Skip to content

OpenAI Closed a $10 Billion Deal With TPG. Anthropic Closed a $1.5 Billion Deal With Blackstone the Same Afternoon.

DS
LDS Team
Let's Data Science
10 min
OpenAI's new firm, The Deployment Company, locked in a 17.5% guaranteed annual return for its private equity backers over five years. Anthropic raised $1.5 billion alongside Blackstone, Hellman & Friedman, and Goldman Sachs to embed engineers inside PE-owned portfolio companies. Both deals target the same opponent: the consulting industry.

On Monday afternoon in Manhattan, two of the largest AI labs in the world filed press releases that read like they had been written by the same lawyer. Both announced that they were starting consulting firms. Both said the new firms would embed engineers inside Fortune 500 companies and PE portfolios to build agents on top of their flagship models. Both named anchor partners that, taken together, control more deployable corporate capital than any other constellation of investors on Wall Street.

The releases hit within hours of each other. Bloomberg posted the OpenAI story shortly after lunch. By 3 p.m., Blackstone's communications team had pushed Anthropic's announcement to the wires. The Anthropic deal had been telegraphed in a Wall Street Journal report on Sunday night, but the OpenAI deal was new. The pairing was not coincidence.

What the two filings made clear is that the next phase of frontier AI is not the model. It is the integration. And the integration belongs, increasingly, to private equity.

OpenAI Built a Deployment Arm and Promised Its Backers 17.5%

OpenAI's new entity is called The Deployment Company, registered in Delaware as a limited liability corporation. The vehicle carries a $10 billion valuation before any new capital was committed.

According to Bloomberg, 19 investors have put up roughly four billion dollars in fresh capital. The anchor investors are TPG, Brookfield Asset Management, Bain Capital, Advent, Dragoneer, and SoftBank.

The most unusual term in the deal is the return floor. OpenAI has guaranteed its private equity backers an annualized return of 17.5% over a five-year hold period. By any normal venture standard, this is not how operating partners structure equity. Private equity vehicles do not typically receive an explicit return commitment from the company they are funding. What the structure looks like, in practice, is closer to a high-yield credit instrument with equity upside attached.

The chief executive of the entity is Brad Lightcap, until recently OpenAI's chief operating officer. OpenAI itself contributed an initial $500 million of equity. There is an option to deploy a further one billion dollars later, bringing the total potential lab commitment to roughly 1.5 billion. The PE consortium fills the rest. OpenAI retains majority control through super-voting shares and operational oversight of strategic direction.

What the new company will actually do is embed teams of OpenAI engineers directly inside client organizations to redesign workflows, build custom agents, and integrate ChatGPT and Codex into core processes. The model is conspicuously similar to Palantir's forward-deployed-engineer playbook, the one that turned Palantir into a profitable company after a decade of skepticism. According to Bloomberg, most of the capital raised will be used to acquire engineering services and consulting firms rather than to build the workforce from scratch.

The PE partners' attached portfolios add up to more than 2,000 client companies. That is the asset The Deployment Company is buying access to. Its first sales pipeline is already populated.

Anthropic Wrote a Smaller Check and Brought in Goldman Sachs

The Anthropic version of the same play is structured differently. The vehicle is unnamed at filing. The total commitment is $1.5 billion.

The three principal anchors are Anthropic, Blackstone, and Hellman & Friedman. Each is putting in roughly three hundred million dollars. Goldman Sachs is the founding bank investor at about one hundred fifty million. The rest is filled by General Atlantic, Leonard Green, Apollo Global Management, GIC, and Sequoia Capital.

The press release language is identical in spirit to OpenAI's: embed engineers inside PE-owned companies, redesign workflows, ship Claude into core operations. The initial sector targets named in the Blackstone release are healthcare, manufacturing, financial services, retail, and real estate. The vehicle does not have a guaranteed return floor.

What the Anthropic deal trades for the smaller check is named relationships. Blackstone manages roughly $1.2 trillion in portfolio company assets. Hellman & Friedman holds about one hundred twenty billion more. Goldman Sachs gets a seat at the operating table for an AI deployment vehicle that will pitch Goldman's middle-market clients directly.

For Anthropic, that is the audience it has been trying to reach since its Series G priced at $380 billion in February and its run rate cleared thirty billion dollars eight weeks later.

Fortune's reporting framed the Anthropic deal as a direct shot at Big Consulting. The named target list inside the partnership documents reportedly includes McKinsey, Bain & Company, and BCG. The pitch to a PE-owned mid-market manufacturer is no longer "hire McKinsey to map your AI strategy and we'll come back in nine months." It is "embed five of our engineers into your operations team for three quarters and we will ship working agents into production."

Why the Same Day, and Why Now

The timing of the dual announcement was not random. Both labs are negotiating with the same consortium of capital sources and the same set of target customers. Both face the same structural problem. Their models are improving on benchmarks faster than their customers can put those models into production. The gap between what Claude Opus 4.7 or GPT-5.5 can do on demand and what a typical insurance underwriter actually has access to inside her workflow is the gap that consulting firms have been pricing for the last eighteen months. Each lab has now decided to capture that fee directly.

The structure also solves a different problem: revenue smoothing. Anthropic's tripled revenue in 90 days and OpenAI's $122 billion fundraise in March both leaned on token consumption growth that swings with model launches. A services arm with embedded engineers, multi-year statements of work, and PE-backed customer pipelines smooths that revenue line. It is less volatile than API consumption. It also pays consultants by the hour, which compounds.

A third motivation is competitive. OpenAI Codex and Claude Code are competing for the same agentic coding budget. McKinsey, Accenture, and Deloitte have been playing both sides, building practices around each platform without committing to either. Both labs are now building their own consulting practices to take that work in-house and steer customers toward their own stack.

The Comparison Table

TermOpenAI: The Deployment CompanyAnthropic: Unnamed JV
Vehicle valuation$10 billion$1.5 billion total commit
New capital raised$4 billion+ from 19 investors$1.5 billion
Anchor investorsTPG, Brookfield, Bain, Advent, Dragoneer, SoftBankBlackstone, Hellman & Friedman, Goldman Sachs
Founding lab equity$500 million initial, $1B optionAnthropic: ~$300 million
Return guarantee17.5% annual, 5-year holdNone disclosed
Operating leadershipBrad Lightcap (former OpenAI COO)Not yet announced
Initial customer pool2,000+ PE portfolio companiesPE portfolios across healthcare, manufacturing, FS, retail, real estate
Acquisition strategyBuy consulting and services firmsBuild engineering teams
Disclosure venueBloomberg, May 4WSJ (May 3) and Blackstone press release (May 4)

What Practitioners Should Read Into This

For data scientists and ML engineers working inside enterprise companies, the practical effect arrives in stages. The first stage is procurement. PE-backed companies will be pitched the appropriate JV by their owners, often as a default option. The second stage is workflow redesign. Embedded engineers will arrive with opinions about how processes should be rebuilt around agents. The third stage is hiring. The JVs will absorb meaningful headcount from the consulting market and from in-house data teams whose work overlaps with what the embedded engineers do.

The PE structure also changes the political economy of AI inside enterprises. In a typical buyer relationship, the vendor sells software and the customer chooses how to deploy it. In this structure, the vendor's investor owns the customer. A Blackstone portfolio company hesitating on Claude adoption now has a board member with $300 million invested in the joint venture telling it to move faster.

There is also the question of intellectual property. When OpenAI engineers redesign a financial services firm's underwriting flow and ship custom agents on top of GPT-5.5, the resulting workflows live on OpenAI's stack. The same is true on the Anthropic side. The customer is rebuilding around a model it does not control, with engineers it did not hire, in service of a return floor it did not negotiate.

The Other Side: Critics Are Skeptical of the Structure

Not everyone reads the deals the same way. Several private equity analysts noted to Bloomberg that the 17.5% guarantee on OpenAI's vehicle resembles a fixed-income product more than an equity participation. If that is the right read, the deal becomes a creative way for OpenAI to access PE balance sheets without diluting itself further at the parent level. The trade-off, those analysts said, is that OpenAI is now effectively guaranteeing returns on capital that may not be there if AI deployment cycles compress or stall.

A second concern, raised in Fortune's coverage, is whether McKinsey, Bain, and BCG will move to defensive postures. The Big Three consulting firms have lost AI deployment revenue to specialists for the last two years. A frontal attack from the model providers themselves, with PE-owned customer lists attached, is materially worse than the previous threat. McKinsey has already announced two waves of internal restructuring in 2026. Whether the firm responds with its own AI partnership or defends through pricing remains open.

The third skeptical line came from Anthropic's own customer base. JPMorgan, Bank of America, and several of the larger asset managers have been building internal AI teams precisely so they would not have to outsource integration to vendors. The JV pitch competes directly with that strategy. Some institutional buyers will treat the new vehicles as a faster path to deployment. Others will treat them as a way for vendors to capture work the buyers were planning to do themselves.

The Bottom Line

OpenAI raised more than four billion dollars on Monday and promised its investors 17.5% annually for the privilege of helping it sell into 2,000 portfolio companies. Anthropic raised $1.5 billion the same afternoon and brought in Goldman Sachs and Blackstone to do roughly the same thing inside a smaller portfolio. The two deals were not coordinated. They did not need to be. The market for AI integration was always going to be too large for one frontier lab to capture, and the labs were always going to fight for it before the customers had finished writing their RFPs.

The longer-term question is what happens to the consulting industry that has been making the integration trade for the last eighteen months. McKinsey reads as the most exposed. Accenture is closer to the engineering layer and may survive by partnering. Deloitte and KPMG sit between. None of them has a balance sheet that competes with TPG's, and none has a primary product that competes with Claude or ChatGPT. The new vehicles do not yet have customers. By the end of Q3, they will.

Brad Lightcap, who built OpenAI's go-to-market team from fifty people to more than seven hundred over the past eighteen months, has reframed his entire job inside OpenAI around this transition. "We're four seconds in this entire shift," he told Fortune last fall. The Deployment Company is what four seconds in looks like when the company doing the building has $4 billion of PE money lined up behind it. The fee structure now follows the engineer, not the project.

Sources

Practice with real FinTech & Trading data

90 SQL & Python problems · 15 industry datasets

250 free problems · No credit card

See all FinTech & Trading problems
Free Career Roadmaps8 PATHS

Step-by-step roadmaps from zero to job-ready — curated courses, salary data, and the exact learning order that gets you hired.

Explore all career paths