Larry Cheng

Co-Founder/Managing Partner, Volition Capital. Journaling general business thoughts so I don't forget them. Christian. Harvard. Several private/public boards.

[Larry Cheng @larryvc]

Some thoughts on how to come up with the "big idea" on which business to start:

Let's think about the difference between being original and being contrarian. An original idea is an idea that no one else has. A contrarian idea is one that everyone disagrees with. They can be related but very different processes to arrive at the end result. One is much easier than the other.

To have an original idea, you need to think of something that no one else has ever thought of before. Originality places an emphasis on being novel - conceiving of things that have not been conceived of. That sounds like a high bar as it is.

To have a contrarian idea, you need to think of something that everyone else agrees with and take the opposite view. It's actually quite easy to think of things that everyone agrees with - the real question is can you get your head around the opposite view.

Often times when people think about being an entrepreneur, they put pressure on themselves to find that "original" idea that no one else has pursued before. They're going down the originality path. That can seem very daunting because every idea seems to have been taken.

I'd suggest an alternate approach that is much easier - rather than focus on being original, focus on being contrarian. By that I mean, think of something that everyone agrees with, and pursue the opposite idea. You will have a contrarian idea - and actually it just might be original as well.

As an example, if you were in a room full of 1,000 people 20 years ago - and you asked yourself the question - what do all 1,000 people in this room agree on?

One of them might have been: don't get into cars with strangers. Every single kid growing up in my generation was taught the same thing - don't get into cars with strangers. All 1,000 people would agree: don't get into cars with strangers.

So, the opposite idea would be: I'm going to build a service to make getting into cars driven by strangers as a normative form of transportation. Exactly 0 out of 1,000 people would have agreed that that's a good idea 20 years ago. And there you have your contrarian idea that became Uber and Lyft.

So if you're trying to figure out your big entrepreneurial idea - try this step. Put on the board a list of ideas or concepts that basically everyone agrees with. And, write next to each idea the contrarian entrepreneurial idea that disagrees with each of those agreed upon concepts. And, see which contrarian idea you think you can build a business around.

You will have plenty of runway to go after that idea because everyone will be stuck in the same worldview. And once the world catches up with your thinking, you may just have a great business on your hands.

[Larry Cheng @larryvc]

Rightly or wrongly, it was the presumptive position of all the LPs I spoke with that AI is in a bubble and the bubble will burst.

[Larry Cheng @larryvc]

LP: What do you think people are getting wrong about AI today?

When we imagine what an AI-native company looks like, we're imagining the wrong thing.

For all of business history, a business has been about bringing together human talent, to leverage human intelligence and capacity, to build products and services for humans. Therefore, when we think about what an AI-native company looks like, it's natural to imagine it as adjacent to the only model we've ever known. We are inherently anchored to a human-centered model.

AI-native companies are built on a completely different set of assumptions. They are foundationally built on machine intelligence, with agents as the primary unit of labor, within a specifically designed agent habitat, to leverage AI capacity to develop products and services for agents (and humans).

They are not a relative of the old model, they are a completely different species.

For certain, human-centered businesses that leverage AI can be very successful, but I wouldn't conflate that with AI-native companies that are built on a company different foundation and worldview.

[Larry Cheng @larryvc]

Met with a number of LPs who are eagerly awaiting the Anthropic IPO. Most appear to be sellers anywhere around $1T or above - too big of a position to risk on any sort of AI correction.

[Larry Cheng @larryvc]

LP: Will intelligence/AI models commoditize?

My response: While I definitely appreciate the common take that models will commoditize, here's my rationale for why they may not.

First of all, I don't think human intelligence has commoditized and that's our only precursor. The global population was 1 billion two centuries ago and it's 8 billion now. Why hasn't human intelligence commoditized over that time frame? It's because there are factors and variables that foster the growth and replication of intelligence that compound over time, not commoditize over time. The same is true for AI models - variables like capital, data, compute, training, etc. are also factors that compound.

Additionally, I think there's a possible flaw in our lens on how we view intelligence. We may inherently set the ceiling on the capacity of intelligence at human intelligence. So, we envision AI models asymptotically headed towards that ceiling and they all commoditize right underneath it.

If we switched our worldview to thinking about the amount of intelligence that could exist above the line of human intelligence, then it's not hard to imagine models blowing through that perceived cap of human intelligence. In some sense, there's infinite room for intelligence above the human ceiling - so there's plenty of space for intelligence to expand. And in that space is where models don't commoditize but they differentiate both generally and with deep and narrow specialization.

The question then becomes in that space above human intelligence that AI models will reside, are the use cases diminishing in value to humans or are they expanding in value in use cases that are hard to imagine since we've inherently been limited thus far? Therein lies the question - I think the latter is more probable than the former and that will be the result of intelligence compounding rather than commoditizing.

We'll see how things play out!

[Larry Cheng @larryvc]

One LP said something interesting to me at the annual meeting - he said that most companies that his GPs are projecting at a ~2x return, e.g. just an okay outcome, he's expecting them to ultimately be <= 1.0x. He thinks the large swath of muddled companies will end up being losses.

There's a lot of experience in that comment. Essentially, he's saying there are a lot of companies that are muddled in their performance for whom there will be little to no market for and performance will continue to erode. In a different market, those companies might have been able to work towards a more positive outcome, but he's more sanguine about their prospects now.

[Larry Cheng @larryvc]

How are SaaS companies valued?

An LP asked this question at our annual meeting - this was my response: 20x forward FCF.

In the public markets, the median SaaS multiple is 20x forward FCF. When you look at the ~65 cos in the BVP Cloud Index and the ~120 cos in the S&P Software/Services Index, it is the same median: 20x forward FCF.

The median public company is a Rule of 25 company in both indexes: in one index it's 18% revenue growth and 7% EBITDA margins, in another index it's 15% revenue growth and 10% EBITDA margins.

The top quartile multiple is 30x-40x forward FCF. The key determiner of a higher multiple is not a financial metric, it's software sub-sector. Security software and infrastructure software companies get higher multiples, and vertical and horizontal software get the lowest multiples - given like-for-like financial performance.

Within the financial metrics, revenue growth is a greater driver of multiple than EBITDA %, but FCF % sits in between in terms of strength of correlation.

The public markets do not value SaaS companies on a revenue multiple as many say. That's like saying Coca Cola trades on a revenue multiple. Yes, if you divide Coca Cola's market cap/revenue, you'll get 7x. But, that's not what it trades at - it's trading at 20x FCF as well. Same as the median SaaS company and same as the historical median of the Nasdaq.

The early private markets value SaaS companies on revenue because there typically aren't earnings. So the key question is between the early private markets paying revenue multiples and the public markets paying FCF multiples, which asset class in the supply of capital along the way will take the multiple compression of paying a revenue multiple at entry and get an earnings/FCF multiple on exit?

I do not think it will be the public markets any longer. That's by virtue of the fact that over half of the SaaS companies in each index are have declined in market cap over the last 5-6 years.

I'm not sure that's apparent to everyone so it bears repeating: more than half of publicly traded SaaS companies have declined in market cap over the last 5+ years. Blue chip names: Salesforce, Servicenow, Atlassian, Hubspot, Adobe, Docusign, etc. among many others have declined over the last half decade+.

That's a re-rating from a revenue multiple to a FCF multiple in plain view in the public markets. I doubt public investors will do that again.

I suspect most private SaaS companies are being valued at a healthy revenue multiple by their existing investors in their respective NAV processes. I expect a high majority of them will exit at a worse multiple than what they are being carried at. That's why TVPI declines towards DPI rather than the other way around.

Which tier of capital is most likely to take the multiple compression of investing a revenue multiple and exiting at an earnings/FCF multiple? That depends on the performance of the company - whichever capital source is holding the company when it goes from >50%+ topline growth to <50% is likely the one that will face multiple compression.

Of course, being valued on a FCF multiple doesn't mean it becomes a bad investment. Most of the economy is valued on an earnings multiple so there can still be big winners of companies that generate real profits. There will definitely be sticky and entrenched SaaS companies for whom AI is a tailwind that can generate meaningful profits and be great outcomes for their investors even if valued on a FCF multiple.

[Larry Cheng @larryvc]

We recently had our annual LP meeting - this is when all of Volition's investors come together and we catch up on the year. It's always a big push but planning to devote some time today to capturing some of my thoughts and our LPs' thoughts from the meeting.

[Larry Cheng @larryvc]

As I've mentioned before, we've begun asking new employees at Volition, as part of their onboarding, to legally replicate a commercial software application in two weeks. Here are two key learnings from it - one of which is counterintuitive.

First, it's surprisingly easy. That's the feedback of those who have done it. Everyone has done it faster than two weeks.

Secondly, here's the counterintuitive one. One of the barriers to entry in SaaS is community. If you have a robust online community of users who are sharing best practices, being a help desk, teaching other users how to optimize the application - that can be a distinct advantage over other SaaS companies. However, it can also create a huge disadvantage.

The reason it can be a disadvantage in an AI world is that communities are creating content about the product. They are showing screenshots, they are doing video demos, they are describing features and use cases all in the public square.

All of that content is getting indexed by the LLMs therefore effectively becoming a free outsourced spec for the world. The more content in the public record on the application, the closer you are to a few prompts to replicate it.

Therefore, while it’s true that robust online community can make a SaaS company harder to displace, it’s also true that it makes the product easier to replicate. Balancing those competing realities will be key for SaaS companies going forward.

[Larry Cheng @larryvc]

This is how things played out in the cloud computing era as well. The Fortune 500 CIOs were the most skeptical of cloud applications as well. As one CIO said to me, "No one will store their data on third party servers." History repeating itself...

[Nick Mehta @nrmehta]

I've had dozens of conversations with CIOs of Global 2000 organizations in the last 3 months. One big learning: many CIOs are the most skeptical execs in their companies about AI.

Some of this comes down to risk - they view AI projects as being so bumpy and that the bumps will mean lumps for them.

Some of this comes down to knowledge - they know that their underlying systems, processes and data need to be rewired for agents to successfully operate.

Some of this comes down to signal - they are inundated with AI pitches and are tuning them out.

My takeaway: startups need to use more plain language - and ideally case studies - in their pitches. Drop the buzzwords - harnesses, models, memory, tools, self-improvement, evals. The AI lead in a customer will care about these.

The CIO cares about the actual specific outcomes you drive (much deeper than the obvious like "faster," "cheaper" and "better") and proof of these in the real world.

[Larry Cheng @larryvc]

This is a sad and avoidable outcome driven by the underwriting philosophy of VCs.

[Callum Williams @econcallum]

Holy moly... this new paper finds that 88% of VC-backed founders receive zero exit value 🤪🤪🤪

[Larry Cheng @larryvc]

My guess is legacy brands will play this dynamic more conservatively and newer brands with less to lose will just go all-in on AI video advertising. The video ad unit that used to cost $10,000-$250,000 to make now costs $9. That will enable hyper-customized video advertising through human-first ad channels like broadcast TV, CTV, DOOH, etc. A national campaign will not be one ad unit distributed nationally but 4,000 custom video ad units targeting the 20,000 zip codes you care about. Expect CPMs to go up in the human channels - this is a boon to their business. Over time (and very quickly) the quality of video will be at parity and the objection will gradually dissipate.

[Larry Cheng @larryvc]

Getting a lot of the smaller things right, but getting the big thing wrong usually leads to a lot worse outcomes than getting the big thing right and a lot of smaller things wrong. This goes for just about everything - life, work, etc.

[Larry Cheng @larryvc]

If a brand or company you buy from used AI video and AI images regularly in its advertising, how would that impact your view of them?

[Larry Cheng @larryvc]

This is practical and interesting data…

[Nick Abraham @NickAbraham12]

Apollo analyzed 5.36M sales activities targeting C-suite execs.

Here’s what gets CEOs to reply:

[Larry Cheng @larryvc]

Token-adjusted EBITDA…

[Delia Lazarescu @tech__unicorn]

overheard in silicon valley:

"we're profitable if you don't factor in the token spend"

[Larry Cheng @larryvc]

This is a dynamic that’s played out over and over in my career at different firms (including BVP). The best investments of any given era when you rewind the tape were often the least liked deals by the broader group. If you’re looking for social proof, prepare to do bad deals.

[Harry Stebbings @HarryStebbings]

Anthropic was the most contentious deal ever done in Bessemer's history.

“According to @bdeeter, it was the lowest-voted deal in the history of the firm during the time he had been there.

But we had conviction.

One of the values of Bessemer is, ‘That is where alpha comes from. If you guys really believe, go do it.’” @spdholakia

How do you think about encouraging non-consensus thinking? @jeremyl @davidcowan @TaliaGold