Anthropic は 15 億ドルを獲得。OpenAI は 40 億ドルを獲得。McKinsey は迂回されました。

Key Takeaways

• Same Day, Same Playbook: On May 4, 2026, Anthropic and OpenAI both announced private-equity-backed enterprise AI services ventures, with no overlapping investors.
• The $5.5 Billion Wedge: Anthropic’s venture is capitalized at $1.5 billion, with $300 million each from Anthropic, Blackstone, and Hellman & Friedman. OpenAI raised more than $4 billion at a $10 billion valuation for The Deployment Company.
• The 6-to-1 Ratio: For every dollar companies spend on software, they spend six on services. That is the exact gap the AI labs are now reaching for.
• Captive Distribution: OpenAI’s PE partners alone control more than 2,000 portfolio companies, all available as a pre-built customer pipeline.

A Coordinated Hour, Not a Coincidence

On May 4, 2026, two announcements crossed the wire within the same news cycle. OpenAI finalized a joint venture called The Deployment Company with TPG, Brookfield Asset Management, Advent, Bain Capital, SoftBank Group, Dragoneer Investment Group, and 13 other backers, raising more than $4 billion against a $10 billion valuation, with OpenAI keeping majority control. Anthropic announced a parallel firm with Blackstone, Hellman & Friedman, Goldman Sachs, Apollo Global Management, General Atlantic, Leonard Green, GIC, and Sequoia Capital, with $1.5 billion in committed capital.

Two AI labs. Two PE consortiums. Zero overlap on the investor list.

The framing in tech press (“Anthropic and OpenAI race for enterprise”) misses what just happened structurally. The alternative-asset industry just carved enterprise AI deployment in half. Blackstone’s portfolio companies will get Claude. TPG’s portfolio companies will get GPT. The capital is positioned, the customer lists are pre-baked, and the entity meant to broker that handshake, the management consultant, is no longer needed in the middle.

That is the move. Not a launch. A bypass.

The 6-to-1 Number That Explains Everything

Fortune buried the most important number in the entire story: “For every dollar companies spend on software, they spend six on services, a ratio that has made consulting a multitrillion-dollar industry and that AI-native firms are now positioning to disrupt.”

That ratio is the entire economic motivation. Anthropic’s API revenue and OpenAI’s API revenue are growing fast, but they are still capturing only the “1” in that 6-to-1 split. The “6” (the implementation work, the workflow redesign, the change-management consulting, the integration engineering) has historically gone to McKinsey, Bain, BCG, Accenture, Deloitte, IBM Consulting, and the broader services ecosystem.

That ecosystem has been busy converting itself into AI consulting at speed. McKinsey’s QuantumBlack, its dedicated AI unit, has 1,700 staff and accounts for an estimated 40% of the firm’s total revenue. Accenture reported $2.2 billion in Advanced AI bookings and $1.1 billion in Advanced AI revenue in its first quarter of fiscal 2026 alone, with bookings up 76% year-over-year and revenue up 120%. Accenture told investors it would stop reporting Advanced AI as a separate line item in future quarters because the category had become inseparable from the rest of the business.

Krishna Rao, Anthropic’s CFO, framed the strategy this way: “Enterprise demand for Claude is significantly outpacing any single delivery model.” The structural read: customers want Claude inside their workflows, and the existing channel (third-party consultants) appears to be the bottleneck the lab is now trying to route around.

So the labs went and built their own consultant. With other people’s money. Targeting other people’s customers.

The Deal Math

The two ventures are structured differently, and the differences are revealing.

Parameter	Anthropic JV	OpenAI Deployment Company
Total disclosed capital	$1.5 billion (includes Anthropic’s own $300M)	Up to $5.5 billion ($4B PE consortium + up to $1.5B from OpenAI)
Lab contribution	$300 million from Anthropic	Up to $1.5 billion from OpenAI ($500M at close, $1B optional)
Valuation	Not disclosed	$10 billion (excluding new raise)
Lab governance	Equal partner	Majority owner, super-voting shares
Return guarantee	None disclosed	17.5% annual return floor over 5 years
Major founding partners	Blackstone, H&F, Goldman Sachs ($300M / $300M / $150M)	TPG (anchor), Brookfield, Advent, Bain Capital, Goanna
Additional backers	Apollo, General Atlantic, Leonard Green, GIC, Sequoia	SoftBank, Dragoneer, and others
Total investors	9 named	19 total
Stated portfolio access	”Hundreds” of portfolio companies	More than 2,000 portfolio companies and clients

OpenAI’s structure is more aggressive on every axis: a larger PE raise, a named valuation, super-voting shares retained by OpenAI, a return floor of 17.5% per year over five years, and a larger captive customer base. OpenAI is also putting up to $1.5 billion of its own capital into the entity (a $500 million equity contribution at close with an option to add a further $1 billion). Anthropic’s deal looks more like an equal partnership, with $300 million each from the three principal players and Goldman in at $150 million. Both labs maintain operational tie-in to the venture; Anthropic specifies that “applied AI engineers from Anthropic collaborate with the firm’s engineering team on implementation and ongoing customer support.”

The combined committed capital is on the order of several billion dollars. The two ventures do not buy a meaningful share of the global management consulting market, which industry analysts size at several hundred billion dollars annually. They buy a wedge into the part of that market that grows fastest, AI implementation for mid-market firms, and they buy that wedge at a fraction of the cost of building distribution from scratch.

The Captive Customer Trick

Marc Nachmann at Goldman Sachs framed the venture as a way to “democratize access to forward-deployed engineers.” Jon Gray at Blackstone described “one of the most significant bottlenecks to enterprise AI adoption” as “the scarcity of engineers.”

These are accurate descriptions. They are also incomplete.

The forward-deployed-engineer model (embed a small team inside a customer’s operations to wire AI directly into their workflows) was popularized by Palantir over the last decade and is now the default playbook for AI services. It is also extremely expensive to scale. To staff one engagement at a community bank, the firm needs to acquire that bank as a customer. Customer acquisition for mid-market enterprise software runs hundreds of thousands of dollars per logo. Multiply by hundreds of bank logos, healthcare systems, manufacturers, and retailers, and the go-to-market cost dwarfs the headline capital raise.

PE partnership solves that line item.

Blackstone’s portfolio includes hundreds of mid-market companies across “healthcare, manufacturing, financial services, retail, real estate, infrastructure, and more.” OpenAI’s PE partners reach more than 2,000 portfolio companies. Each of those companies is owned by a fund manager whose returns improve every time one of those portcos cuts costs through AI deployment. The fund manager is, simultaneously, an investor in the AI deployment firm and the controlling shareholder of the customer. The pitch is not a sales pitch. It is a board memo.

A pure-play AI services startup would spend years getting in the door at 2,000 mid-market firms. A PE-funded AI services firm gets in the door the day the deal closes.

This is why every named PE investor on both deals is an alternative-asset manager with an enormous portfolio company footprint. Not a strategic. Not a corporate. The portfolio is the asset.

What the Mainstream Take Misses

The reflexive read is “Anthropic versus OpenAI in enterprise.” That is happening. It is also the smaller story.

The bigger story is what happens to the services layer when the model maker and the customer’s owner are aligned and the consultant is not.

Consider the existing chain. A mid-market manufacturer wants to deploy AI agents in its accounts payable workflow. Under the legacy model, the manufacturer hires Accenture or a regional implementation partner. That partner uses an underlying model from Anthropic or OpenAI under an API agreement. The partner takes the bulk of the project economics; the model maker gets the API line item. The PE owner of the manufacturer pays the bill out of the portfolio company’s operating budget and watches its NAV improve at the end of the quarter when the cost saves materialize.

In the new chain, the PE owner directs the same project to the JV it is also a partner in. The JV deploys Claude or GPT through the embedded-engineer engagement model both labs explicitly describe in their announcements. Revenue that an external implementation partner would otherwise have captured now flows through an entity the PE owner sits inside. The PE owner collects a share of the services revenue at the JV level alongside the cost savings at the portco level. The model maker collects the API revenue alongside a share of the JV’s services margin. The party most exposed in this rewiring is the third-party implementation partner that used to sit between them.

Carnegie did this in steel. Vertical integration consolidates the value chain when one player has captive demand and the technology to satisfy it. PE has captive demand in its portfolio companies. The AI labs have the technology. The JV is the new vertical layer that captures both.

The Counterargument: McKinsey Is Not Standing Still

The case against this article’s thesis is concrete and worth taking seriously.

The largest consulting firms have spent the last three years cannibalizing themselves to become AI deployers. McKinsey’s QuantumBlack went from a 45-person acquisition in 2015 to an estimated 40% of firm revenue. McKinsey CEO Bob Sternfels has said the firm runs 25,000 AI agents alongside its 40,000 human consultants, with a stated target to reach parity by the end of 2026. Accenture’s Advanced AI revenue grew 120% in its first quarter of fiscal 2026 and reached $1.1 billion in that single quarter. Total Accenture revenue in the same quarter was $18.7 billion.

Big consulting did not miss AI. It is the largest single buyer of frontier-model APIs and a vocal champion of the technology to its corporate clients. The firms have hired tens of thousands of AI engineers, restructured graduate hiring around AI roles, and built proprietary delivery platforms on top of OpenAI, Anthropic, and Google models. Across the Big Four, graduate hiring fell an estimated 44% year-on-year through the 2025 cycle as junior generalist roles were absorbed by AI tooling. The firms know exactly what is happening; they are running the same play.

The PE-backed JVs are not going to vaporize a half-trillion-dollar global management-consulting industry. McKinsey will keep its CEO-level relationships, its scenario-planning work, and its regulatory-strategy practice. Accenture will keep its multi-year integration contracts at Fortune 100 customers. The new ventures are not pitching for those.

What the JVs can take is the mid-market AI implementation work, the segment where consulting margins are richest and customer relationships are weakest. PE-owned mid-market companies are the softest target on the consulting menu. Their owners can simply redirect the budget. Even a small dent in the AI implementation revenue line at the Big Four would be material; a deeper one would reset the industry.

The Historical Rhyme

The 1990s outsourcing wave produced a similar pattern. EDS and IBM Global Services did not kill internal IT departments by being technically superior. They killed them by offering a captive bundle: hardware, integration, ongoing operations, and a single throat to choke, all priced as a multi-year contract that Wall Street rewarded. The customer’s CFO loved the predictability. The customer’s IT director was disintermediated.

What the AI labs and PE just built is the inverse bundle: model, deployment, and operating control all under one cap table, sold to a captive portfolio company instead of an arms-length enterprise customer. The customer’s CFO does not need to be sold; the customer’s CFO works for one of the deal’s investors.

The risk for the labs is that they get pulled into low-margin services work that pollutes the model-business income statement. OpenAI’s structural answer is super-voting shares plus a 17.5% annual return floor to its PE backers over a five-year window. That combination lets OpenAI keep strategic control while the financial sponsors take the venture’s economics as an income-oriented investment. Anthropic’s structure looks more genuinely shared, with the three principal partners each writing a $300 million check.

What to Watch Over the Next Four Quarters

A few specific tells will indicate whether the bypass is working.

First, watch Accenture’s reported AI bookings against the trend line. Accenture itself flagged that it will stop separately reporting Advanced AI revenue, which removes the public tracking signal at exactly the moment it would matter most. If the firm is suddenly less specific about mid-market AI growth in the second half of 2026, that is the canary.

Second, watch which named portfolio companies the JVs disclose as anchor customers. The Anthropic announcement promises engagements that “start with a small team working closely with the customer to understand where Claude can have the biggest impact.” The first wave of named customers will signal whether the JV is hitting its captive PE-portfolio base or struggling for outside deals.

Third, watch hiring. Forward-deployed-engineer models live or die on senior implementation talent. If Anthropic’s JV starts pulling McKinsey QuantumBlack senior partners and Accenture practice leads, the talent flow is the leading indicator of where the next wave of AI implementation revenue ends up.

Fourth, watch the smaller boutique AI consultancies, the ones that have raised tens of millions on the premise of being the model-agnostic Palantir for mid-market AI. They are the cleanest casualties if PE-owned mid-market customers get redirected to the labs’ own services arms.

The Bottom Line

The consulting industry’s AI strategy was to absorb AI into its own delivery model, taking a wedge of every Claude or GPT deployment by being the partner the customer hires to install it. That works only if the customer hires the consultant. PE just built a structure where its own portfolio companies do not have to.

The two AI labs did not announce competing products on May 4. They announced parallel forward-integration plays into the services layer, financed by the people who own the customer base. The unit economics, six dollars of services for every dollar of software, make the move obvious in hindsight. The simultaneity makes it official: this is the new shape of the enterprise AI value chain.

Anthropic wrote a $300 million check. OpenAI kept super-voting shares. PE handed over a customer list McKinsey took a century to build.

McKinsey will not vanish. But it will spend the next two years explaining to its mid-market clients why the consultant in the room and the controlling shareholder of the company are not the same person any more.