Codex is OpenAI's bet to prove it can make money
There is a pattern that repeats itself in the history of technology companies seeking access to capital markets: the moment when the narrative of massive users is no longer enough and they need to show something more concrete. OpenAI is there. And the tool it chose to make that argument is not ChatGPT, but Codex, its software development assistance product, which in the last two months has received updates at a frequency that no competitor has matched.
From late March through May 21, Codex incorporated integrated browsing, operating system operations, pull request reviews, remote SSH connections, mobile access, a Chrome extension, access tokens for enterprise workflows, team-shared plugins, administrator usage tracking, remote desktop control with a locked screen, and an extended execution mode that allows the tool to work for hours without user intervention. This is not a cosmetic list of features. It is an architecture that describes what type of customer OpenAI is targeting.
The jump in active users is consistent with that direction: from 1.6 million weekly users in March to more than 4 million in May, according to data from the company itself. More than a growth metric, it is a signal that the market Codex is targeting — engineering teams at companies that pay for productivity — has a demand that responds to the product.
The argument that ChatGPT cannot make alone
ChatGPT is OpenAI's most recognized product and its greatest brand asset. It is also, in terms of financial architecture, a complex burden: every conversation consumes inference, every active user adds computational cost, and the equation between subscription revenue and operating costs remains difficult to close at massive scale. According to data available in KuCoin's analysis, OpenAI's adjusted operating margin in the first quarter of 2026 was approximately -122%. For every dollar of revenue, the operation cost around 2.22 dollars.
That number cannot be resolved with more ChatGPT users. It is resolved, at least partially, with enterprise customers who pay higher rates, who have contracts, who integrate the tool into productive workflows they cannot easily abandon, and who generate more predictable revenue than mass-market subscriptions.
Codex is designed for that type of commercial relationship. Not because it is "more advanced" than ChatGPT in abstract terms, but because its most recent features are built to fit into the processes that engineering departments already have: code review, continuous integration, permissions management, usage auditing, approval workflow automation. Each of those features responds to a real objection that a chief technology officer raises before approving a purchase. The fact that OpenAI has resolved those objections in the form of a product, and not just a promise, is what distinguishes this round of updates from an ordinary roadmap.
The underlying financial argument is that software development is one of the few sectors where the cost of skilled labor is sufficiently high and measurable to justify the price of a sophisticated automation tool. A senior engineer in markets like the United States or Europe costs between 150,000 and 300,000 dollars per year in total compensation. If Codex can consistently accelerate their output by 20 or 30 percent, the math for the corporate buyer becomes relatively straightforward, and the price of an enterprise license falls within the margin of what is already being spent.
The shadow coming from Anthropic
The pressure on OpenAI has a concrete origin: Anthropic is closer to operational profitability than the market anticipated twelve months ago. According to reports from The Wall Street Journal cited in available sources, Anthropic expected to surpass 10.9 billion dollars in revenue in the second quarter of 2026 and to approach its first quarterly operating profit, with an estimate of 559 million dollars. For a company that until recently was described almost exclusively as a computational black hole with good security intentions, that figure reshapes the competitive landscape.
The path Anthropic took was not one of mass popularity. It does not have a product with the recognition of ChatGPT, nor a comparable user base among consumers. What it built was a concentration on high-value enterprise use cases, and Claude Code was the most visible vehicle of that bet in the software development segment. The sequence was gradual but coherent: developers adopted it individually, teams followed, and eventually the product entered corporate procurement budgets. In April 2026, Anthropic's adoption rate among companies using the Ramp payment platform rose to 34.4 percent, surpassing OpenAI at 32.3 percent, according to data included in KuCoin's analysis. It is not a global market study, but the direction it points to is clear enough for OpenAI to take seriously.
Codex is doing, with more resources and a faster update cadence, what Claude Code did first. The difference is that OpenAI arrives with a broader brand, a larger installed user base, and a potential integration with ChatGPT Enterprise that Anthropic cannot directly replicate. The disadvantage is that it arrives later, to a market where Anthropic has already established expectations of quality and frequency of improvement.
What is technically interesting is not the confrontation between the two companies as if it were a competition of models on benchmarks. It is that both are converging toward the same business thesis: that the software engineering workflow is the most sustainable entry point into the enterprise budget, because it combines high willingness to pay, high exit friction once integrated, and a value chain where savings are quantifiable. If that thesis is correct, the market will reward whoever achieves greater depth of integration before whoever has the model with the highest score on technical evaluations.
What the pace of updates reveals about the real strategy
There is something that deserves attention beyond the list of features: the frequency with which they were released. Almost one update per week for two months is not the pace of a product team working at cruising speed. It is the pace of a team executing against a very specific deadline or external pressure.
OpenAI is preparing its opening to capital markets. The exact timing is not confirmed in available sources, but the context is explicit: the company needs to build an argument for investors that goes beyond the popularity of its chatbots. The thesis that Codex makes it possible to present is different from that of ChatGPT: it does not speak of millions of free users or mass-market subscriptions, but rather of integration into productive workflows for which companies with real engineering budgets are already paying.
That is the threshold that changes the conversation with an institutional investor. It does not matter whether ChatGPT has one hundred million active users if the revenue architecture behind that cannot scale without costs scaling proportionally or faster. What a capital market wants to see is at least one line of business where the revenue model is understandable, the customer is stable, and the economic unit can improve over time. Codex in the enterprise segment can make that argument in a way that ChatGPT cannot make on its own.
The CEO of Codex summarized the plan with irony, according to available sources: better models, products that are updated every week, more computing capacity. What he did not say, but the pattern indicates, is that each of those three elements points to a specific audience. The better model justifies technical adoption. The product that is updated weekly justifies retention and contract expansion. And computing capacity is the necessary condition for all of the above not to collapse when scale arrives.
The shift that Codex represents is not that artificial intelligence entered software development. That already happened. What is taking place is that AI tools for engineering are moving from being individual options that developers use on their own to becoming infrastructure that companies purchase, manage, and govern in a centralized manner. That transition — from personal tool to manageable enterprise asset — is the moment when the market begins to generate predictable revenue and contracts with duration. OpenAI, with two months of very precise updates, is betting that that moment has arrived and that Codex can be the product that captures it before its competitors consolidate their advantage.
