Anthropic announced a joint venture on May 4 with Goldman Sachs, Blackstone, and Hellman & Friedman to build "a new enterprise AI services company." The venture has secured $1.5 billion in capital commitments and will embed Anthropic engineers inside mid-sized businesses — many of them PE-owned — to deploy Claude Code and Anthropic's broader product suite as the operating layer of those companies.
The deal lands the same week as Anthropic's exclusive 300 MW compute agreement with SpaceX, a reported $30 billion run-rate revenue line, and an enterprise customer roster that now includes a meaningful chunk of the top tier of American finance.
I have spent a lot of time around enterprise consulting firms, both as a customer of one and as a vendor competing against several. So when I read the words "embed engineers inside mid-sized businesses to implement AI systems," what I see is not an AI venture. What I see is McKinsey with model weights.
The McKinsey model — and I use it as a category, not a critique — has been the highest-margin services business in the world for decades. The pattern: assemble a small team of expensive engineers, place them inside a customer's organization, they sit on the executive floor, and they build whatever the executive wanted built — sometimes a strategy, sometimes a transformation, sometimes a piece of software. The customer pays for the time. The customer pays a lot for the time. And because the engagement reshapes how the customer operates, the consulting firm acquires institutional knowledge that makes the next engagement, at the same customer or a similar one, more lucrative.
What's new in 2026 is that the people doing the embedded work are AI engineers, and what they're embedding is software that gets cheaper to operate every quarter while remaining licensed by their employer. The economic asymmetry is striking: the consulting firm captures both the engagement margin (people-time) and the software margin (Claude API spend) for the same customer, with the customer's own data forming the moat against substitution. That is not how Accenture works. That is how a company with the cost structure of a SaaS business and the pricing power of a Big Four consultancy works.
The financial sponsors here are doing something specific. Blackstone, Hellman, and Goldman aren't AI investors in any meaningful sense — they're owners of large rosters of mid-sized PE portfolio companies. Their incentive is to use Anthropic's product to compress operating costs and increase EBITDA multiples across their books. The JV gives them a captive AI-deployment shop with margin economics they understand from selling consulting to their own portcos for years. That's a coherent thesis, and it's also the kind of thesis that historically produces an oligopoly: three or four firms, mostly indistinguishable in marketing, with deep relationships to the top thirty-percent-by-revenue of mid-market American business.
So when people ask whether the AI cycle is a "bubble," I think they're asking the wrong question. The right question is what kind of business OpenAI and Anthropic are becoming. Neither of them is starting to look like a chip company, a software company, or an internet company. They are starting to look like the firms they originally claimed they would disrupt — consultancies, agency businesses, white-glove service firms that bill hourly and never give up the customer relationship.
If the analogy is right, the comparable name on the New York Stock Exchange isn't Nvidia. It's Accenture.
The straightforward counter-argument is that consulting is a low-margin business once you take out the very top firms, and Anthropic isn't going to look like McKinsey in five years because the engineer-embedded model is just the on-ramp for self-serve product. Under that view, the JV is a customer acquisition mechanism, not a business model — Anthropic uses Goldman and Blackstone to land into mid-cap PE portcos, the portcos run on Claude for two years, and then the renewal is API-based and the engineer-time goes away. That is also a coherent story. I don't know which one is right.
- See Anthropic's own framing of the venture
- The financial sponsors collectively hold over a thousand mid-sized PE portfolio companies — the addressable market for the JV's services is, in practice, captive
- OpenAI's structurally similar "The Development Company" raise this week, at $4B from 19 investors, suggests the labs are coordinating on this organizational shape
If you operate an enterprise SaaS business inside a vertical the AI labs are now embedding in — finance, ops, procurement, HR — assume your software-only competitive moat is now competing with someone else's software-plus-McKinsey moat. Plan accordingly.
— Hank