A Few Thoughts On VC-Backed Rollups

Justified VC enthusiasm around AI-enabled services has turned into questionable capital allocation

I typically write and re-write long-form pieces over weeks-long periods, but I have a six-hour flight from NY to SF in the midst of this Crowdstrike fiasco and wanted to put out a few quick ideas (cross-posted from Twitter) around the newest trend in Silicon Valley — VC-backed rollups — that I’ve been grappling with over the last few months. Please reach out with feedback; my email is below.

“Services” was a dirty word in venture until 5 minutes ago. Now, it feels like every VC firm has a thesis for why they’re investable in the era of LLMs. This is both exciting (because I think this view is correct) and amusing (because it became consensus overnight, like everything else in the SF venture matrix).

Along the way, we somehow experienced a not-so-subtle shift from “AI-forward services companies are venture-investable” (a view I’ve held for a long time) to “VCs should invest in rollup strategies for AI-enabled services businesses” (no). The latter statement is decidedly not a logical extension of the former. I know I’m far from the only person who feels this way. But I’ve heard it so much over the last few months that I feel compelled to offer a few thoughts.

To start, we ought to consider why the VC-backed rollup opportunity is exciting to some. Some of the excitement is warranted: (i) if AI reduces the barrier to building software products, distribution becomes relatively more valuable and (ii) selling AI tools to old-school services businesses run by boomers is extremely difficult. For these reasons, there are compelling reasons for software companies to “buy their customers” and own the end-user relationship directly.

Of course, a lot of the excitement is attributable to typical VC hype-chasing, momentum investing, mimesis, etc. I’m sure some vibes-driven investor (who has never seen a balance sheet) came across Mark Leonard’s letters earlier this year, posted a Twitter thread, and whipped his followers into a frenzy. Constellation, Transdigm, Acrisure… the thoughtboi case study threads started flying. The smartest people in the business (Thrive) funding something in the space created an even greater sense of FOMO.

The nature of venture as an asset class has contributed to the frenzy as well. Venture is structurally broken. There’s never been more capital chasing so few companies that matter. Many brand-name firms are waking up to the reality that they have no right to win deals in the ultra-competitive landscape. It’s obvious that only a tiny number of firms will continue to produce strong returns and that everyone else ought to pack up shop. They’re desperate to put money to work. Why not roll the dice and try their luck at rollups? I can already envision the pithy taglines for the website: “we purse a private equity strategy in the venture markets.”

Buying, transforming, and rolling up old-school services businesses is going to produce wonderful outcomes for some investors. I’d be shocked if any of them are VCs (at least as most VC firms are constructed). In the same way that the ZIRP VCs couldn’t differentiate between a good product and a good business (let alone a good investment), they can’t differentiate between a good strategy and a good strategy for VCs.

VCs don’t have the expertise

Investing in the unleveled equity of high-growth, unprofitable startups requires an entirely different underwriting and post-investment support skillset than running a serial M&A strategy. VCs (rightfully, for the most part) roll their eyes at Wall Street types who read a Paul Graham essay and decide to build a startup. But expertise-strategy mismatch cuts both ways.

The fund math doesn’t pencil

If you are running a venture fund in the classic mold, you expect most investments to be worthless or return a very small amount. Fund-level returns are driven by a small handful of companies that become very big.

Any given investment must have the potential to become very big: return all (or at least a very large chunk) of the fund. Rollup strategies simply don’t have the required risk/return profile to make sense for a VC investor. They aren’t taking the sort of market, execution, and technical risks that berth venture scale outcomes. VC-backed companies become huge because their founders take on at least one of these risks, conquer them, and build something that nobody else can: differentiated, durable, moaty businesses.

If you talk to founders pursuing a rollup strategy, they’ll usually tell you that AI transformation can be achieved by implementing off-the-shelf AI tooling and some lightweight in-house AI development. They promise to be highly capital-efficient: “we’ll deliver great venture-scale returns because this will be the last money we ever raise!” This is a huge tell with respect to risk-reward potential. Capital efficiency is a double-edged sword. On one hand, the ability to fund growth via FCF and non-dilutive debt is highly attractive for initial equity investors. On the other, it indicates why the return potential for these businesses is much lower than most venture-backed companies. True venture-scale companies constantly need capital because they’re making very expensive R&D and long-term customer acquisition investments in the service of creating something hard, novel, and durable. AI-enabled rollups implementing third-party tools or building cheap solutions in-house simply indicates that what they’re doing isn’t very differentiated.

I’m skeptical that “seeding” rollup strategies with tiny checks makes much sense

Most VC-backed rollups are raising small amounts of money (like normal seed rounds) to purchase companies. I think this is a flawed approach for a variety of reasons. Investors ought to start by making large platform acquisitions, not tiny sub-scale ones.

First, one of the driving beliefs behind the “buy and build” thesis is that distribution trumps product in post-AI software markets. If this is true, you want to buy businesses with as much existing distribution as possible.

Second, scaled companies have much better access to large quantities of proprietary, domain-specific data to create powerful (and hopefully differentiated) AI tools. If data is the new gold, you want to buy businesses with as much of it as possible.

Third, buying a large business to start limits the amount of M&A execution required — it’s much easier to identify, execute, and integrate one target business vs. doing this serially with many tiny companies. This is especially true for venture investors with 10-year fund lives; VC-backed companies need to achieve huge scale within a certain timeframe, and M&A is quite slow.

Fourth, buying a large business makes the ROI math pencil — it offers a large existing cost base over which to amortize up-front investments (losses) in AI tooling. A crude example: if it will cost $5M of R&D expense to build great AI tools that reduce costs by 50%, you’d rather be working with a $10M cost structure than a $1M one.

Fifth, buying a large business allows you to take advantage of multiple arbitrage, a key driver of value creation for rollup strategies. It works if you start with a large platform business that commands an attractive valuation and subsequently buy sub-scale players at low prices, integrate them into the platform, and have them become more valuable given scale advantages, professionalized management, etc.

Sixth, structuring these investments like typical seed deals is downright silly. The typical seed construct makes sense because (i) so much of potential future returns are attributable to founder labor, knowledge, insight, etc. rather than pure capital and (ii) there is always an implicit bet that founders are taking at least one meaningful risk, giving them an opportunity to build something novel and huge if they’re able to conquer that risk. It’s true that developing a strong, repeatable AI-led transformation playbook isn’t easy, but ripping out huge cost centers via automation (an inherently “N to 1” process) is nowhere near as difficult as building something truly zero-to-one. Funding these investments like typical seed deals is essentially paying an enormous premium to book value for a business with no reasonable expectation of growing book value per share so quickly to make that premium worth it. Consider VCs purchasing 20% of a VC-backed rollup: if ~100% of the capital finances acquisitions, VCs have paid a 5x premium to book. A 20x return on the underlying asset would be legendary, and the VCs would make 4x their investment. Silly!

All this is to say: I am extraordinarily excited about capital allocators pursuing an AI-forward services transformation strategy, especially those who pursue serial M&A in certain end markets. But this strategy makes no sense for VCs, at least how they're pursuing it today. Like the e-commerce rollup phase during COVID, I suspect this new phase will end in tears for many.

If you have questions, comments, or feedback, please reach out: andrewziperski [at] gmail [dot] com.

The views expressed herein are the author’s own and are not the views of Craft Ventures Management, LP or its affiliates.