The Artificial Investor #54: AI drives a new VC model
✍️ The prologue
My name is Aris Xenofontos and I am an investor at Seaya Ventures. This is the deepdive version of the Artificial Investor that comments on one of the top AI developments of the last few days.
This Week’s Story: AI is pushing the most prominent VC funds away from VC
Thrive Capital, the late-stage VC fund, announced it has acquired Crete Professional, an accounting firm, and plans to spend 500 million dollars in acquiring smaller competitors and integrating AI into their business. This announcement came a few days after Elad Gil, the renowned early-stage investor, stated that “AI-powered rollups is his next big bet” stating that this strategy helps accelerate AI integration and improve business margins from 10% to 40%. Other top-tier US investors, such as General Catalyst (commited a 1-billion-dollar Creation Fund) and Khosla Ventures, had already expressed their interest in this “new asset class”.
What is this new AI VC trend all about? What are the drivers for this trend? What are the opportunities and challenges in transforming large businesses using AI? In what cases does it make more sense to focus a VC fund's strategy in transforming large businesses with AI as opposed to funding AI-native startups?
Let’s dive in.
Introduction
Venture capital firms are increasingly adopting a hybrid VC/Private Equity strategy, acquiring established traditional businesses and infusing them with AI to unlock value. This trend gained significant momentum in the past year, driven by both the explosive potential of AI and shifting conditions in the venture ecosystem.
General Catalyst names these AI-driven acquisitions a new asset class and has already backed seven such roll-up platforms. One, Long Lake, is buying up homeowners association management companies to streamline community management with AI; it has raised 670 million dollars within two years. Thrive Capital has reportedly raised a 1 billion dollars of permanent capital vehicle dedicated to acquiring legacy service businesses and transforming them with AI at the core. Khosla Ventures, known for bold tech bets, is cautiously entering the space by targeting companies like call centers or accounting firms that have steady revenues but limited tech – “very unlikely to lose money” – and then enhancing them with AI. Even prominent angel investors such as Elad Gil have shifted focus to this model.
Underlying all these moves is the promise of capitalising on AI’s transformational power while mitigating venture risk with the safety provided by established businesses.
The drivers
🏄 Riding the AI wave
We are experiencing a new industrial evolution. The new wave of AI is reducing the costs of creation to nearly zero while AI’s capabilities have been improving exponentially to the point of exceeding humans.
Many traditional sectors, such as call centers, accounting firms and staffing agencies, remain inefficient and largely analog in their operations. Investors see an opportunity to inject AI and automation to streamline workflows, cut costs and expand services. The expected outcomes are primarily higher margins, as well as business growth.
🐘 Challenges in selling AI to incumbents
Selling software and AI solutions to incumbents is a challenging task: 1) long sales cycles driven by multiple layers of approval, 2) budget constraints due to other priorities and existing vendor lock-in, 3) integration complexity driven by existing legacy systems, 4) governance concerns over explainability, data privacy and operational disruption, and 5) lack of startup credibility.
Rather than trying to sell AI to large companies, some VC firms believe owning a traditional business outright and rebuilding it with AI is a faster path to transformation.
📉 Declining VC returns
VC returns have recently hit decade lows, even turning negative in some quarters. This is driven by 1) higher entry valuations due to increasing competition, and 2) dry-up of exit opportunities (fewer IPOs and big acquisitions) in 2022-2024. There is pressure for firms to look for new investment avenues in order to maximise returns for their investors.
💰 Difficulties in deploying billions of dollars
Income for VC employees is driven by a combination of a commission on the returns of their investments (the “carried interest” or “carry”) and an annual fee linked to the fund’s size.
With returns declining, inevitably some VC firms have turned their focus on fund size: many leading VC firms raised huge funds during the 2021 boom. However, as large exits have become scarce, the late-stage opportunities, which have traditionally been the main way to deploy large sums of money, have also become more limited.
Acquiring stable, cash-generating businesses requires a lot of money and thus provides a new channel to deploy large funds.
😧 Hard to know where to invest nowadays
We have to admit it: as a venture capitalist, it’s not so easy to know where to invest nowadays. The technological landscape moves so rapidly, that it’s hard to have conviction on whether a product that thrives now will be relevant in 2-3 years time. At the same time, every day it gets easier to build and launch products resulting in very high competition.
Many venture capitalists believe in the AI trend, but have found it hard to invest in AI products. An apparent solution at the beginning of this AI wave (early 2023) was to invest a layer down in the Tech stack, i.e. in AI Infrastructure, such as OpenAI, Anthropic, Groq, CoreWeave, etc (the “picks and shovels” approach).
Since the funding saturation of that segment, another alternative has risen: investing a layer up in the Tech stack: the client base (transformation).
An additional benefit here vs. the Infrastructure Layer is that it’s less risky, because since day one investors sit on cashflow-generating, stable businesses (as opposed to startups burning cash for years).
Opportunities
What if you could leverage millions of historical proprietary data points to train a top-notch AI model and use it to launch an AI product that would get adopted overnight by tens of thousands of users? This is the promise of AI rollupst.
👥 Immediate user distribution
AI rollups enable the deployment of new Tech solutions across a ready-made large client base. In legacy industries that are relationship-driven or trust-based, incumbents have a significant edge: they’ve spent years (or decades) building customer trust, brand recognition, and distribution channels. AI startups could take years to accumulate a comparable customer reach.
🏰 Data moat
Traditional businesses sit on valuable proprietary data collected over many years of operations. For instance, a hospital chain has extensive patient outcome data that could power healthcare AI models. Data is a goldmine for AI model training and can become a source of a strong moat, helping AI products to improve with feedback loops.
🫏 Faster adoption in stubborn industries
In industries that have been historically resistant to adopting new technology, such as Real Estate/ Construction, acquiring and transforming a traditional business can accelerate technological adoption by “packaging” it as an in-house improvement. Effectively, end-users (employees and clients of the acquired firm) are carried along through the transformation rather than asked to buy a product from an unknown startup.
Challenges
It’s obviously not all rosy. How do processes get reengineered? What happens when employees are afraid of change and resist it? How does AI interact with legacy systems? How much does the transformation cost and how certain is the return on its investment?
Big company transformation certainly comes with its own challenges.
🛑 Change management & cultural resistance
Long-time employees may fear automation will replace their jobs. A survey found over half of executives cite cultural resistance as a top barrier to AI implementation. Overcoming such barriers requires a slow, phased approach to implementation to win user buy-in.
📚 Employee training & skill gaps
Traditional firms typically lack in-house AI expertise. If employees are not properly trained or engaged, AI tools might get underutilised or misused. So, workforce upskilling becomes essential, which is expensive, time-consuming and not always successful.
➿ Process reengineering & systems integration
Large or traditional companies need to redesign workflows to embed AI. Tasks need to be redistributed between humans and AI (e.g. AI handles routine inquiries while humans handle exceptions), which requires careful planning to avoid service issues during the transition.
Also, most traditional businesses run on outdated software infrastructure not compatible with modern AI applications. Plus, that data that is critical for AI may be trapped in siloed, disparate systems, requiring extensive data cleaning and integration work before AI can even be deployed. All of this process, system and data reengineering is resource-intensive and can disrupt daily operations.
As noted by Elad Gil after meeting numerous AI roll-up teams: “many still need to sort some things out” [before they are ready].
💵 High upfront costs & resource intensity
Many companies will find themselves facing significant capital expenditure for new technology, infrastructure upgrades, cloud migration (as they often operate on premise), data management and hiring AI talent or consultants. A McKinsey study found AI integration projects in sectors like banking and manufacturing cost 1.5-5 million dollars on average.
Additionally, management’s attention is limited and focusing on an AI transformation initiative can distract leadership from the core business.
A Khosla partner noted: “We wouldn’t do it alone, we don’t have that expertise,” referring to the need to partner with experienced operators to execute acquisitions and integrations.
👷 Uncertain return on investment
And on top of all tangible and hidden costs, as well as the risk of execution, there is a risk that AI does not deliver the efficiency gains expected.
Limitations come from both ends of the spectrum of automation; no or too much automation.
On one hand, some workflows are highly specialised or reliant on human judgment and prevent AI from improving them. On the other hand, in some cases AI commoditises the entire service all together. For instance, in the case of language translation, automation can make a previously high-margin service cheap and ubiquitous impacting significantly an incumbent’s pricing power. What would be the point in an AI rollup strategy in the translation services sector?
Finally, value destruction may come also from increasing acquisition prices, as more VC capital gets directed to AI rollup strategies chasing the same targets.
An “AI roll-up” investment guide
There are great opportunities and challenges in a strategy of acquiring, merging and transforming traditional business using AI.
It’s not as simple as saying “it works” or “it doesn’t work”. It depends on the attributes of a specific market in question. The main factors to consider are: 1) Market fragmentation, 2) Purchasing process, 3) Regulation, 4) Data landscape, 5) Customer lock-in and 6) AI impact.
🧩 Market fragmentation
Consolidated markets favor AI roll-ups because incumbents control distribution and data at scale.
On the other hand, in fragmented markets startups can tap on underserved niches and scale bottom-up, while roll-ups are less effective because acquisition synergies are weaker.
At the other extreme, nascent/ “greenfield” markets with no dominant players or legacy systems, such as GenAI Security Platforms, are a perfect playground for startups.
🤝 Human touch in sales
AI roll-ups succeed in relationship-driven markets by integrating into long-standing client/ vendor partnerships. A good example is Trading Software for large financial institutions, where a large part of innovation has come either from incumbents (e.g. Bloomberg) or industry consortia.
On the other hand, startups tend to do better in self-serve or low-touch environments where users try and adopt products directly.
🏛️ Regulation
Heavily regulated sectors (e.g. medical devices) favor roll-ups because incumbents have built compliance infrastructure and customer trust. At the same time, regulation slows down procurement, which makes life harder for startups, often forcing them to look for partnerships or narrow non-core use cases.
Expect AI-native startups to excel in low-regulation domains where rapid experimentation and short product cycles are the norm.
📄 Data landscape
It’s easy here to talk about a data advantage for AI rollups, as incumbents often hold rich, proprietary data sets that AI rollups can leverage to build differentiated internal tools.
However, the devil is in the details. AI rollup success depends on access to clean, usable and compliant historical data, which rarely is the case.
There are also cases where new usage data is more powerful than existing data sets. Take AI Coding tools as an example. There is so much coding data out there: from Github code repositories to Stack Overflow debugging discussions. However, the quality of that data varies significantly. On the other hand, the new AI coding tools, such as Cursor and Windsurf, are capable of “handpicking” the best human coding data, as well as access proprietary data on how users interact with AI.
🔐 Customer lock-in
AI roll-ups have a higher chance to win in high-lock-in environments, driven by long-term contracts and deep system integration. A good example can be Industrial Automation Software, where legacy manufacturing systems are difficult to replace. There, incumbents grow by expanding wallet share and cross-selling within already locked-in accounts.
Startups have more room to grow in markets with low or moderate switching costs, and ideally where they can offer plug-and-play solutions. Subsequently, startups do have the opportunity to create lock-in through network effects (e.g. data, platform, community, etc).
🤖 AI impact
High-trust sectors (e.g. medicine, legal), in which full automation is risky or unwelcome, favour AI augmentation. In these cases, startups have more limited advantages vs. incumbent transformation. The ideal startup territory consists of repetitive, high-volume and low-risk tasks, where the AI adoption curve is likely to be faster, and could even lead to broader market commoditisation.
🔮 Looking forward
Defining whether a market is more suitable for AI incumbent transformation or funding AI startups is complex. The factors one should consider include market fragmentation, the level of human touch needed in the purchasing process, the amount of regulation, the characteristics of the data landscape and how it affects product moats, the depth of customer lock-in, as well as whether AI is capable of an end-to-end automation or not.
As a result, it is not as simple as picking an industry (e.g. Healthcare or Financial Services) and rule it out for AI-native startups because it is regulated. Investors need to go down a level and analyse each market sub-segment separately.
For instance AI Medical Scribing for outpatient clinics (e.g., DeepScribe, Nabla) is probably a startup-friendly segment thanks to lighter compliance, smaller buying units and fast ROI. This could also be the case with SMB Lending using alternative data or KYC/KYB Automation for Fintech companies - markets that offer clear opportunities for startups to win with nimble, API-first AI solutions.
Given that there is no “perfect” market for AI rollups or startups, probably the most rational approach would be to build a score card and prioritise the higher scores on each side of the spectrum for a given strategy. Having said that, the final decision is likely to be a combination of art and science.
Declining VC returns and limited late-stage dealflow are pushing large funds towards alternative strategies. Pre-2022 it was about pre-IPO transactions, SPAC deals and public investments (cross-over funds). In 2025 it is about AI roll-ups. It remains to be seen whether this strategy is a winning one; but this time around, don’t expect it to be a quick verdict. Large traditional businesses take years to get transformed.
See you next time for more AI insights.
great piece Aris!