The Rise of AI Holding Companies

We are witnessing the emergence of a new asset class. Why?

This asset class seeks to answer two fundamental questions in business: 1) How do you create value? And 2) How do you align incentives? Holding companies like Long Lake Holdings, Highline Power, and Thrive Holdings propose a new answer for the AI era. 

What is an AI Holding Company?

Long Lake is the best early example. Presently valued north of $12 billion and fresh off the $6.3 billion acquisition of Amex Global Business Travel, it’s easy to forget the holding company is only three years old. CEO Alexander Taubman (HBS ’14) started the company alongside CTO Rasmus Wissman and Varun Shenoy.

Long Lake acquires established services businesses as subsidiary operating companies and deploys Nexus, an AI platform developed at the Holding Company level, inside these businesses to create value. Its engineers embed deeply in the acquired operating companies, working alongside the people managing properties, providing tax services, designing buildings, and arranging corporate travel. Long Lake assumes the risk of transforming these businesses and shares in the upside.

Scope varies across different AI HoldCos. Highline concentrates on power services, while Thrive Holdings, founded by Joshua Kushner (HBS ‘11), creates industry focused platforms in areas such as accounting and IT services. Long Lake’s Nexus is designed to work across many different service industries, allowing them to move from HOA management services to include platforms for HR services, tax services, architecture services, and now travel services.

When I discovered AI HoldCos, despite having a significantly different business model, I immediately recognized some similarity to Palantir’s value creation approach. Forward Deployed Engineers (FDEs) embed with operators to build AI products closely aligned to their needs. Developers turn reusable solutions into platform features, creating technology leverage that accelerates the next deployment.

Palantir’s financials and operating results might explain why others seem to have taken notice. The company’s quarterly filings show that revenue has almost tripled in the last three years and continues to accelerate with annual revenue growth of 93% Y/Y in Q2 2026. At the same time, the company has improved adjusted operating margins from 34% to 62% over the same period, indicating accelerating and highly profitable growth. This helps to explain why Palantir’s enterprise value has skyrocketed from $35 billion to $500 billion over the same period.

It’s no surprise then that Palantir talent is viewed as maybe the hottest commodity when it comes to AI transformation. If you look at the firms deploying AI transformation strategies across all forms of ownership, you will find a diaspora of Palantir alumni. Long Lake’s CTO worked at Palantir. Thrive Capital spent significant time with Palantir while developing the Thrive Holdings concept; it hired former Palantir CIO Jim Siders as CEO of one of Thrive’s core platforms. Thrive has also hired many senior engineers from Palantir to fill key technical roles. A new PE firm, Aldaron Capital, recently announced an initial fund of $1 billion, with a long-tenured Palantir FDE as founding partner; the firm styles itself as a Transformation Equity firm. The CTO of Highline Power is a seasoned Palantir alumni FDE. OpenAI’s DeployCo recently acquired Northslope, a Palantir service partner with over 50 Palantir alumni FDEs. OpenAI DeployCo and Ode, Anthropic’s DeployCo, have both been in a hiring frenzy over former Palantir FDEs. The Head of Forward Deployed Engineering at Wonderful, a Europe focused DeployCo valued at over $5 billion, is also a seasoned Palantir alumni FDE. ElevenLabs, the voice AI company valued at $22 billion that recently moved into the enterprise agent deployment space, was founded by Palantir alum Mati Staniszewski. When it comes to AI transformation, there seems to be Palantir alumni in key roles everywhere you look.

The Difference Between AI DeployCos and AI HoldCos

So what is the difference between Palantir, which I would characterize as the original AI DeployCo, and the new branch of AI HoldCos? They provide different answers to the second business question: how do you align incentives? From my personal experience at Palantir, and talking to many friends who worked on different deployment teams, the primary blockers to value creation were frequently not the technology itself. The primary blockers were often key stakeholder buy-in, access to data systems, change management, changing daily operating behaviors of the worker, and attributing outcomes to the technology—essentially, aligning incentives to unblock the transformation.

The HoldCo model provides interesting answers to these blockers. AI HoldCos have ownership over the subsidiary operating companies with the governance rights required to overcome these barriers. A HoldCo can screen for management’s strategic alignment during the acquisition process, ensure engineers have access to the required systems, and underwrite an AI transformation that management may be unable or unwilling to attempt. 

The operating company and tech team can work together toward the same financial outcome, while the HoldCo warehouses the risk of the transformation. That does not mean adoption will be automatic. In fact, Long Lake CEO Alex Taubman explicitly rejects forcing employees to use their tools, though many of the common barriers to transformation can be overcome through a new ownership and governance structure.

Why This is Happening Now

Sequoia’s Pat Grady recently highlighted the growing AI diffusion gap: technology capability is accelerating faster than it can be absorbed into specific industries. As AI continues to accelerate, this means the gap between the value that could be created and the value actually being realized will continue to grow. This is another form of William Gibson’s famous saying that “the future is here; it’s just not evenly distributed.”

So what is causing this diffusion gap? Well, presently, very specific knowledge and expertise is required to get the most value out of AI in an enterprise context (see the Palantir note above).

This elevates the concept of relative potential impact. Two teams with access to the same model and capabilities can produce very different results. Software has always given a skilled person immense leverage, because a useful solution can be multiplied across many machines and many users. This remains true in an AI world, and we’re still on the cusp of a new technology trend where the specific knowledge to maximize impact is highly concentrated among the groups that have discovered the breakthrough. 

My thesis is that due to these dynamics, the range of relative potential impact is unusually wide today. Teams that possess the specific knowledge required to enact a successful transformation can create value in ways that others are simply incapable of. Under normal circumstances, one might recognize the relative potential impact of someone who is the best and someone who is average on a linear scale—for example, a cashier who performs twice as quickly as the average cashier. But in this scenario, the scale is logarithmic, and the teams that are the best can create impact that is magnitudes greater than the average.

From a financial perspective, if you look at a business deal through the lens of a simple returns bridge, value is generally captured or lost across three levers: 1) EBITDA growth, 2) multiple change, 3) cash flow generation. Cash flow generation has to do with existing cash flows, and multiple change is a proxy for the quality and reliability of cashflows going forward. As a result, it intuitively makes sense that AI transformation should impact the EBITDA growth lever most significantly (though there will be some residual impact on multiples as business model quality improves and cash flow generation as cash flow conversion improves).

I mention this because we have historically seen new investment platforms emerge when EBITDA growth was a primary value creation lever. When Mitt Romney (HBS ’74) and Bob White (HBS ’82) started Bain Capital in 1984, the founding team believed that their consulting and operating expertise could be applied with company ownership to create more value in the companies that they invest in. Additionally, founder Geoff Rehnert specifically recalled the idea that investing might allow the team to earn more from the business transformation than collecting consulting fees alone.

The same questions Bain Capital asked apply to AI transformation. How much value can be created? What barriers exist to prevent value creation? How much are customers willing to pay for this value creation? What is the best model for sharing in this value creation?

As you can see, the barriers to entry are much lower for a deployment company, but the tradeoff is less participation in the value creation and different binding constraints on scalability. One way to put this is: the barriers to entry are lower for deployment companies, while the barriers to value creation are much higher.

The tradeoffs can be roughly expressed:

Participation in value creation ≈ Businesses reached × Value created per business × Share captured – Costs

As a result, I anticipate a divergence going forward, where specific market spaces will highly favor deployment companies while other market spaces will significantly favor an ownership model. This distinction will be distinctly tied to market structures and the binding constraints for each approach.

Finally, there are questions around willingness-to-pay, the cost of transformation, and value-to-be-created. As the ratio between willingness-to-pay and value-to-be-created grows, the ownership model is further incentivized as share of participation becomes a more meaningful term in the equation above. As the cost of transformation grows, the ownership model is also incentivized—as the costs might present a prohibitive business risk for management to accept on their own. It is also prudent to compare the cost of transformation to a business’s existing free cash flows to understand if they have the ability to self-finance the transformation. There is also the gap between each party’s perception of how much value should be expected from the transformation. The larger the gap between the two parties, the further incentivized the ownership model is and vice versa.

Why a Holding Company vs. a Private Equity Fund?

There are many different structures to an ownership approach for transformation. As mentioned above, Aldaron Capital recently announced itself as a PE firm focused on AI transformation. Transparently, I do not attempt to have the answer to the question of holding companies as opposed to private equity funds—but I do have some intuition on the benefits of each structure.

In my mind, two significant distinctions between private equity funds and holding companies are the length of hold period, and the fluidity of being able to deploy resources across the portfolio. I am sure there are other important considerations, and I would welcome expert opinions on this.

An instructive approach has been comparing Danaher with Bain Capital, who both (coincidentally?) were founded in 1984. Both are businesses with models focused on improving the companies they invest in, but they took different ownership paths: investment partnerships at Bain Capital and a holding company at Danaher.

Taubman names Danaher as an inspiration for Long Lake. The legendary Danaher Business System provided Danaher a repeatable approach to growth, lean processes, and continuous operational improvement across acquired businesses. Long Lake believes AI expertise and a reusable technology platform will provide a similar source of advantage across the HoldCo. Long Lake also styles itself as never wanting to sell the companies that it acquires, allowing the operating assets to continue compounding and benefiting from the proprietary technology platform that’s shared from the HoldCo. One could imagine the complexity of selling a company when their technology operating system is owned by the HoldCo. While the complexity could be resolved, there would have to be a good reason for inducing it upon the model.

Another example might be Berkshire Hathaway, with their principle of central capital allocation and decentralized operations. Available cash from one business can fund better opportunities elsewhere, which significantly lowers the cost of capital across the entire HoldCo. In the AI HoldCo model, in addition to capital the FDE talent could be considered an extremely valuable resource that could be centrally allocated by the HoldCo to the location of greatest potential impact.

A holding company with lasting capital can preserve the technology platform, retain deployment experience, and reinvest across businesses without a scheduled fund wind down, which can limit long-term strategic focus. These are serious and plausible advantages for the holding company model. But private equity already has control rights, operating teams, long hold strategies with permanent capital vehicles, and an advantaged fundraising model that LPs are very comfortable with.

Which structure will prove better for supporting AI transformation remains an open question. The clearer point is that the diffusion gap is significant and growing. So much so that the ownership model has become quite attractive in this moment for specific industries. The teams that prove they can create the most value will be able to finance their transformations, assume the risk, and share in the upside.

Eric Menser (HBS ‘27) hails from Stafford, Virginia. He graduated with a B.S. from the Air Force Academy. He worked at The White House under the CEA, did 5 years as an officer in the Marine Corps working with the joint special operations community in the Pacific, then worked in the Office of the CEO at Palantir Technologies when the company went from $45 billion to $450 billion TEV. Eric has been an investor at Journey, a technology private equity firm with an AI Holding Company thesis. Journey was a founding partner to Highline Power.