OpenAI is moving deeper into enterprise deployment through DeployCo, a subsidiary built around engineers who work directly inside large companies. The goal is not just to sell models or APIs, but to help businesses wire AI into real workflows, compliance processes, and monitoring systems.
That hands-on strategy comes as Codex adoption accelerates worldwide and as companies ask harder questions about cost, regulation, and return on investment. Arnaud Fournier, OpenAI’s deployment chief and CTO of the OpenAI Deployment Company since April 2026, frames the moment as a shift from AI experiments toward operational systems.
DeployCo is built for implementation, not just advice
Fournier co-founded OpenAI’s Forward Deployed Engineering team two years ago and was its first FDE. He later ran the function for EMEA and global verticals before taking the CTO role at DeployCo, known internally as DeployCo.
OpenAI unveiled the subsidiary in May alongside 19 private equity firms plus global systems integrators and consultancies. Its first acquisition is British consulting firm Tomoro, which brings roughly 150 forward deployed engineers and deployment specialists into DeployCo.
The European role is central to the plan. Fournier told THE DECODER, "This is going to be a global company with a key leadership presence in Europe." The team already has sites in Paris, London, and Munich, and it runs joint projects with German firms.
The company’s premise is direct: AI only creates value when it is embedded into the way a business actually works. Fournier’s view is that a model or API on its own is not enough. The technology has to fit into processes, remain compliant, and be observable after launch.
Customer work becomes a feedback loop
DeployCo’s engineers sit between customers on one side and OpenAI’s product and research teams on the other. Their work has two stated purposes: solve customer problems and understand what still breaks when advanced AI is used in real organizations.
Fournier is clear about one boundary. "We do not train on our customer data unless someone explicitly asks us to." When that happens, he says it becomes a research partnership, and those cases are very rare and require extensive regulatory review on both sides.
The feedback loop works in other ways. If deployed systems show that document understanding performs poorly, that weakness can be passed back to research teams, which then acquire data and improve the model. Fournier links that process to a BBVA solution that improved sharply from GPT-5.0 to 5.5.
Tooling needs also flow back into OpenAI. The source article describes how demand for multi-agent orchestration led to the open-source repository Swarm, which later became the Agent SDK.
The BBVA example shows the difference between automation and redesign. The bank initially wanted help automating credit document writing. OpenAI instead proposed a broader system for continuously assessing credit risk, giving the bank a more frequent view of exposure when events such as the war in Ukraine or an escalation in the Strait of Hormuz affect counterparties or portfolios.
Consultants remain close to the strategy
DeployCo does not exist apart from the consulting world. Fournier points to the Frontier Alliance ecosystem with Accenture, Capgemini, BCG, and McKinsey, describing a market where major consultancies have built large AI units but still struggle to keep up with the technology.
OpenAI’s role, in his telling, is to help those partners deliver the current state of the art, plus the tools and systems needed for clients. The same firms investing in DeployCo and supplying Alliance partners is presented as evidence that they want to put the model into practice with OpenAI.
One point remains less defined: the exact split between OpenAI’s own forward deployed engineers and consulting partners that may be pursuing similar client work. The article notes that this division of labor stays fuzzy in the conversation.
Europe is not slowing adoption, Fournier says
European AI adoption is often discussed through the lens of strict rules such as the EU AI Act and data privacy concerns. Fournier does not present those as major blockers from what he sees in the field.
He adds a caveat: "I'm not the biggest expert in regulatory items, but I speak to what I see in the field every day." From that vantage point, he says regulation is barely slowing companies down. OpenAI has introduced EU data residency and enterprise key management for European firms, after which the conversation turns quickly to outcomes.
France, Germany, and the UK are among OpenAI’s ten largest markets globally, according to Fournier. The German company Stadler, a maker of waste-sorting systems with more than 650 employees, is offered as one example. According to OpenAI’s case study, over 85 percent of its workforce uses ChatGPT actively every day.
Codex is also growing quickly. OpenAI shared that Codex has more than four million weekly users worldwide, after a fivefold jump within three months at over 70 percent monthly growth. Germany plays an especially large role: weekly active Codex users there have grown more than sevenfold since January 2026, while the key points in the source describe growth of 720 percent since January 2026.
- Germany leads Europe in weekly active Codex users.
- Germany ranks among the global top five.
- Germany sits in the global top three for both paid subscriptions and developers.
- For weekly active Codex users, Germany is among the top five markets.
The ROI question is still unresolved
As Codex and other agentic systems spread, cost is becoming harder for companies to ignore. These systems can work through tasks over many steps and use far more compute than a single chat call. The source also notes broader industry movement toward higher API fees and credit-based billing.
Fournier argues that the price of AI intelligence has dropped sharply, saying it has fallen a hundredfold over the past 18 months. At the same time, newer models like GPT-5.5 cost more because more complex compute is involved.
That creates a practical tension for buyers. AI may be becoming cheaper in one sense, while advanced systems still require companies to think carefully about usage, workflow design, and measurable value.
OpenAI does not offer a universal formula for calculating ROI across AI projects. That may be the most important enterprise takeaway: deployment is becoming more concrete, adoption is rising, and tools such as Codex are gaining users fast, but the value case still depends on the specific process being changed.