The 4 million number sounds like a victory. OpenAI's Codex has attracted 4 million weekly active users since its public launch—a metric the company highlighted in Tuesday's announcement about Codex Labs, its new enterprise division. But the real story lives in what OpenAI did not announce: a single enterprise contract.
Codex Labs represents OpenAI's first serious attempt to move Codex from a tool individual developers experiment with into software that enterprises actually pay for. The company partnered with Accenture, PwC, and Infosys to help deploy Codex across large organizations. These consulting giants will integrate Codex into their clients' software development workflows—the same approach GitHub used to land its first Fortune 500 Copilot deals.
Here lies the paradox. OpenAI has the users. GitHub Copilot has the enterprise lock-in. Copilot launched in 2021 and has spent two years embedding itself into corporate development environments at $19 per user per month. Microsoft's advantage is not technical superiority—it is institutional momentum. Companies have already negotiated vendor agreements, updated procurement systems, and trained developers on Copilot. Switching costs are real.
OpenAI's path to enterprise revenue runs through its existing user base. The 4 million weekly actives represent developers who have already chosen Codex over alternatives. Some are already paid Pro subscribers at the rumored $30 per month price point. The question is whether OpenAI can convert these individual enthusiasts into organizational buyers before GitHub converts their employers.
This conversion problem explains the Labs branding. Codex Labs is not just a product tier—it is a sales motion. Enterprise software requires security reviews, procurement cycles, and integration support. Partners like Accenture do not just implement technology; they provide the credibility and hand-holding that corporate IT departments demand. This is how OpenAI competes against Microsoft's entrenched enterprise relationships: by becoming a consulting firm's preferred vendor rather than a direct sales competitor.
Whether this strategy works depends on factors OpenAI cannot control. Enterprise buyers care about liability, data handling, and support SLAs—not just benchmark performance. GitHub Copilot's two-year head start means it has already weathered the early adoption risks that enterprise procurement teams fear most. OpenAI enters a market where "we've been doing this for two years" is already someone else's talking point.
The 4M WAU number proves Codex works well enough for developers to keep using it. It does not prove OpenAI can sell it to enterprises. That conversion is the only metric that matters for the company's AI coding ambitions—and it is the one number OpenAI has not shared.