The next AI battleground: Is AI Bringing Services Back as the Stickiness Layer

For a long time, the AI race looked like a model race. Who has the best model? Who has the biggest context window? Who can write better code, reason better, or answer more accurately? But the latest moves from Anthropic and OpenAI suggest the race is shifting. The hardest part is to get those models to actually work inside enterprises.

Anthropic has announced a new AI-native enterprise services company with Blackstone, Hellman & Friedman, and Goldman Sachs to help companies bring Claude into core business operations. Reuters has also reported that OpenAI and Anthropic are in talks to acquire AI services firms to build more deployment capacity.

To unpack this, I’m using a simple SWOT lens. It is not meant to be a perfect framework, but it is a useful way to look at the strengths, weaknesses, opportunities, and threats behind this new services push.

Strengths

• The obvious strength is brand. OpenAI and Anthropic are among the most recognized names in AI, and that matters when enterprises are deciding whose technology to bet on
• They also have deep investor backing, which gives them room to experiment
• They have technical credibility. They are not coming to market as generic implementation or consulting firms; they are coming with powerful model capabilities and the ability to build more customized agents around enterprise workflows.

Weaknesses

• Services is a very different business from software. It is people-heavy, delivery-heavy, relationship-heavy, and often lower margin. Running a services model also requires domain knowledge, change management expertise, program governance, and the ability to manage messy enterprise realities over time
• There is also an architecture concern. If model companies lead deployment, they will naturally favour their own models. That may be acceptable for some buyers, but many enterprises will prefer model-flexible architectures
• Enterprise AI is not only about probabilistic models. It needs deterministic workflows, integrations, auditability, controls, exception handling, and governance around the model. This is less of an external threat and more of an internal capability gap they will need to close.

Opportunities

• The opportunity is real and big. Many enterprises are stuck between pilots and production. They have experimented with AI, but turning that into repeatable business value is proving harder
• The mid-market may be especially attractive. Many mid-sized companies have been underserved by large consulting firms, and some have less legacy baggage than larger enterprises. They may be more willing to adopt AI-native ways of working if someone can help them move quickly
• Deployment also gives model companies something valuable: direct product and workflow insight. I would be careful saying this gives them enterprise data to improve their models broadly, because data usage boundaries will be sensitive. But where permitted, deployment work can certainly improve enterprise-specific agents and help these labs understand where their models succeed or fail in real business processes.

Threats

• The biggest threat is differentiation. Tech vendors already have professional services arms. Service providers already have client relationships. The question enterprises will ask is simple: Do I trust an AI lab to run transformation, or do I trust it to provide the model?
• There is also a coopetition problem. OpenAI and Anthropic already work with hyperscalers, platforms, and service providers. If they push too far into independent deployment, they may create friction with the same ecosystem players they need for scale
• They also risk becoming too broad. If they try to solve every industry problem directly, they could end up as generalists in a market that still rewards domain depth.

Implications

For service providers, this is a warning shot. The right response might be to double down on domain depth, open architecture, strengthening existing relationships, governance, integration, and measurable outcomes. Mid-market clients, in particular, may become more contested if AI labs offer a faster and more focused deployment path.

For enterprises, the good news is more choice. The harder part is evaluation. Buyers will need to assess not only model quality, but delivery credibility, architecture neutrality, data boundaries, long-term support, and whether the partner understands the business problem deeply enough.

My bottom line: Anthropic and OpenAI are not necessarily trying to become traditional service providers. They are trying to solve the deployment gap that stands between model capability and enterprise adoption. If they succeed, their models become stickier and harder to replace. If they fail, the reliance on traditional service providers is further reinstated.