On 4 May, OpenAI and Anthropic announced enterprise services joint ventures on the same day.
OpenAI launched DeployCo. $4 billion raised from 19 investors, led by TPG, Brookfield, and Bain Capital, at a $10 billion valuation. The unusual detail is the deal structure: 17.5% guaranteed annual return over five years, with OpenAI subsidizing the downside. The lab is buying a captive enterprise customer base ahead of its IPO scheduled for autumn 2026.
Anthropic responded with $1.5 billion, with Blackstone, Goldman Sachs and Hellman & Friedman leading, and Apollo, General Atlantic, and Sequoia participating. Their own IPO is in preparation in parallel.
The model both labs landed on is identical: lab engineers embedded in the portfolio companies of their investors, redesigning workflows and integrating AI into core processes. This is not classic consulting. It is the model maker rewriting your operation with direct access to its own stack.
The interesting thing is not the money. $5.5 billion combined is large but not unprecedented for AI rounds. The interesting thing is what the money is paying for.
The 17.5% guarantee reveals the real move
A 17.5% annual return over five years is not a return on equity. It is a debt-like instrument structured as preferred paper. OpenAI is issuing structurally subordinated debt, in effect, in order to buy enterprise distribution before its IPO. The investors get a near-fixed-income product; OpenAI gets a captive customer base whose contracts can be shown to public-market investors as recurring revenue.
The label "venture round" obscures the substance. This is a customer acquisition cost, financed through a financial instrument, paid for by the lab itself. Anthropic's structure is different in form but identical in intent: the multi-bank syndicate (Blackstone, Goldman, H&F) is providing distribution into the PE-owned portfolio companies that those firms control. Apollo, General Atlantic and Sequoia bring more.
The labs need a particular kind of revenue ahead of their IPOs. Not API tokens — those are volatile, low-margin per unit, and increasingly commoditized. They need recurring services revenue that public-market investors recognize as a real business. The DeployCo and Anthropic JV revenue lines will show up as multi-year enterprise contracts. That is what gets priced into the IPO.
The 17.5% number is not generosity. It is the price the labs are willing to pay, in cost of capital, to convert the next twelve months of pre-IPO uncertainty into a guaranteed enterprise pipeline. It is also a marker of how badly they want this market.
The target is PE portfolio companies, not startups
Read the named targets carefully. Anthropic's announcement explicitly mentions community banks, mid-size manufacturers, and regional health systems. OpenAI's investor list — TPG, Brookfield, Bain — points at the same kind of company. These are not the customers that have been driving API revenue for the last two years. Those were tech startups. The new target is the operational backbone of the global mid-market, with 200 to 2,000 employees, owned by private equity, under cost-cut pressure from their funds.
In Spain, this profile maps to the businesses currently paying €1,500–2,500/day to Big Four consultancies for AI implementations that take six to twelve months. The labs are looking at the same total addressable market. When the lab itself shows up with engineers and preferred access to its own stack, the intermediary becomes a margin tax with no defensible function.
This is the substance of the move. The labs are not competing on model quality at this layer. They are competing on implementation depth: how fast can a workflow be redesigned, how completely can the AI be integrated into core operations, how much business risk can be transferred from the customer onto the lab itself. Consultancies cannot match the third one because they do not own the stack. The model maker can.
Labs are competing for a $500B/year services market
The total addressable market for professional services worldwide is around $500 billion annually. Most of it is captured by Big Four firms, advisory boutiques, and system integrators who get paid to bridge the gap between off-the-shelf software and a company's actual operations.
The labs do not need to win the whole market to win the strategic battle. A 20-30% slice on AI-implementation work alone is a new $100B+ business. And that slice is precisely the slice where consulting margins have been highest — because integration work has always been priced as a discretionary buy, not a commodity.
Sierra is the parallel signal worth noticing. Last week they closed $950 million at a $15 billion valuation, with 40%+ of the Fortune 50 already running their agents in production. Sierra is not a foundation model lab. It is the services layer on top of foundation models, productized. Its valuation has just doubled because the market is pricing in that the AI services layer is where the next wave of enterprise spend lives — and it is consolidating fast.
The labs have read the same signal. The difference is that the labs have something Sierra does not: direct access to the underlying model and its roadmap. That is the asymmetry the JVs are designed to exploit.
What this means for the consulting layer
The structural implication for the consulting world is uncomfortable. The most sophisticated AI consultancy in the world, without direct access to the lab developing the model, is fighting with a road map when the lab has access to aerial navigation charts. The asymmetry is permanent unless consultancies can build their own model relationships — and the JV structure makes those relationships exclusive by design.
Three scenarios from here.
Consultancies double down on what only humans do. Change management, organizational redesign, regulatory navigation. The parts of an AI implementation that are not about the model. This is defensible but compressed: the margin pool shrinks because the model integration was where the billable hours were.
Consultancies build deep partnerships with one lab at a time. Become the preferred implementer for OpenAI or for Anthropic, accepting the loss of vendor neutrality. The risk is becoming a single-vendor channel and having no leverage in the relationship.
Consultancies build their own models. Theoretical option. Capital requirements are extreme; foundation-model economics do not work below a certain scale. Not a real move for the existing players, possible for a new entrant funded by PE itself.
Most large consultancies will end up somewhere between option one and option two, with the margin compression that implies. The valuation rerating, when it comes, will be brutal for firms whose AI practice is the growth engine of the last three years.
What this looks like from the operator seat
For anyone running AI inside a mid-market company today, the calculation has shifted.
The cost of waiting six months has gone up. The labs are now competing on implementation timelines, which means the price of a good integration is dropping and the speed at which it can be done is increasing. Companies that signed €1.5M consulting contracts in 2024 to redesign a workflow can now get the same outcome in half the time, at a fraction of the cost, from a vendor with stronger access to the underlying model.
The cost of vendor lock-in is more nuanced. A DeployCo or Anthropic JV engagement embeds you deeply with one lab's stack. The integration depth that makes these arrangements valuable is the same depth that makes leaving expensive. The pricing power moves, eventually, from customer to vendor. Operators who go in with eyes open negotiate exit clauses; operators who go in chasing speed end up married.
The biggest unknown is the regulatory layer. The AI Act treats high-risk systems differently, and a lab-led implementation in a high-risk vertical — health, hiring, credit — will need to pass scrutiny that is currently being defined inside sandboxes like the one I run for Shakers. Whether DeployCo can take a community bank from pilot to production under AI Act constraints is an open question, and one the labs will hit harder than they expect over the next twelve months.
Closing
The labs are not entering the consulting business. They are eating a specific slice of it — the AI integration layer — and they have $5.5 billion to do it with. The middleman that survives this cycle is the one that owns something the labs cannot easily replicate: deep regulatory knowledge, local operational presence, sector-specific eval data. The middleman that does not survive is the one that has been billing for the integration work itself.
The consulting industry has seen technology disruptions before. SaaS compressed margins in software implementation. Cloud reshaped infrastructure consulting. AI is different in one important way: the technology vendor is not selling a tool that consultants then deploy. The technology vendor is selling the deployment itself.
This is the part most consulting partners are not pricing into next year's budget. They should.