conventional IPO. Its Public Benefit Corporation structure creates different governance obligations than a standard Delaware corporation, designed to allow prioritizing mission outcomes over shareholder returns in situations where those objectives conflict. Staying private preserves the ability to make decisions that might be correct long-term but would be difficult to defend on a quarterly earnings call.
If Anthropic stays private while OpenAI, xAI, and others list publicly, the public markets will effectively fund the AI labs that have chosen conventional corporate structures. The safety-focused lab with the most explicit commitments around alignment research remains outside the reach of retail investors and standard institutional capital flows.
This is not inherently a problem. Private companies can do important work without public market pressure. But it creates a visible divergence in what equity markets will and will not fund. The labs most willing to accept public market accountability for quarterly results will have the broadest access to capital, and public market accountability is not primarily calibrated around long-term safety research or careful deployment decisions.
Whether that matters depends on how much you think capital access drives research direction. The answer is probably: more than it should, less than critics assume.
The Timing Problem Nobody Can Solve
Both SpaceX and OpenAI are reportedly watching each other's timing. IPO markets are open. Conditions are favorable. But going public before the business model is proven means accepting sustained scrutiny of numbers that do not yet support the valuation. Waiting means those conditions could change, and the window that exists today has no guaranteed shelf life.
The risk of moving too fast is concrete. A public company that misses earnings guidance in its first two or three years as a public entity faces a reset that is painful for employees, damaging to brand perception, and genuinely slow to recover from. The OpenAI brand carries enormous consumer recognition. Disappointing public market performance would be a different kind of story than disappointing private round performance, visible in ways that affect enterprise sales conversations and talent recruitment.
The risk of waiting is equally concrete. Capital market conditions change without warning. A significant AI safety incident, a regulatory intervention, or a broader market correction could close the window that exists right now for years. Sitting on a $300 billion private valuation while market conditions deteriorate is not a neutral choice.
What an OpenAI IPO would actually mean for the broader industry is a public price discovery moment for the agentic internet thesis. If the stock performs well, it validates the infrastructure bet and draws capital toward the category broadly. If it struggles, the narrative reset affects valuations across the sector, including companies with better current economics than OpenAI.
The $3.6 trillion number reflects expectations set in private markets.
Public markets will run their own analysis.
The gap between those two numbers is where the real story happens.
It will be a large gap.