ChatGPT Changed the World. So Why Is OpenAI Burning $14 Billion a Year?

Let me map out how I expect this to unfold across three time horizons, starting with the near-term dynamics and working forward to 2030 and beyond.

The IPO timeline itself is the first embedded risk, and one I think the market is systematically underpricing. Under confidential S-1 filing rules, the full prospectus remains sealed until 15 days before the roadshow begins. That puts the likely S-1 public disclosure window in late July to early August 2026. The timing problem I keep coming back to is this: the EU AI Act enters full enforcement on August 2, 2026 — almost exactly the day OpenAI's prospectus will appear for the first time. Institutional investors absorbing the S-1 will simultaneously be processing that this company operates under a regulatory framework capable of levying fines up to 7% of global revenue for serious violations. That dual-information moment — IPO prospectus and major new liability framework arriving together — is not an environment that underwriters want, and I believe this timing overlap represents a meaningful, systematically underappreciated drag on institutional book-building.

The lockup expiration window creates a second near-term vulnerability that is easier to quantify. Standard post-IPO lockup periods run 90 to 180 days after listing. In OpenAI's case, that means the supply held by Microsoft at 27%, the Foundation at 26%, and the investors from the March 2026 private funding round — where $122 billion was raised at an $852 billion post-money valuation — could all become eligible for public sale within six months of the listing date. My estimate is 15% to 25% downward price pressure at or around the lockup expiration window. The relevant precedent is ARM Holdings, which surged roughly 70% in the weeks after its 2023 IPO and then corrected more than 20% when the lockup window opened. OpenAI's pre-IPO shareholder base is considerably larger and more concentrated than ARM's was, which amplifies rather than reduces the overhang risk. This should be a central input to any timing strategy around the IPO entry.

Moving to the medium-term picture, roughly the 2027 window, this is where the real verdict on the IPO narrative begins to form. OpenAI's internal advertising revenue target for 2027 is $11 billion. The pilot program gave a genuinely encouraging signal — $100 million ARR in six weeks is a striking early velocity for any digital advertising product. But scaling from $100 million to $11 billion requires 110-fold growth in a single year in one of the most competitive advertising markets on earth. Google's search advertising franchise generates over $200 billion annually, and if ChatGPT can capture even 5% of that addressable market, $10 billion is arithmetically achievable. The complication is that Google is pushing Gemini into AI-native search with the full weight of its advertising infrastructure and distribution advantages. At the same time, hyperscaler AI capital expenditure is projected to surpass $800 billion in 2027, meaning OpenAI's infrastructure costs keep rising even as it tries to demonstrate meaningful unit economic improvement. Revenue must grow substantially faster than costs, for an extended period, for the bull narrative to hold.

On the competitive landscape, Anthropic is likely to complete its own public market debut by mid-2027. Anthropic's valuation already exceeds OpenAI's pre-IPO mark at $900 billion to $950 billion, its models train at one-quarter of OpenAI's cost, and as of April 2026 it already generates more ARR. When two public companies with comparable P/S multiples trade simultaneously in the same AI sector, capital allocators naturally gravitate toward the more capital-efficient operator — this is a mechanical market dynamic, not a speculative prediction. Adding xAI at a $244 billion valuation and Meta's open-source Llama models applying downward pressure on API pricing across the industry, the CNBC analysis headlined "Cheap AI could derail OpenAI and Anthropic's IPOs" reads not as sensationalism but as a sober structural observation about where the competitive floor is moving.

The bull case — which I assign a 20% to 25% probability — requires something close to the Amazon playbook replaying in the AI context. The necessary and simultaneous conditions are: GPU-driven inference cost reductions accelerating sharply from 2027 onward as next-generation chipsets achieve meaningful efficiency gains, ChatGPT capturing material and durable share in the global search advertising market before Google fully transitions its own search product to Gemini, and paid conversion rates climbing from today's 6% toward 15% or better by 2029. If all three materialize on schedule, OpenAI's internal road map — $100 billion to $125 billion in 2029 revenue at a 70% gross margin producing $14 billion in net profit — becomes achievable. At that outcome, a 2029 P/E ratio in the 70-times range is aggressive but not irrational for a platform at inflection. The critical caveat that makes me assign only 20% to 25% probability: all three variables must cooperate simultaneously, each has independent failure modes, and any single miss slides the profitability timeline to 2031 or later, which at that point may not be early enough to satisfy the investors who bought in at $1 trillion.

The base case — which I put at 45% to 50%, making it my modal single forecast — sees intensifying competition moderating revenue growth while inference costs improve only gradually, pushing the path to sustained profitability into 2031 or 2032. The key structural assumptions are: ChatGPT's market share stabilizing in the 45% to 50% range by 2028 rather than collapsing further, the revenue gap with Anthropic narrowing back into a genuine two-player duopoly dynamic rather than a continuing Anthropic lead, and the advertising ramp delivering somewhere in the range of $5 billion to $7 billion by 2027 rather than the full $11 billion internal target. HSBC's own analysis supports this framing: even projecting $213 billion in 2030 revenue in an optimistic scenario, it still expects cumulative free cash flow to remain negative through the end of that year. Under this scenario, sustained high interest rates through 2027 could push actual IPO pricing 20% to 30% below the stated target range — landing the deal at $600 billion to $700 billion. This is a survival outcome rather than a triumph, and it is the most probable single path based on current evidence.

The bear case — which I assign 25% to 30% probability, a figure higher than most mainstream commentary would acknowledge — is structurally a WeWork replay under different surface conditions. The specific catalyst combination that triggers this scenario: EU AI Act enforcement creating immediate compliance cost spikes and investor hesitation at the precise moment of the IPO roadshow, open-source AI models reaching commercial-grade performance and stripping OpenAI's enterprise pricing power in the most profitable segment, and a governance crisis resembling the November 2023 Sam Altman board-firing episode recurring at maximum possible damage to investor confidence. WeWork filed at a $47 billion valuation, watched governance disclosures destroy institutional trust within weeks of the S-1 appearing publicly, withdrew the offering entirely, and filed for bankruptcy in late 2023. SoftBank's total loss reached $14.4 billion. An analogous event at a $1 trillion valuation would produce a catastrophic outcome on an order of magnitude approximately ten times larger. The Foundation's 26% stake creates a structural tension between shareholder value maximization and a humanity-benefit mandate that legal restructuring alone cannot fully resolve.

Over the longer horizon — three to five years — 2029 becomes the year of maximum informational value and the moment when the company's trajectory becomes definitively legible. OpenAI has explicitly committed to profitability that year in its internal financial documents. Sustaining the projected $44 billion in cumulative losses from 2023 through 2028 while remaining operational requires not just IPO proceeds but repeated additional capital raises along the way. Each successive raise dilutes prior shareholders. An investor entering at the 2026 IPO could realistically see their ownership stake reduced by 20% to 30% through subsequent equity issuances before the 2029 profitability date arrives — whether or not the profitability actually arrives. Amazon spent roughly seven years essentially flat after its 1997 IPO before AWS unlocked a parabolic upside nobody at listing time anticipated. Whether OpenAI has an equivalent hidden revenue stream — a category-defining business not yet visible in its current product mix — remains genuinely unknown. That is not necessarily a reason to avoid the investment; it is a reason to price the uncertainty correctly rather than assuming the answer is yes.

Looking beyond 2030 at the market structure question, I expect the current ChatGPT-centric AI landscape to reorganize around three to five dominant platforms: Google, Microsoft in partnership with OpenAI, Anthropic, Meta, and potentially one Chinese platform — ByteDance or Baidu — depending on geopolitical developments. There is no structural guarantee OpenAI holds the top position in that reorganized landscape, particularly given that Anthropic already leads on ARR in mid-2026. If Microsoft builds out its own AI capabilities more aggressively and begins reducing strategic dependence on OpenAI — a process that becomes much easier after the technical IP license expires in 2032 — the entire strategic rationale for the current partnership transforms in ways that are hard to model from today's vantage point. I put the probability of OpenAI existing as an independent company at or above $1 trillion market cap in 2030 at roughly 40%. The probability it gets absorbed into a larger platform through M&A: approximately 30%. The probability it survives as an independent company but at a materially lower valuation: the remaining 30%.

There are wild cards capable of invalidating every scenario I've laid out, and intellectual honesty requires acknowledging them. The most consequential would be an AGI-level or AGI-adjacent capability breakthrough arriving in 2027 or 2028. Every financial projection above assumes gradual, incremental improvement in model capability and cost efficiency. A genuine intelligence discontinuity — a qualitative leap rather than an incremental step — would expand the total addressable market by an order of magnitude and potentially make today's $1 trillion valuation look like an obvious bargain in retrospect. The second wild card is geopolitical: if escalating U.S.-China AI competition leads Washington to formally designate OpenAI as a national AI champion and direct large-scale federal procurement its way, the revenue trajectory could decouple from commercial market dynamics entirely. Both scenarios are real, neither is easy to assign a precise probability to, and both should factor into any honest accounting of what this company might become.

If you are an individual investor considering participation in this IPO, my practical advice is to avoid rushing the entry. Watch two full earnings cycles after the listing before committing capital. Three metrics deserve particularly close attention in those quarters: whether operating margin improves past negative 100% on a clear sequential trend line showing consistent movement, whether quarterly advertising revenue exceeds $500 million showing the 2027 target is within range, and whether paid conversion rates have crossed 8% showing the monetization engine is genuinely accelerating. If all three indicators are moving in the right direction by mid-2027, you will have real forward-looking evidence to work with rather than optimistic projections embedded in an S-1. Factoring in lockup expiration dynamics, the first half of 2027 may represent a structurally better entry point than the IPO itself, once the supply overhang from Microsoft, the Foundation, and private investors has had time to clear. The fundamental question this IPO poses is not whether you believe AI will transform the world. The question is whether this specific company can bend its cost structure toward profitability within three years — and whether whatever price the market assigns at listing properly compensates for the considerable execution risk that transformation requires.