OpenAI pivots to coding and enterprise

This marks OpenAI choosing the highest ROI use of scarce GPUs, where coding and enterprise agents create repeat usage, bigger contracts, and tighter workflow lock in than broad consumer experiments. The practical shift is from showing what AI can do to selling tools that write code, operate desktops, and slot into existing company buying channels like ChatGPT Enterprise, cloud marketplaces, and systems integrators.1 2 3 4

• Coding had already become the revenue engine on the other side of the market. Claude Code was estimated at $2.5B in annualized revenue by February 2026, 50% from enterprise, which showed OpenAI exactly where demand was deepest and most monetizable.1 5
• The product is also changing shape. Codex moved beyond autocomplete into a fuller work agent with computer use, desktop control, web previews, and image generation, which makes it useful not just to engineers in an IDE, but to product, ops, and other desk workers handling multi step tasks.1 3 4
• Shutting down Sora is less about one failed feature and more about capital allocation. In a compute constrained market, every GPU pointed at video is a GPU not serving code generation, agent loops, or enterprise workloads, where demand is more durable and distribution is stronger.1 2 6

The next phase is a fight to become the default work surface for AI inside companies. The winner will pair the best coding model with the best enterprise distribution and enough compute to stay reliable during demand spikes, which favors labs concentrating spend on coding, agents, and enterprise deployment instead of broad consumer side projects.1 2 3 5