OpenAI's $600 Billion Reality Check

How the company that started the AI era is losing the revenue race to a competitor that spends 4x less, and what it proves about where AI value actually lives

Author

Justin Wright
May 04, 2026 • Est. Reading Time: 5 minutes

Happy Monday!

Last week, the Wall Street Journal reported that OpenAI missed its own targets for both revenue and user growth. The specifics are worth sitting with: the company fell short of an internal goal of one billion weekly active ChatGPT users, it missed multiple monthly revenue targets earlier this year, and its CFO, Sarah Friar, privately warned leadership that revenue may not grow fast enough to cover the company's compute commitments.

Those commitments total roughly $600 billion.

Last week, we covered SpaceX paying $10 billion for a call option on Cursor because Grok could not crack the AI coding market. The thesis was straightforward: the product is the moat, not the model. OpenAI is now proving the same point from the opposite direction. It has the most recognized AI brand on earth, 900 million weekly users, and more locked-in compute than any company in history. And it is losing the revenue race to Anthropic, which just passed it at $30 billion in annualized revenue while spending a fraction of what OpenAI spends to train its models.

OpenAI missed internal revenue and user growth targets, with CFO Sarah Friar warning that $600 billion in compute commitments may outpace revenue growth. Meanwhile, Anthropic passed OpenAI in revenue at $30 billion ARR while spending 4x less on training. The difference: Anthropic earns 80% of revenue from enterprise customers, OpenAI earns 60% from consumers. OpenAI built the biggest audience in AI. Anthropic built the products that enterprises pay for. The revenue followed the product, not the brand.

TL;DR

The Numbers Behind the Miss

OpenAI is running at roughly $25 billion in annualized revenue as of March 2026. That is an extraordinary number by any historical standard. It is also below the company's own projections, and the trajectory is the deeper problem.

ChatGPT's consumer growth slowed late last year as Google's Gemini gained ground. In coding and enterprise, Anthropic's Claude took market share on multiple fronts. Claude Code now holds over half the AI coding market and generates $2.5 billion in annualized revenue on its own. The result: OpenAI missed several monthly revenue targets this year despite having the largest user base of any AI product ever built.

The revenue composition tells the deeper story. Roughly 60% of OpenAI's revenue comes from consumer subscriptions: ChatGPT Plus, Pro, and Team. Enterprise makes up about 40% and is growing, but the consumer-heavy mix creates a structural challenge. Consumer AI subscriptions are low-margin, high-churn, and vulnerable to free alternatives. Enterprise contracts are stickier, expand over time, and carry higher retention rates.

OpenAI vs. Anthropic: The Revenue Picture

Metric

OpenAI

Anthropic

Annualized revenue (April 2026)

~$25B

$30B

Revenue from enterprise

~40%

~80%

Revenue from consumer

~60%

~20%

Projected 2026 losses

$14B

Targeting FCF positive by 2027

Training cost efficiency

Baseline

~4x more efficient

AI coding market share

Minority (Codex)

Majority (Claude Code)

Weekly active users

900M

Not disclosed

The $600 Billion Overhang

The missed targets are concerning on their own, but they become alarming in context. OpenAI has locked in approximately $600 billion in future compute contracts, accumulated through years of aggressive dealmaking under Sam Altman's thesis that compute scarcity was the fundamental constraint on AI progress.

The largest single commitment is a five-year, $300 billion agreement with Oracle tied to the Stargate data center project. That deal alone requires $60 billion annually starting in 2027. Add to that an expanded $100 billion Amazon agreement, an $11.9 billion CoreWeave contract, and additional Nvidia commitments.

Friar warned that revenue may not grow fast enough to fund these contracts. She and Altman are reportedly at odds over spending discipline and IPO timing. Altman wants to go public by year end. Friar has cautioned that internal controls are not ready for public market scrutiny. Both called the report "ridiculous" in a joint statement.

The math does not care about alignment. OpenAI projects $14 billion in losses for 2026 and has pushed its breakeven target to 2030. Revenue needs to not just grow, but accelerate rapidly to cover obligations that start compounding next year.

Why Anthropic Is Winning the Revenue Race

Anthropic grew from $1 billion to $30 billion in annualized revenue in fifteen months. It passed OpenAI in April while spending roughly four times less on model training. OpenAI is projected to spend $125 billion annually on training by 2030. Anthropic's projection for the same period is around $30 billion.

The difference is structural. Anthropic built an enterprise company that happens to have a consumer product. OpenAI built a consumer company that is trying to become an enterprise company. Those are fundamentally different businesses with different economics.

Anthropic now has more than 1,000 customers spending over $1 million annually, a figure that has doubled since February. Claude powers the developer tools where engineers spend money: Cursor, Windsurf, and Claude Code. Eighty percent of Anthropic's revenue comes from enterprise API usage and developer contracts. That revenue base has higher retention, better expansion economics, and lower churn than consumer subscriptions.

OpenAI has 900 million weekly users, but Anthropic has the revenue advantage. That inversion captures something important about where AI value is consolidating. The largest audience does not automatically produce the largest revenue when the paying customers are choosing a different product.

What This Means for Practitioners

This is the second week in a row where the same thesis has surfaced from a different angle. Last week, SpaceX paid $10 billion for Cursor because the product layer, and not the model layer, is where revenue converts in AI. This week, OpenAI demonstrates the corollary: even the most powerful brand and the largest user base in AI cannot compensate for losing the product competition in the segments where customers actually pay.

If you are evaluating AI tools for your organization, the revenue inversion should inform your procurement strategy. Enterprise buyers are increasingly choosing Claude for coding, analysis, and agentic workflows. The company with the most users is not necessarily building the best product for your use case.

If you are building an AI startup, the lesson is about capital efficiency. Product quality, developer experience, and enterprise go-to-market matter more than raw compute expenditure once models reach competitive parity on benchmarks. Spending more on training does not automatically produce more revenue.

If you are an investor, the $600 billion overhang is the number to watch. OpenAI needs revenue to accelerate significantly by 2027 when the Oracle contract begins. If it does not, the gap between commitments and revenue becomes a balance sheet problem and not just a growth story.

The Bottom Line

OpenAI pioneered the AI era, built the most recognized AI brand in the world, and attracted 900 million weekly users. It also locked in $600 billion in compute contracts, missed its own revenue targets, and watched Anthropic pass it while spending a fraction on training.

The pattern is unmistakable: Musk paid $10 billion because Grok could not build the coding product. OpenAI is losing revenue to a company that builds better enterprise products on models that cost less to train. The model is necessary but not sufficient. The product layer is where revenue converts. The company with the best product is pulling ahead of the company with the biggest audience.

In motion,
Justin Wright

If 900 million weekly users and $600 billion in compute commitments cannot outpace a competitor spending 4x less on training, what exactly is OpenAI's moat, and how long can the assumption that scale alone wins hold up against a rival that is simply building better products?

Food for Thought

If you haven’t listened to my podcast Mostly Humans: An AI and business podcast for everyone yet, new episodes drop every week!

Episodes can be found below - please like, subscribe, and comment!