Weekly active users of Codex in Germany have multiplied twentyfold since January 2026, according to OpenAI. This metric highlights a decisive reallocation of developer resources away from general multimedia consumption. Behind this shift lies a fundamental restructuring of infrastructure priorities. Speaking at DevDay Exchange 2026 in Berlin, Romain Huet, head of developer experience at OpenAI, stated that the company abandoned video processing models due to extreme resource costs and weak client demand. The freed-up compute capacity was redirected toward coding tools and autonomous agents.
Huet noted that the Dots AI assistant is scheduled for broader distribution in Europe within weeks, pending the resolution of compliance reviews with regulators. Regarding regulatory barriers in the EU, enterprise accounts navigate compliance through bespoke legal agreements and data processing addenda. Meanwhile, retail Pro accounts face total restrictions while strict consumer protection rules undergo evaluation. Despite the upcoming rollout, personal Pro accounts in the European Economic Area, Switzerland, and the UK remain blocked, whereas selected enterprise clients have already gained access.
Management noted that enterprise customers show minimal demand for video sequence processing on par with Google's Gemini. This dynamic forced the company to prioritize single-frame analysis. Panelists confirmed that extreme resource requirements led to the termination of the Sora video generator in March, proving the economic unviability of heavy multimodal features under current compute constraints.
Platform Lock-In and the Agents API
By releasing products such as the Agents API and providing direct access to 1.2 billion weekly active ChatGPT users, OpenAI aims to anchor developers and enterprises firmly within its proprietary infrastructure. Because such agents operate in dedicated virtual environments and AI-generated code volumes surge rapidly, demand for traditional CPU compute grows in parallel. Consequently, a dedicated engineering team audits OpenAI's internal codebases weekly solely to minimize CPU load and cut infrastructure maintenance costs. The heavy computational overhead of autonomous loops strains server clusters and inflates operational expenditures per user request.
OpenAI aims to lock developers into its platform, not just its models, using managed services and distribution channels for long-term ecosystem retention.
This architecture mirrors the classic app store economy, where third-party developers build products while OpenAI controls the core infrastructure and distribution channels. Nearly one in five desktop users regularly executes skills for documents, spreadsheets, and presentations, demonstrating that programming utilities are rapidly absorbing standard office workflows.
Ecosystem Control Versus Open Runtimes
European developers operating under strict regulatory regimes face a strategic contradiction regarding vendor lock-in, particularly given the availability of alternative open runtime environments. By keeping the core Codex Harness runtime open, the company allows developers to inspect code and execute it independently rather than relying exclusively on managed services.
Nevertheless, the economic realities of competing on OpenAI's terms remain asymmetric. Market observations indicate that developers relying on platform algorithms and customer acquisition via ChatGPT will face severe commercial friction if they attempt to migrate ecosystems later, turning infrastructure dependency into a permanent balance sheet constraint.
Side by side
Compute costs are now shaping the product roadmap far more strictly than consumer marketing ambitions, as dropping the Sora video generator in March freed up hardware for practical coding tools. While Google still bets on video sequences with Gemini, OpenAI is cutting central processor loads just to keep developers locked inside its proprietary infrastructure. Abandoning resource-heavy media signals a pragmatic shift toward autonomous agent loops that generate measurable returns across enterprise workflows.