The AI labs are becoming the consultants they promised not to be

In February, OpenAI's chief revenue officer made the company's position explicit as it recruited the consulting industry's biggest names as partners.

We do not want to build a model where we are doing the work. We want our customers to become self-sufficient. — Denise Dresser

The Frontier Alliance paired OpenAI with BCG, McKinsey, Accenture, and Capgemini to train their staff and help customers become "self-sufficient" ^[1]. By July, OpenAI had raised $4 billion for a Deployment Company that does the work itself — forward-deployed engineers embedded inside mid-sized enterprises ^[2]. Anthropic followed with a $1.5 billion venture, Ode, aimed at the same mid-market ^[3]. The labs that promised to stay out of the implementation business are now in it, and in direct competition with the firms they partnered with months earlier. The arithmetic behind the reversal is not subtle. Big Tech has committed $740 billion to AI infrastructure this year, a 69% jump from 2025, while an MIT study found 95% of businesses investing in AI have failed to make money on it ^[4]. The cash flow is showing the strain: Alphabet posted its first-ever negative free cash flow, minus $5.9 billion, and Meta's quarterly free cash flow fell 91% to $784 million ^[5]. The model layer that was supposed to pay for all of it is collapsing in price — OpenAI cut fees 80% in August, Anthropic released a top-tier model at half cost, both answering Chinese competitors like DeepSeek and Alibaba's Qwen ^[6][7]. So the labs are moving down the stack, in three distinct strokes. Meta launched Meta Compute in July to sell excess GPU capacity to outside customers, positioning itself against AWS, Azure, and the neoclouds ^[8]; CoreWeave, the compute landlord Meta once rented from, has lost half its value from the 52-week high, about $49 billion in market cap ^[9]. The embedded-engineer ventures put OpenAI and Anthropic inside client firms doing the integration work consultancies like TCS, Infosys, and Wipro used to sell ^[2]. And on April 24, all four labs — OpenAI, Microsoft, Google, Anthropic — launched enterprise agent platforms on the same day, selling automated workflows that compete directly with the services those consulting firms provide ^[10]. The hinge is the Frontier Alliance itself. In February the labs needed the consultants as the implementation layer; by summer they were the implementation layer. The promise not to compete was made when the labs believed model revenue would cover the bill. It was abandoned when it became clear it wouldn't. That doesn't mean the consultancies are finished. Wipro is pivoting to a platform-plus-services model with a $500 million AI investment and grew bookings 14% ^[11]. PwC's data shows the most AI-exposed firms adding headcount and productivity at once ^[12]. The U.S. added roughly 3 million white-collar jobs from late 2022 into early 2026 ^[13]. The collision is real, but it is not a clean sweep. The betrayal, if that's the word, is arithmetic. When the model layer could no longer carry the cost of $740 billion in infrastructure, the labs turned to the layer below it too — the compute, the engineers, the workflows that were supposed to belong to someone else.