Your CIO opens the all-hands with a slide called "AI Transformation Roadmap." It has four swimlanes, six workstreams, and a Gantt chart that extends past your mortgage. Three months in, the only artifact that ships on schedule is the invoice from Accenture or Deloitte. Nobody on the board finds this suspicious. You probably shouldn't either — because this is the business model now.

OpenAI and Anthropic are the names on every headline. But if you trace where your company's AI budget actually routes, it doesn't land at a model lab. It lands at a systems integrator — a consulting firm that installs software for a living. And the margin, it turns out, is not in the model.

On April 13, 2026, analyst firm Horses for Sources published a synthesis with a blunt title: "How Anthropic is Devouring IT Services." The numbers are structural, not cyclical. Across the Big 4 and strategy firms, more than one million practitioners are now formally committed to Claude — Anthropic's LLM (large language model — the brain behind a chatbot like ChatGPT or Claude). Deloitte alone has 470,000 employees on the platform. Cognizant has 350,000. Accenture has trained 30,000 and staffs 85,000 AI professionals, per its Q2 FY2026 earnings on March 19, which also logged $2.2B in Gen AI bookings in a single quarter. Anthropic's ARR (annual recurring revenue — the subscription run rate) went from $9B at end of 2025 to $30B by this April.

Here's the flywheel. Model labs ship deliberately rich platforms — Managed Agents, Agent Engines, Codex SDK, MCP (Model Context Protocol — a universal plug standard for AI tools, like USB for data). Non-technical buyers cannot deploy them alone. The platforms need sandbox ops, IAM wiring (IAM — identity and access management, the corporate permissions layer), change management, and legal cover. So consultants embed 40 to 200 bodies per Fortune 500 client at $400–$1,500 per hour to translate the complexity. Labs then cite the named customer — Notion, Rakuten, JPMorgan — as product-market fit. Everybody wins, except procurement.

Anthropic made the arrangement official on March 12, 2026, launching the Claude Partner Network with a $100M commitment. Steve Corfield, Anthropic's Head of Global Business Development, called it the most committed partner ecosystem in AI. OpenAI did the same thing on February 23 with its "Frontier Alliances" — formal multi-year tie-ups with McKinsey, BCG, Accenture, and Capgemini.

Here is the quiet part. Model access is the commodity. Professional services are the margin. Enterprise AI in April 2026 looks structurally identical to SAP implementations in 1998: the software vendor takes the credit, the systems integrator takes the check. The "transformation engagement" calcifies into a five-year retainer. Your prompts and playbooks don't transfer when you want to switch labs. And the labs quietly prefer this arrangement — because it masks how hard their own platforms are to adopt without a six-figure SI partner.

If your 2026 AI spend breaks down to roughly 80% services and 20% tools (directional, not audited), you are funding a consulting business with an AI coat of paint. The fix is boring and effective: an internal AI platform team of three competent staff engineers, vendor-neutral procurement criteria, and a standing refusal to pay for slideware that MCP plus a weekend can replace.

The model wars were the distraction. The real AI industry in 2026 is the services layer stacked on top of cheap inference. Calm systems beat expensive panic. 🫶 ⚙️