Unpacking an OpenAI IPO

Venture investing and retail stock investing are two very different things. Venture investors, depending on the stage, look at a lot of factors that are not specific to the economics of the underlying business. Early stage venture investors like us look at the founder and whether we think they both have a vision worth backing and whether they have the ability, strength, resilience, charisma, management skills and a dozen other factors to then turn that vision into reality. We look at the total addressable market to see how big the overall opportunity is and then we work backwards to see what success would look like if the company captures a relatively small share of that market. We look at their team, their tech, if they have other investors on the cap table that can bankroll future rounds and a bunch of other considerations.

Later stage venture investors look at most of those factors too, but are also sometimes driven by the need to simply deploy large amounts of capital because their fund sizes are massive and their economics are driven more by collecting 2% of assets under management every year than by the carry earned by making money for LPs. That often means investing at really high valuations, whether the fundamentals warrant it or not.

In contrast, retail investors typically focus — or at least should focus — on the fundamentals of the company itself. What are revenues, what’s EBITDA, what are the margins, where are the opportunities for growth, what are the market risks, what are the risks from competition and so on. In most cases, the market eventually correlates the fundamentals with the share price (Tesla consistently seems to be one very high profile exception). That’s why quarterly earning calls are important – they detail the fundamentals every three months and keep the market honest.

In fact, many high profile IPOs ultimately result in lower valuations – the market sees the disparity between the valuation assigned to the company by venture investors and the reality of the company’s underlying economics and corrects it. If you look at tech IPOs with valuations of $10 billion or more from 2015-2025, after twelve months, the median valuation was 32% below the IPO price, covering two thirds of those companies. After twenty four months, the median remained down 32%, but now covering 75% of those companies. So in other words, the first day is often the best day, at least for the first few years after going public.

That’s why the much rumored (but not yet confirmed) September IPO from OpenAI is so interesting. In many ways, OpenAI is an absolutely incredible company. They are at the forefront of perhaps the most transformational technology ever. They have over 800 million weekly active users — that’s well over the entire US population two times over. They handle over 2.5 billion prompts per day. And the ads they’re running during the NBA playoffs are truly outstanding.

But with that said, choosing to invest in OpenAI will not be a simple equation based on revenue, multiples, profits and margins. It requires making multiple bets and assumptions on multiple fronts. The expected IPO valuation is around $1 trillion. There’s no reason for anyone to invest in OpenAI (or anything) unless they think they can generate a meaningful profit in a reasonable amount of time. So if we assume someone wants to make 25% year over year on the stock – and given the risk, if anything, the expectation should probably be even higher – here’s what you have to take into account (to be clear, this is just my analysis and not investment advice):

1. The entry price: A retail investor likely won’t be able to buy in at the expected $1 trillion offer because there will almost certainly be a first day pop. Assume 30% and the real cost is $1.3 trillion, which is around 54x OpenAI’s current run rate of $24 billion (if the pop doesn’t happen or the initial IPO price reflects a lower valuation, the rest of the math adjusts accordingly).

2. The market cap needed to generate a reasonable return: Earning 25% a year for three years means reaching a valuation of about $2.54 trillion in 2029. Here are the companies currently worth more than $2.5 trillion: Amazon, Alphabet, Apple, Nvidia and Microsoft. They have an average of $400 billion in annual revenue and profits of over $100 billion.
3. Or look at Dell. As of this writing, they have a market cap of around $297 billion. They expect to generate $165 billion in revenue this year. Multiples for a hardware company are very different from a platform but the contrast is still stark and striking (by comparison, last year, OpenAI had $13 billion in revenue and $9 billion in losses.)
4. The challenge of high multiples at scale: A business doing $100 billion or more in revenue doesn’t trade at 54 times sales. It’s far more like 10 to 15 times (Nvidia is more like 25x but runs a 50% profit margin; the rest are mainly in the 10-15x range). The bigger the revenue base, the lower the multiple so you can’t rationally assume both at the same time.
5. SAAS companies with high multiples tend to enjoy 70-80% gross margins. OpenAI projects tearing through $665 billion in cash over the next four years. Companies with that kind of capex and opex can’t command SAAS multiples (maybe a new type of agentic ecosystem model can be created, but that’s idle speculation at best).
6. The revenue needed: Just to not lose money at a $1.3 trillion valuation means that revenue has to more than triple to $100 billion from OpenAI’s projected $30 billion in revenue this year. To make 25% year over year on the stock means that revenue has to reach $195 billion in three years at a 13x multiple. That exceeds even OpenAI’s own projections by 56% (they project $125 billion in revenue in 2029).
7. OpenAI projects roughly $280 billion in revenue in 2030, so by then, assuming they can actually reach those projections, the math starts to work. But it assumes that everything they think will happen actually does happen — and that none of the underlying risks truly materialize ranging from brutal competition to increasingly stringent regulation to ever changing geopolitics to massive debt service and a lot more.
8. It took Amazon, the company most often cited for chasing revenue over profit, about a decade to do the same thing at comparable scale: from roughly $25 billion in 2009 to $280 billion in 2019, across far more business lines than OpenAI runs, and with a fraction of the debt. And while you can argue AI has more upside than any technology before it, Amazon was effectively one of one in its market — Walmart only caught up later and only partially. And while the sectors are different, Amazon trades at a multiple of just 4x revenue.
9. Debt stacked on top of it: OpenAI already has $1.4 trillion in compute commitments. It’s technically not debt in the traditional sense, but it is contractually obligated spending, which functions similarly. OpenAI has also said it plans to spend $600 billion in compute operating expenses through 2030. The two figures aren’t additive, but they’re also not interchangeable: the $1.4 trillion is the capital commitment to build the infrastructure and the $600 billion is the ongoing cost to use it. Both are real obligations stacked ahead of equity holders, and neither disappears when the company goes public.
10. While the capex isn’t stacked directly on top of the market cap, it does reflect costs that effectively require a much higher revenue target to make the math work. At a 15x multiple on revenue, the capex means OpenAI really has to reach more like $260 billion in three years, which is now more than twice what even they project to have by then. If the multiple is more like 10x, which is reflective of Alphabet (12x), Microsoft (11x), Apple (11x) and Meta (9x), then almost $400 billion in revenue is needed to produce a reasonable rate of return on the stock. Removing state-owned enterprises of China and Saudi Arabia, only five companies have that kind of revenue: Walmart (which is 64 years old), Amazon (32 years), Apple (50 years), Alphabet (28 years) and UnitedHealth Group (49 years).
11. Basically, to have any chance of reaching the needed $2.54 trillion market cap, let alone something that factors in the massive capex, OpenAI has to at least increase revenue by 10-15x and still command a healthy enough multiple to get there. Maybe they can but it’s a herculean task.
12. Competition: OpenAI does not operate in a vacuum. It has to compete against Anthropic, Gemini and Grok. That means going up against a company in Anthropic with a significantly higher run-rate of $47 billion but without the $1.4 trillion in compute commitments and a far better public reputation. Then, on top of that, you’re taking on both Google (owner of Gemini) and Elon Musk (owner of Grok). Microsoft does own 27% of OpenAI, but the relationship keeps loosening and shifting – in fact, it has been renegotiated twice in the last seven months alone. So the competition will be fierce.

7. Regulation. AI is deeply unpopular and only getting more so. In a February–March 2026 Annenberg/University of Pennsylvania survey, just 17% of Americans said AI will have a positive impact on the country over the next decade against 42% who expect it to be negative. Nearly two-thirds — 65% — said the government has done too little to regulate it, a complaint shared across the spectrum: 77% of Democrats, 72% of independents, and even 53% of Republicans. Other polling puts the share of Americans who fear AI could eventually threaten humanity at over 75%. And on the question that most impacts OpenAI’s business model, nearly half oppose new data centers being built in their own communities. If data centers aren’t built, OpenAI can’t reach the scale it needs to come anywhere near the revenue projections needed to justify their IPO valuation let alone any expectation of growth.
8. New rules and regulations are hitting from every direction: states are passing everything from data center permitting and zoning limits to constraints on what chatbots can say and do to restrictions on AI in hiring and health care, the EU has enacted sweeping AI regulation, and China just does whatever it wants, whenever it wants (assuming they ever even give OpenAI market access).
9. And all of this is happening before the job losses land at scale. Even the optimistic estimates aren’t particularly reassuring — Goldman Sachs pegs near-term direct displacement risk at around 2.5% of US employment, rising to 6–7% under broader AI adoption. Other forecasts are far more pessimistic. 41% of working Americans already say they’re worried about losing their job or having their hours cut because of AI. For OpenAI to have the kind of revenue they envision, their products need to replace employees across the economy to justify their sales projections. But the more that happens, the more they will face even tougher regulatory scrutiny and greater public antipathy.
10. Politicians respond to exactly one thing: what impacts their next election. When voters hate something this much, politicians hate it too. Which means the regulatory environment OpenAI enjoys today is probably the most permissive it will ever see. And if a Democrat wins the presidency in 2028, it very well may come amid a rising wave of anti-AI sentiment — leading to tough new federal restrictions stacked on top of everything that states and cities are already doing.
11. Geopolitics. We just watched what happens when a single country threatens one critical chokepoint in the Strait of Hormuz. Now imagine China invades Taiwan — which produces roughly 90% of the world’s most advanced semiconductors. The US is racing to build domestic capacity, but on the leading-edge nodes that AI depends on, American fabs will run four to five years behind Taiwan for the rest of the decade. The US isn’t projected to reach even 17–20% of global chip production until the early 2030s. If Nvidia can’t get its chips made, OpenAI having the compute it needs to operate will be incredibly difficult.
12. Layer on the complete unpredictability of Trump, a never-ending Russia–Ukraine war, the safety risks of AI itself, worsening climate change, the spread of nuclear weapons, the falling cost of assembling bioweapons and a world drifting from global cooperation toward zero-sum nationalism — and assuming that nothing disrupts OpenAI’s business may be the riskiest assumption of all.

As a venture investor, you can enter a company at an exponentially lower valuation than retail investors see, so many of the risks above don’t really matter. If you can get into a company at a sub-$100 million valuation and it eventually gets to $1 billion in annual run rate (let alone $280 billion), then we’re looking at an IPO valuation of $5-10 billion (or more, depending on the sector and its multiples). Even if you materially discount the share price until the lockup expires six months after the IPO, you’re still going to make a lot of money (of course, finding that company is very hard).

As a retail investor, you’re a price taker. You have to wait until the shares are publicly available and then factor in all of the relevant variables to see whether you reasonably believe the share price increases enough to merit the risk. To rationally invest in OpenAI after the IPO, it means believing that all of the following will happen:

• Generating massive, unprecedented revenue growth of 10x within just a few years, eclipsing any rate of growth at this pace seen to date;

• Assuming that a new agentic ecosystem category with multiples will come into existence, defying how multiples across the rest of the market works;

• Bucking the trend of major new public companies trading below their IPO price in the first few years;

• Being able to service $1.4 trillion in capex and $600 billion in cash burn and still reach profitability;

• Achieving this in the face of massive, unrelenting, well funded competition that is better funded (Google), better liked (Anthropic) or incredibly powerful with an intense hatred of OpenAI (Elon/ Grok);

• Believing that public sentiment on AI will flip on its head rather than only getting worse despite the major job loss that effectively has to happen for OpenAI to meet its revenue targets;

• Believing that regulation at every level of government and in every part of the world will subside rather than increase despite all evidence to the contrary;

• Believing that all of the data centers needed to reach revenue targets can actually be built as quickly as envisioned (or at all);

• That the chips needed for compute will continue to be readily available; and

• That nothing else terrible happens geopolitically (or happens generally because of something that went wrong with OpenAI or just AI in general).

That’s a lot to ask a retail investor to take on, especially when they can just as easily decide to wait and see how everything unfolds.

Now, maybe ChatGPT does become the fastest scaling product of all time. Even if most of their users aren’t paying customers, it’s still a very popular product. Maybe AI is as absolutely transformative as promised and OpenAI is integral to much of that transformation and the revenue generated by it. Maybe an agentic ecosystem comes about and is priced differently than anything else. If that’s true, then perhaps buying OpenAI stock as soon as you can makes sense.

But keep in mind, once the company is public and the shares are fully accessible, you can invest at literally any time. Given all of the challenges detailed above, to me, buying OpenAI shares on day one seems like a very risky decision.

‍