OpenAI and Anthropic launch dedicated AI services firms to speed enterprise agent deployment

OpenAI and Anthropic have each formed separate services companies—OpenAI's 'DeployCo' backed by Bain, Capgemini, and McKinsey, and Anthropic's tailored enterprise deployment arm—to accelerate the adoption of agentic AI in large organizations. These entities field 'Forward Deployed Engineer' teams that embed directly with enterprise clients to integrate frontier models with messy legacy systems. The model mirrors what cloud providers like AWS did with ProServe after their platforms matured: a high-touch services layer that captures deployment data, generates high-margin revenue (30-50% margins typical for such firms), and deepens the dependency on the underlying platform. OpenAI's DeployCo has reportedly raised $200M from its consulting partners, while Anthropic's arm is bootstrapped from its $7.5B raise, aiming for a 'land and expand' approach with Fortune 500 pilots.

The services arms position OpenAI and Anthropic against established enterprise AI providers like Microsoft, which bundles Azure AI with its consulting arm, and Google Cloud, which offers Vertex AI Agent Builder alongside professional services. But the key differentiator is that OpenAI and Anthropic are not offering a cloud platform—they are offering model-agnostic integration into whatever infrastructure the client already has. This is a deliberate strategy to avoid competing with cloud giants on infrastructure and instead capture value at the application layer. Historically, AI labs have followed this pattern: DeepMind created DeepMind Applied in 2016, Cohere set up a solutions architecture team in 2023. Yet the scale of capital and the explicit partnership with top-tier consultancies (Bain, Capgemini, McKinsey) signals a more aggressive push to own the enterprise relationship from the ground up.

The obvious narrative is that services arms accelerate enterprise deployment and generate lucrative revenue. But the hidden implications are more consequential. First, 'Forward Deployed Engineer' talent is drawn from the same pool that improves the core models—creating a talent drain that could slow research progress at a critical moment. Second, these arms may cannibalize existing systems integrator partnerships (Accenture, Deloitte), risking ecosystem friction. Third, deep integration with enterprise data creates security and compliance surface area; a breach at a services deployment could expose sensitive client workflows and erode trust. Most critically, token cost volatility could lead enterprises to prefer on-premise or open-source alternatives after initial lock-in—defeating the purpose of the services hook. Ben Thompson has noted the risk of diverting engineering talent from core model work, while Sarah Wang of a16z highlighted vendor lock-in as a real concern for enterprises that should demand multi-model strategies. The enterprise AI agent market is projected to reach $30B by 2027 (IDC), and these services arms are the vanguard of that growth. But they also represent a bet that integration depth matters more than model performance—a bet that may prove right until it doesn't.