When Anthropic and OpenAI announced their PE-backed services ventures this month, markets reacted as if a predator had entered the ecosystem. IT services stocks fell sharply. The read: the firms that build the models are coming for the firms that deploy them. If that is the interpretation, it is exactly wrong.

This is not new territory. As Investec equity research noted in a recent sector note, Microsoft, SAP, Oracle, AWS, Salesforce, and Workday have all operated services arms alongside their core platforms for years, bridging the adoption gap, de-risking complex implementations, and deepening the partner ecosystem rather than displacing it. Did Microsoft’s consulting arm eliminate Accenture? Did SAP services replace Deloitte? No. It multiplied deployment demand and made the SI ecosystem’s pitch easier. The platform and the partner need each other.

What these ventures actually represent is an “admission”. Anthropic’s new services company and the OpenAI Deployment Company, whose first acquisition added approximately 150 engineers, are not being formed because model companies want to compete with system integrators. They are being formed because those companies have discovered, at the edge of real enterprise deployments, that building a capable model and deploying it as a working business system are two entirely different disciplines. They need the services layer. They are acknowledging it by investing in it.

Demonstrating the capability of a tool is not the same as taking accountability for driving an economic outcome. AI model companies are now discovering this in the field, and that discovery validates, rather than threatens, the firms built to carry that accountability.

Every enterprise CIO I speak with is impressed by what today’s models can do. The problem is not capability demonstration. The problem is everything that comes after the demonstration.

Large, regulated enterprises are not greenfields. They run on decades of accumulated architecture—legacy systems, proprietary data models, compliance workflows, and interdependent integrations that no foundation model has seen or can be prompted to understand. When AI generates code or decisions in these environments, the questions that matter are not just “does it work?” They are: how does it work? Can a regulator audit it? Is it maintainable by a team that did not build it? Is it explainable to a compliance function? Is it secure within a production stack that was never designed with AI in mind?

These are questions that require institutional knowledge, engineering rigour, and domain accountability, none of which can be scaled by deploying forward-deployed engineers into lighthouse engagements.

This is what the industry means by agency. The business definition, not the technical one. The ability to operate with authority inside a client’s environment, take accountability for the outcome, and govern the system front to back. The distance between a model’s capability and an enterprise’s governed, auditable, measurable business outcome is precisely the space that specialist services firms occupy. AI model companies are not entering that space to displace the specialists. They are entering it because they have finally seen how wide it is.

The CIO community grasps this intuitively. The consistent observation I hear: AI has become a hammer looking for a nail. Models are accessible, pilots are everywhere, budgets are committed. And almost none of it has structurally changed how the business operates at scale. The gap is not access to a capable model. The gap is a deployment framework that makes AI scalable, repeatable, secure, and governed, and that someone is willing to take accountability for.

That accountability is not a forward-deployed engineer embedded in a client team. It is years of accumulated enterprise context: knowing how a client’s legacy systems actually behave in production; understanding what a regulatory sign-off requires in practice; engineering AI-integrated outputs that are not just functional but explainable, maintainable, and secure within a brownfield operating environment. That is not built by a capital allocation decision. It is built over decades of institutional engagement, and it is exactly what the model companies are discovering they need.

The stated objective of these ventures—to accelerate model adoption through lighthouse deployments that create proof points for the broader partner ecosystem, is not a competitive posture. It is a go-to-market strategy that structurally depends on the SI ecosystem to execute at scale. Every major model provider has simultaneously announced deep strategic partnerships with the world’s largest system integrators: training hundreds of thousands of their practitioners, co-developing joint enterprise offerings, and embedding model capabilities into established delivery platforms. You do not build that partner network while planning to displace it. The lighthouse and the ecosystem are the same strategy, not competing ones.

The market sold IT services stocks because it assumed the model companies had decided that services were worth owning. The actual conclusion is different and more important. The market should have concluded that delivering enterprise AI outcomes requires the enterprise context, domain knowledge, accountability structures, and platform architecture that specialist firms have spent years building. For firms that genuinely own that layer, these announcements are not a threat.

They are public validation. The market panicked, it should have bought.

(The author is the chief executive officer and managing director, Mphasis. Views are personal.)