Department Store (1958), Michael Ramus
Whilst the traditional Berkshire Hathaway story begins with the formation of the Buffett Partnership in 1962 as a result of the merger of Warren Buffett’s interest in 11 partnerships, there are two crucial junctures in the Buffett-Berkshire story that may just as well mark its birth as we know it today.
The first is in 1951, when a 20-year-old Buffett discovered that Benjamin Graham (at the time his mentor, not yet his colleague or employer) was the Chairman of GEICO. Buffett got the train to GEICO’s headquarters in Washington, D.C. on a Saturday and had a janitor permit him access to the building. Whilst there, he would serendipitously meet GEICO VP Lorimer Davidson, and the two wound up having a half-day-long conversation discussing the ins and outs of the insurance business.
This conversation would culminate first in Berkshire’s acquisition of National Indemnity Co for US$8.6mm in 1967. It was here that Berkshire first began its forays into the delights of the high-float business. It’s not impossible that it was from this conversation with Davidson that Buffett was introduced to (or at least sold on the appeal of) the substantial duration arbitrage that insurance can operate with.
National Indemnity collects all of its premiums upfront, with any potential liability on this occurring months or years later. As the premiums from customers rolled in, Berkshire had time to reinvest the cash before any payouts had to be made. Since the premiums were already priced to cover any claims that may arise, an insurance firm that placed any faith in its actuarial department’s ability to price this risk could invest a growing pool of other people’s money at close to zero cost of capital.
25 years after his conversation with Davidson, Buffett would finally establish a significant position in GEICO after the insurer underestimated claims losses driven by inflation. 20 years after that, Berkshire would take full control of GEICO.
US$1,000 invested in Berkshire Hathaway in 1964 would have yielded the investor $55mm by the end of 2024. Comparing this to the ~$390,000 that the same investment would have returned if invested in the S&P 500, the attraction of the business model is undeniable (to understate it a lot).
As the Berkshire asset base has grown, however, the arbitrage it can reap on its float has eroded somewhat. Berkshire stock has underperformed the index over each of the past three, five and ten years.
Though it has made some prescient investments in the technology sector (see: Snowflake, Apple), Berkshire’s original edge is a relic of the times it was born in. The age in which we live now offers a range of new float-like structural opportunities, new sellers, new operating models and new capital structures that can be used to repeat the Berkshire playbook .
The changes that have occurred in the past half-century seem to have culminated in this point in time now, where the advent of AI, the tailing off of recent generations of software companies and vigorous experimentation with novel capital formation techniques have led to the revisiting of Berkshire’s holding company model in an array of different manifestations.
This piece seeks to tease out the unique operating advantages of each of these early players, understand the operating floats they are trying to establish and establish a gauge on where the models might logically be headed as we enter a new Industrial Age.
What makes a Holding Company (HoldCo) a Holding Company?
As Berkshire was before them, the new generation of HoldCos are acquirers first and foremost, at least in the short term, as these models get up and running. HoldCos differ from other forms of mass or repeat acquirers on a range of fronts, however.
The first is the time horizon. Where a private equity firm buys to sell and return capital to investors within the window of a fund’s lifetime, the HoldCo is defined by its intent to operate and generate long-term cash flows from the businesses it owns.
The second is the way that capital moves in the layers below the HoldCo. The HoldCo subsidiaries can send cash up to the HoldCo, which then decides where to reinvest that cash. Again, not a possibility under the private equity model.
The last of the most salient differentiators is the capital structure. HoldCos typically invest off of their own balance sheet rather than making use of client money. How they procure these balance sheet funds is a design choice for the HoldCo – the cash for further acquisitions can be generated through any mix of float, other subsidiary cash flows, shareholders’ equity or debt. The distinction between the debt that a HoldCo takes on and the leverage that a Private Equity firm uses is that the HoldCo borrows against the cash flows of its full suite of subsidiaries in order to redirect those funds as it chooses, whereas the Private Equity firm loads debt onto specific companies in the portfolio.
The astute reader might notice that all of the above don’t uniquely delineate the HoldCo from conglomerates or roll-ups. Whilst both of these share all of the same characteristics that have been outlined above, both have their own minor distinctions that I will make for the purposes of this essay.
In the case of roll-ups, this is that they typically operate within one vertical (e.g., rolling up a range of disparate lawnmowing services in Connecticut in order to improve operational efficiencies and pricing power at the group level).
It is worth noting that HoldCos can still retain roll-up style structures under their umbrella – Long Lake is an example today of how the new-age HoldCo can mix both broad exposure across a portfolio with some concentrated exposure into certain industries where they believe they have significant advantage. Berkshire did this as well, with its ownership of BH Energy and BH Finance among others as subsidiaries (see below the example in the case of Home Services of America’s existence as a subsidiary to Berkshire Hathaway).
Source: TMCNet
In the case of conglomerates, I distinguish them from these HoldCos mainly based on connotation – a lot of the powerful conglomerates of the 1960s (e.g., Honeywell, General Electric, Toshiba) would often buy companies based on size in order to expand their sphere of influence and in doing so, hopefully jack up share prices. All of the examples provided above ended up unwinding into disparate entities for particular industries (e.g. Energy, Aerospace, Medical, you name it). It is worth noting here that many of the leaders of the HoldCos of today do repeatedly reference Teledyne and its founder Henry Singleton in podcast interviews as substantial influences. It may well be that the inevitable path of the successful HoldCo is to eventually morph into one of these conglomerates and spin off its different business arms to retain the advantages of operating smaller.
To finish this section, one way to think of the HoldCo, as it will be defined throughout this essay, is as a department store owner that owns all of its tenants. The real estate and building itself in this analogy act as the operating centre of the HoldCo, whilst its owned-and-operated tenants sell their wares to a variety of customers and pass these revenues right back to the HoldCo.
Today’s Players – An Introduction
Note: The below is not a comprehensive list of all the companies in the market, but a representation of some of those that are receiving the most attention as a starting point.
The new generation of HoldCos covered in this piece all share the characteristics outlined in the previous section, but maintain discretion over a few differentiating factors.
These can be briefly summarised as (1) business/vertical selection, (2) capital sources, (3) operating philosophy (i.e. hands-on versus hands-off, management changes, headcount changes) and (4) core value accretion methodology.
A brief overview below:
Inside the Operating Machines
While these AI / internet-age Berkshire successors all pitch themselves on differentiated operating USPs (as exhibited above), there are a few core tenets that some or most of these firms share.
The first is their focus on utilising the centre of the HoldCo as a product team rather than exclusively as an investment firm. Where savvy reinvestment of capital from subsidiaries was key to the long-term success and growth of Berkshire, the likes of Bending Spoons, Long Lake and Ithaca all make use of embedded staff to engineer, design and re-price the products that they have acquired. This means that rather than having to operate as pure value investors, they can create fundamental changes in the products and (to some degree) the business models themselves.
Aside from these fixes improving the value of these businesses, the promise of being able to incite meaningful change on well-known businesses makes these firms a honeypot for talent. This is not dissimilar to the rationale for why consulting firms have been able to attract and retain talented people in the past, but now with a more technological bent, an outcomes-first bent. For the HoldCos, talent density in their centres can also drive meaningful headcount cost reductions in the subsidiaries (though there is varying evidence on this based on the early stages of this model).
Another distinguishing feature of these firms is the pace and volume at which the larger entities in this cohort in particular are doing their deals. Bending Spoons deployed $2.01bn on acquisitions in the first quarter of this year alone, Long Lake has closed close to 40 deals in its three years of operation, and Thrive Holdings reportedly rolled up over 45 professional services businesses in its first year of operation.
Each marginal acquisition gives the acquirer better information on expected baselines and improvements across core operating metrics in their given verticals (e.g. for Thrive, this might be payroll processing times, reconciliation times, cost-per-report etc.) that they can apply to pricing and improving value on future acquisitions.
Further, the marginal cost of the operating centre of the HoldCo declines with each new acquisition. Long Lake cites that its Nexus AI platform infrastructure is shared across around 70-80% of the portfolio, creating a flywheel whereby each new vertical becomes cheaper to transform than the last.
Finally, a stated goal of some of these new-age HoldCos is to expedite the adoption of AI in older-school businesses where management may be hesitant to upend the entire business model. What better way to drive this change than by owning these businesses with that particular end in mind?
One key issue with AI-related value accretion in businesses is that attempts to date have largely been limited to businesses bolting on hodgepodge AI products to try and improve certain workflows. This has led to ~95% of surveyed organisations stating that they saw no measurable return from the integration of generative AI tools. This new age of HoldCos is essentially a structured bet that integrating AI at core structural levels, and redesigning workflows rather than mildly augmenting them, will deliver vastly greater returns. The business fundamentally has to change in this scenario, but customer bases with persistent problems they pay a third party to solve will remain and continue paying for new, improved versions of these providers.
Under the Hood: A Case Study of Two Acquisitions
Given the relative nascence of these models in practice, there is relatively limited data to show the reader what the numbers look like post-acquisition. To begin with, we can observe what has to be true for these deals to make sense. For each, we can observe what this looks like in the case of two acquisitions made by acquirers with two very distinct profiles: Bending Spoons’ 2022 purchase of Evernote and the more recent 2026 takeover of American Express Global Business Travel (AmexGBT) by Long Lake Management.
For some context before digging in, Bending Spoons closed its purchase of the consumer note-taking app Evernote for a sum reportedly just under $200mm in January 2023. At the time, Evernote had 250 million users that it was converting into $100mm of annual revenue. It had raised ~$290mm of venture capital over its lifetime, and the internal figures weren’t promising the kind of ten-bagger outcome that venture funds live or die by.
After having acquired a wide range of physical services businesses, Long Lake completed its marquee take-private of AmexGBT for $6.3bn in cash (a 65.1% premium to the 30-day trailing VWAP). Unlike Evernote, AmexGBT was not a business in decline – it had produced $532mm in adjusted EBITA in 2025 on $2.7bn revenue, sitting comfortably as the leader in the corporate travel management sector.
Now, the first condition for these deals to be worthwhile is that the price paid must be well below the transformed value of the company. An important distinction to make between this and other forms of acquisition (i.e. by the likes of Private Equity firms et al) is that transformed value here implies that it does not necessarily have to be below fair value or some perceived fair value by the firm. This distinction is driven by the modern HoldCo’s emphasis on its desire to hold companies for the long term. In the case of Evernote, Bending Spoons had paid $200mm for a business that was losing money – any turn to profitability would massively expand the value of the business on the market. Long Lake’s acquisition of AmexGBT was made at ~12x its 2025 adjusted EBITDA number and roughly 10 times AmexGBT’s internal guidance. This is by no means a bargain, but it does represent a bet on there being a transformed price that Long Lake can compensate for this market-leading price with.
Which elicits the second point – the transformation levers must show a reasonable path towards shifting the dial on business outcomes. In the case of Evernote, this is simple to observe – upon completion of the acquisition, at least 129 people were laid off within weeks, with every original employee gone within around a year (a headcount reduction from 341 to 60 staff, mostly supplied by Bending Spoons). By applying their embedded engineering teams, Bending Spoons was able to accelerate feature releases by a reported 50% in 2023 before doubling in 2024. Amidst all this, it also jacked prices up by 86% on its premium plans whilst vastly reducing core feature availability on free plans.
Transformation in the case of Long Lake’s AmexGBT stewardship looks radically different to the above. It has ruled out headcount cuts already, leaving integration with Long Lake’s platform, productivity gains and new technology integrations as the key strings it has to pull on. On the integration side of things, the 2025 merger of AmexGBT with competitor CWT identified $155mm in potential cost savings to be realised, meaning that there is already some transformed value lying in the business for Long Lake to take advantage of. Secondly, Long Lake’s AI focus introduces a prime opportunity for reinventing exactly the kind of workflows that today’s travel agents use themselves – over half of travellers’ requests arrive by email, the kind of thing one sees AI agents beginning to optimise for all over their X feeds. If such agents can handle even 25% more requests each on far quicker turnarounds, revenue can grow without a matching rise in staff costs. This is a key to unlocking margin expansion.
The technology lever of the above argument is trusting that the transition of AmexGBT’s technology costs onto Long Lake’s Nexus AI platform drives substantial cost reductions and enlarges the share of earnings that convert to cash.
This provides a convenient segue to the third success condition of these new-age HoldCo deals: that earnings must convert to cash at a rate above the cost of capital. Returning to the Evernote deal, the business went from a loss-making entity to producing $11.3mm in pre-tax profit (including restructuring costs) in its first year under Bending Spoons’ ownership. Against the estimated purchase price of $200mm, this represents a 6% yield on the funds used for the acquisition in Year 1. This is short of Bending Spoons’ stated 25% unlevered return hurdle on the asset, but it is worth noting that it was the year of acquisition and that little data has been released on Evernote’s isolated performance since 2023.
This conversion remains the biggest test for Long Lake’s AmexGBT acquisition as well. In the year prior to acquisition, the business only converted $104mm of its $532mm EBITDA figure into cash. This figure is largely depressed as a result of $299mm in tax, transaction, working capital & restructuring costs – much of which relate to the aforementioned merger with CWT. For this to improve in ongoing years, a lot rides on Long Lake’s ability to restructure the business as it redesigns the business around its AI platform at minimal cost. Further, with $129mm in capitalised technology costs eating away at EBITDA, successful utilisation of the groupwide Nexus AI platform is a key fulcrum in the earnings-to-cash conversion equation. Further, given that the acquisition was known to have been funded with some debt, and that AmexGBT already carried net debt equivalent to ~1.9x EBITDA prior to the acquisition, earnings must grow at a rate far exceeding their interest bills. This means that any productivity gains from the application of its AI platform must not exclusively go towards improved customer service capabilities, but also have a meaningful impact on the bottom line. In any case, the permanent capital structure of the modern HoldCo also protects the buyer somewhat in that the owner has time to iron out these issues.
Lastly, paying customers have to stick around. Even in the case of Bending Spoons’ more traditional acquisition playbook, it managed to grow mobile revenue on Evernote by 39% even in the event of new customer downloads declining by 64%. It achieved this (as mentioned above) because of customer stickiness, allowing it to raise prices substantially whilst retaining a significant portion of the customer base, and conversions from non-payers to payers improving as a result of stripping back the Free products feature completeness.
In the case of AmexGBT, Long Lake begins from a strong starting point. AmexGBT reports 96% YoY customer retention, with many of its contracts being multi-year and client relationship durations for hundreds of clients approaching somewhere around 15 years (the business itself was started 111 years ago to help get AmEx travellers’ cheque customers home who were stranded in WWI-era Europe). Like many of the other service businesses that both Long Lake and Thrive Holdings (in accounting & finance services) have acquired, these customer-facing service businesses are a brilliant attraction for this very purpose – customer relationships often have a far stickier human element that AI can augment rather than replace as it cuts costs elsewhere.
This last point raises a very interesting corollary about these four success conditions for many of the HoldCos emerging today that rely on technological advancements to drive growth. Given that these conditions hold (price < transformed value, transformation actually works, profits convert to cash at a rate > WACC, strong customer retention), a lot of the long-term residual value in these businesses lies not just in margin expansion and size, but also in the preservation of human-to-human trust and augmentation that they offer in this regard.
Where Long Lake or Thrive’s edge in the services business here is improved customer service and low-error-rate bespoke support. For Aries Global’s businesses, it will be customer trust in the subsidiary products’ undergirding cybersecurity & infrastructure support. For Ithaca, it will manifest in the augmentation & prevalence of human connection on Bumble. Such relationship advantages may prove to be one of the great market dominance strategies that the early AI-enabled business can offer.
Logical Endpoints & Future Directions
Earlier in this piece, I described these new-age HoldCos as essentially representing a structured bet on the AI-native business coming to dominate the economy that interacts with real-world users today (i.e. businesses that exist entirely outside of the realm of tech industry circles).
In the event that this occurs, what does this mean for the direction of the economy and investments if one or more of these businesses comes to assume Berkshire’s position on the HoldCo throne in 10 or more years’ time? What other verticals will logically come to fall under their grasp? What other novel structures will rise to take advantage of this trend?
Cross-Sector Expansion
Given the trend of the most recently publicised HoldCos having more of an AI bent than earlier predecessors like Constellation or Bending Spoons, the first place to look is in where AI & agentic systems are already demonstrating significant adoption as tools and exhibit capability gaps with their human counterparts.
Produced by Claude
As the above image shows, there remains low-hanging fruit where agentic capabilities can substitute for human drudgery for key workflows. Obviously, these agentic systems need to be up to snuff with regard to fulfilling the trust architectures that these businesses rely on, but platform fidelity is already a key part of the value transformation in those business lines denoted as ‘Already targeted by AI holdcos’ above.
Placing these sectors against the four success conditions for a high transformation value business outlined in the previous section above we can look at a series of sample acquisition candidates as follows:
Sample Target List produced by Claude
The above is just to show where the most logical short-term targets may sit based on core agentic value-add today. To factor in the final point of the above section and the stated intentions of certain modern HoldCos to refactor and improve largely human-facing businesses with AI as an aide:
Both samples produced by Claude
While much of the focus of the discussed HoldCos (and thus, this piece) has centred around the application of AI-enabled operating centres to acquired businesses, what about alternative value-accretive technologies?
Crypto-Centric Acquirers
Adjacent to the AI pitch is the idea that programmable settlement of both financial & logical flows, composable applications and minimal transaction costs could transform the cost structures & service delivery frameworks of certain businesses were they to be re-wired on crypto rails.
This idea first manifested with the launch of Inversion Capital in 2024. Initially launched as a PE-style acquisition vehicle and later also operating an Avalanche L1 with the same end goal in mind, Inversion’s stated purpose at inception was to acquire telecoms & remittances businesses primarily in high-crypto-adoption LATAM economies. Whilst there have been no acquisitions publicly announced by the firms in two years, the concept itself is compelling.
Tether’s expansion into alternative business lines (public communications infrastructure, energy, invoice financing) is a more concrete, active example of such a framework being applied, but in this case more analogous to the conglomerate approach than that of the purpose-built HoldCo.
There are certain idiosyncratic limitations to this model, caused by the very benefits that crypto espouses. The ‘float’ argument for the crypto-native acquirer would be aided by cash being released from quicker settlement; however, it would also be its own enemy in the case that settlement friction is exactly one feature that allows certain businesses (for example, in the case of trade & invoice financing) to hold customer funds and reinvest or earn interest on that, which instant settlement would eat into.
Programmability offers the prospect of new products (particularly in the case of trade financing, again), but onchain rails require that all parties in the interaction live on them, again blunting the transformative value of the rewiring.
HoldCo as GP Partner
A third interesting design space into which the modern HoldCo could encroach is that of the search fund or small-to-medium rollup business. There is up to $5T (trillion!) in enterprise value tied up in over 1 million small-to-medium businesses with ageing (read: boomer) owners seeking but not finding immediate successors – this is in the US alone, with Japan and Europe showing similarly large figures with respect to the number of businesses suffering or anticipating this problem.
The first argument for this being a viable strategy for the modern HoldCo is that it expands their sphere of influence without corresponding increases in headcount. They can expand the application of their proprietary software in their chosen verticals (i.e. services businesses in the case of Long Lake, finance & accounting for Thrive) across geographies more rapidly.
By expanding this sphere of influence, the HoldCo compounds in platform advantages without substantial marginal cost given that the core platform used in its existing PortCos can be shared. In this case, scale becomes the moat, as every business in the network adds valuable data on what key performance metrics should look like when an HVAC, landscaping, insurance agency or dental practice is operating at its most optimal.
GP stakes in these smaller acquirers also decentralise the Investment function for the HoldCos, with returns tied to manager outcomes rather than salary + bonus compensation structures. It also diversifies the risk held by HoldCo platforms across geographies and verticals (given that there is still value to be gained from these verticals using the core HoldCo tools). On the manager’s end, having a major HoldCo as a partner has massive benefits for improving seller confidence, operational support, and sourcing & underwriting deals. This negates the core source of failure for search funds, which is that many only around half ever complete an acquisition, whilst giving the HoldCo access to the solid (aggregate 33.9%) returns that these search funds have historically demonstrated post-close.
In a world where the work of some of these HoldCos may serve to displace human wage labour, what better substitute could there be than operating as a capital partner to people looking to build their own equity value in businesses that are desperate for new ownership? In the case that the search manager wants an exit, they have an obvious path to liquidity through selling the remainder of their stake into the HoldCo itself.
Tangentially, if one of the reasons for the spread of the modern HoldCo is the hastening of the broad adoption of AI models at the foundation levels of a company’s business processes (see: OpenAI’s stake in Thrive Holdings), what better way could there be to accelerate this adoption than by ensuring that these operating centres are spread as far and wide as possible?
In Summary
Berkshire Hathaway’s origins as a holding company trace back to the mill that lends it its name to this day, but its compounding float advantage did not become apparent until it got its teeth into the insurance business. The mill would shut down in 1985, while the entity itself may be the highest idol of corporate success this century. Given the rapid nature of change in the space the HoldCos of today operate in, there is every prospect that there is a similar disconnect between mill and float, where the true compounding advantages do not reveal themselves transparently until either (1) they find themselves in full effect or (2) they are applied to their most fitting use cases.
Whilst there is an easy line to draw from Berkshire to this growing generation of HoldCos, the similarities can be most easily conceptualised by three commonalities: permanent ownership, patient capital and a unique source of compounding advantage. Whilst the first two components of this equation remain more-or-less similar, the vast technological advances and new business models cropping up mean that the third (sources of compounding advantage) look starkly different today.
Though there are some exclusions, the key compounding advantage that these firms are bringing to the market is AI-enabled operating centres that promise to rewire their subsidiaries on more efficient and effective rails. Where the Buffett float was defined by being able to reinvest cash before its use was mandated in the business, these businesses are seeking to augment similar cash float (subscriptions, long-term contracts etc.) with some kind of intelligence float instead – applying intelligence to businesses with strong customer moats before anyone else has applied it to compound these advantages.
Whilst a lot of the business logic remains the same from previous eras of acquirers and other forms of corporate conquistador – buy at a price less than you think it might be worth, make sure you can actually improve the value of the business with the tools at your disposal, convert profits into reusable cash and don’t piss off your customers enough that they leave as you go through the preceding steps – there is a human element that might easily be overlooked. By applying frontier tools to high-retention businesses, the businesses that endure will most likely be those that can capture, retain, and delight their customers with these tools rather than harvesting them.