my thesis on ai rollups and smbs

TL;DR
We're seeing a transfer of wealth + no successors for SMBs in service based economies; Harsh macro conditions (cheaper multiples); RL environments scaled for SMBs; Forward Deployed + Search Fund type consortium

I tweeted this thought a while back:

hating ai is like hating electricity

I’m starting to consider seeing AI through a public utility lens and it got me thinking: when electricity started being introduced, did the shops who had better functioning light bulbs get better sales than the stores that still relied on candles? Could the stores/office who had gas heating start work earlier than those who didn’t during cold harsh winters? Did restaurants who had electricity for refrigerators first get margins because they could store leftovers overnight?

These questions seem rather stupid but in a sense seem quite interesting to think about even knowing there will never be a satisfying answer. A lot of small medium businesses (SMBs) definitely hear about AI, but don’t think much about how to implement it. Maybe because they don’t care or maybe they’re content the way they are operating. Somtimes it’s the idea that, okay if I could automate this, what else would I do with the saved time? They frame it as that they prefer the satisfaction of actually running their own businesses.

It seems to me AI can only ever be adopted are for medium corps where the C-suite wants it, or the employees implement it from the ground up or getting someone who wants to implement AI in a businesses to straight up buying out and owning the business. I started looking more into the latter and figured I’d a brain dump on AI acquisitions and rollups.

The traditional search fund way of buying and owning one business is worthy of an essay itself but I’d focus on AI rollups here — basically, scooping up small, fragmented service businesses and juicing them with AI to crank up efficiency and margins. There’s also some financial engineering optimizing to do in the valuations, but in a sense, this feels like a private equity meets venture capital play, but with a tech twist.

a palantir for SMBs + special acquisition vehicle? The core idea: Instead of just building software and hoping SMBs (small and medium businesses) buy it, why not buy the businesses themselves?

Acquire them cheap, layer on AI to automate the grunt work, and turn low-margin ops into something scalable. I first got fairly hooked reading about how this is blowing up in the US, with VCs like General Catalyst earmarking 1.5B for their Creation Fund in AI-enabled rollups, or Thrive Capital backing plays like Long Lake with 18 HOA management acquisitions and 600M raised, plus Crete targeting 500M in accounting firm buys. They’re not just really just slashing costs like old-school PE; they’re using AI to rethink the whole service delivery to be AI-native. Or a full stack AI company as YC puts it (we’re seeing a rise in these companies backed in later batches, see Cranston and Claybird)

When you launch a firm that’s born around agentic workflows, every email, model, and meeting transcript is captured in a clean, structured form from day one. That ‘data exhaust’ feeds the very agents that automate the next engagement, so gross margin expands with each client you serve. (Lyseng)

vc vs pe investment approach Let’s break down the approaches, because not all rollups are created equal. Traditional private equity (PE) rollups are about buying cheap (1-3x EBITDA would be a steal), consolidating, cutting fat, and flipping at higher multiples — think heavy debt and operational tweaks, but not rehualing the company completely. VC-backed ones? They’re betting on AI. Less leverage, more tech infusion for margin expansion without gutting the team.

Diving deeper, PE's bread and butter is multiple arbitrage: snag assets at low multiples in fragmented markets, bolt them together for scale, and exit at 5-10x EBITDA thanks to synergies and professionalization acriss the board. But AI rollups amp this up — VCs like Thrive and General Catalyst are blending in venture-scale upside by embedding AI for gross margin jumps, say from 20-30% in services to software-like 60-80% through automation of knowledge work. The idea is that it is not just cost cutting; AI enables revenue growth via better products, like in staffing where models handle unlimited multilingual calls with zero bias. Risks lurk in integration— PE pros know the drill with cultural clashes and debt loads in high-interest eras, but VCs bring the AI hype, often underestimating ops complexity.

From what I’ve pieced together, VCs like Thrive Capital and General Catalyst are going PE-style but AI-native: acquire fragmented players, embed AI for automation, and scale outcomes. Thrive’s got over 1B for this, partnering with OpenAI to customize models for acquired firms in accounting and IT. General Catalyst’s “creation strategy” and Perceptra arm incubates AI companies then buys services biz to transform (100-150M checks for stuff like Crescendo in call centers, which acquired PartnerHero and hit ~90M revenue by May 2025)

AI commoditizes software, so owning the customers (via acquisitions) beats selling tools they might ignore. PE pros handle integration perils, but VCs bring the AI hype for upside.

why i’m bullish, especially for hk’s aging services scene I know Hong Kong’s economy and business landscape the best and here are my two cents of how it might work here: there could be a play here where the economy’s all services, the population’s graying fast, and SMBs are scrambling to stay afloat without enough young hands (the children going to be doctors, lawyers or bankers instead of taking over an old-school business.

Refining my take: AI rollups aren’t just a trend; they could be a lifeline for service-heavy economies staring down demographic cliffs. Hong Kong’s a prime example—financial services, tourism, retail dominate, accounting for 93.5% of GDP in 2023, but with an aging pop and talent drain, SMBs are hurting. AI could pump HK 260B into the economy, juicing finance by HK 83B via better fraud detection and modeling, tourism by HK 15B through personalized itineraries, and e-commerce by HK 5.7B with optimized ads and automation. (HK census and gov data).

Zooming in on the AI mechanics, integration often starts with forward-deployed engineers (FDEs) embedding in acquired firms to map workflows—think shadowing workers to spot automatable tasks like data entry or content gen, then layering LLMs for 20-30% efficiency gains. In PE terms, this drives EBITDA lifts of 2-3x by automating core services, not just back-office, creating defensible moats via proprietary datasets from rollups. For HK's silver economy (which is a major policy direction push), where the elderly (65+) hit 22.4% or 1.68M in 2023 and are projected to reach 35% or 2.67M by 2043, with median age at 47.4 in 2025, AI handles repetitive elder care like fall detection or personalized plans, freeing scarce labor while scaling services in a market where 20%+ are over 65.

Pushback I’ve heard: critics often say AI rollups won’t stick because tech vendors already flood these markets, or integration’s a nightmare—like how 42% of enterprise AI initiatives got axed in 2024, up from 17% in 2023. Fair, but in aging spots like HK, where SMBs lag on AI adoption (only 2% fully prepped per Cisco), rollups force the upgrade. Plus, data from owned assets trains better ctiadel models, creating potential moats.

unrelated, BVP's memo on ServiceTitan has relevant ideas on service transformation; memo

investment professionals don’t think too deply about implementation What often rollups bull cases miss, or rather just assume is that the tech with some smart engineers will diffuse automatically. Most attempts still revolve around superficial orchestration: wrapping large language models around existing workflows via simplistic retrieval-augmented generation or basic tool-calling scaffolds. But the durable benefits emerge in a different way.

I could see Prime Intellect’s verifiers and RL environments being quite interesting to consider for SMB use cases. The subtle bottleneck here is context and proprietary data.

Most AI rollups essays/memos focus more actually on medium businesses rather than the small. By small I mean accounting firms with 3-5 people or family businesses run by 3 siblings. Admitedly, finding them are quite hard but sometimes if they’re looking for exit liquidity or have no successor, an opportunity might present itself. I haven’t found any good community/broker for this yet tho. HK business has a lot to do with guanxi.

That said, if you happen to be/have friends who’s a cracked sales/ops guy, a cracked engineer and someone who knows their M&A well, this seems like the thing to do.

TLDR; Acquire cash-flowing SMBs, use AI for efficiency (e.g., automate back-office, expand outcomes), fund more buys. In aging economies, it compensates for workforce shrinks — AI creates abundance, dropping costs for goods/services while upskilling what’s left. For HK SMBs, it’s high leverage. I know the Hong Kong General Chamber of Commerce wants to help SMBs, maybe there is a conversation to be had there.

Folks/Players to watch:

Multiplier Holdings, Percepta (General Catalyst), Thrive Holdings
Related: Chamath’s 8090, Tenex (more consultancy/FDE style), Varick Agents
Our favorite tweets/X people on adoption: ~Liam Ottley~, ~vas~, ~amirmxt~, ~Bryan Kim~, @rohitdotmittal
VCs: Hemant Taneja/Marc Bhargava (General Catalyst): Leading the 1.5B Creation Fund, Josh Kushner (Thrive Capital)/Crete (accounting acquisitions), Vinod Khosla (Khosla Ventures), Trace Cohen (Six Point Ventures), Chris Young (ex-Microsoft venture & acqusition lead)
Sahil Patwa, Evan Lyseng (does really good substack pieces)
Sceptics: Dave Friedman, John Hwang

Reading list: