Long on shovels, short on gold

When everyone is betting the house on AI, there are probably some good deals on the other side of the line. Plus, the real problem with liquidation preferences.

Snowshovel.

One day, the urge to spend more time with your family comes for us all. 

Last week, Frank Slootman stepped down from his role as CEO of Snowflake. Slootman, the Rihanna of Silicon Valley CEOs—a proven hitmaker, ostensibly retired, but available for hire for the right price¹—said that the company will be fine without him. Wall Street hated it. Content farmers made threads about it. Databricks missed their chance to make a funny commercial.²

And Snowflake announced that their new CEO will be Sridhar Ramaswamy:

Since joining Snowflake in May 2023 in connection with the company’s acquisition of Neeva, the world’s first private AI powered search engine, Ramaswamy has been spearheading Snowflake’s AI strategy. He led the launch of Snowflake Cortex, Snowflake’s new fully managed service that makes AI simple and secure for all users to quickly drive business value. Prior to joining Snowflake, Ramaswamy co-founded Neeva in 2019.

Two years ago, Snowflake’s ambitions seemed clear: Be Oracle. Be the next dominant data warehouse vendor. Be boring, but in the best sort of way—by predictably generating $10 billion to $15 billion of profit every year. Turn Snowflake Summit into the next Oracle OpenWorld; conquer the ballooning market for enterprise data management; drink Larry Ellison’s $140 billion milkshake (and, probably, beat his sailing team too). 

Now, that goal might be changing? Slootman—who built his fortune and reputation on being a ruthless businessman—is being replaced by a former ML and AI engineer.³ Though I’m sure that Ramaswamy will continue to invest in Snowflake’s meat-and-potatoes warehouse business, it’s hard not to interpret the choice as the soft launch of a new priority, as a gentle reboot, as a trickle-down pivot, all towards the AI supernova. An engineer from Nvidia reacted to the news:

I was hosting Snowflake's Cortex team at NVIDIA HQ when the news broke that AI Legend Sridhar Ramaswamy would be replacing the Enterprise General Frank Slootman. …

This is a truly exciting time. AI is eating the world and will soon consume the enterprise in particular. We at NVIDIA have known this for years and Snowflake's leadership change signals in very clear terms that they too are serious about revolutionizing the enterprise with AI.

For the last 18 months, the tech world has been hyperventilating about AI. Thirty percent of all U.S. venture funding went to AI companies in 2023. Earlier stage companies have an even more extreme skew, with AI companies making up 60 percent of Y Combinator’s recent cohorts. The 100x multiples are back. Elon Musk is raising $6 billion for his AI company; Sam Altman is raising $7 trillion for his.⁴ 

On one hand, this is normal enough—venture capitalists are prone to irrational exuberance;⁵ it’s kind of their whole thing. If AI is indeed a bubble—and there are rumblings—it won’t be the first time that Silicon Valley overfunded an immature and unproven technology.

On the other hand, there is at least one very notable difference between the AI circus and prior technological megatrends, including both successes like the cloud and the internet itself, and disappointments like the metaverse and the web3 carnival: This time, it’s not just startups that are betting the house on the new new thing. 

Ramaswamy's promotion at Snowflake is just one example of an established company declaring themselves all in on AI. Meta is spending tens of billions of dollars on AI investments.⁶ Alphabet CEO Sudhar Pichai is under fire for botching the Gemini release, and people are wondering if Google—a company that makes $240 billion from ads, and manages a half-dozen products with billions of users each—should discard the person who used to run Search, Chrome, Gmail, Google Drive, and Android and replace him with someone who’s done none of that, but is an AI expert. In 2017, just as Bitcoin was hitting its first peak, AWS CEO Andy Jassey said he didn’t see any need for Amazon to build blockchain services yet. We don’t “build technology because we think it is cool,” he said. With AI, Amazon’s tone changed: Amid the OpenAI-inspired chatbot craze, AWS released Q, a generative AI-powered assistant that will “reimagine the future of work.”

I mean, maybe? Maybe chatbots will reimagine the future of work. Maybe AI is eating the world; maybe it will consume the enterprise in particular. Maybe Google does need to replace Pichai with an AI researcher. Maybe Ramaswamy will turn Snowflake not into Oracle, but into Apple or Google. Maybe people will brag on their podcasts about being among the first $6 billion in xAI. I don’t know.

But for as much as tech Twitter and the financial markets have become enamored by AI, it doesn’t have that many success stories yet? Though some companies have made very real money from it—Nvidia is printing cash; OpenAI’s business is booming; Microsoft is the most valuable company in the world, at least until Nvidia passes it next month—these are the shovel salesmen, and shovel salesmen can still get rich on hype. Where is the gold? Where are the companies that have gotten rich using AI rather than selling AI?

There are some examples. Microsoft Copilot’s makes more than $100 million in annual recurring revenue. Midjourney might make twice that. Last week, Klarna claimed that its support chatbot can do the work of 700 agents. And in a recent blog post, OpenAI talked about how free versions of GPT-4 were helping people:

We provide broad access to today's most powerful AI, including a free version that hundreds of millions of people use every day. For example, Albania is using OpenAI’s tools to accelerate its EU accession by as much as 5.5 years; Digital Green is helping boost farmer income in Kenya and India by dropping the cost of agricultural extension services 100x by building on OpenAI; Lifespan, the largest healthcare provider in Rhode Island, uses GPT-4 to simplify its surgical consent forms from a college reading level to a 6th grade one; Iceland is using GPT-4 to preserve the Icelandic language.

Yes, these are all impressive and useful things. Yes, it’s early, and everyone’s still figuring all of this out. Yes, there are probably lots of other examples that I haven’t heard about, or stories that haven’t been publicly told. Yes, it’s probably smart to talk about preserving languages and making surgical consent forms more accessible when you want to win a PR battle against the richest man in the world and his lawsuit that accuses you of working for yourselves and not the betterment of humanity. Yes, AI may well live up to its trillions of dollars of promise. 

But it’s still just a promise.⁷ And it seems notable that, despite AI also producing its fair share of flops, nobody is betting against it. There are, to my knowledge, no Blockbusters saying that the AI hysteria is completely overblown. There aren’t any clips of people saying that it’s a newfangled radio, wondering why it costs so much and doesn’t come with a keyboard, or doubting it’s so safe and easy after all. Every CEO, it seems, is instead bending over backwards to talk about how much they’re spending in AI. 

In other words, all the action is on one side of the line. That doesn’t mean the bet is wrong, of course; the internet is bigger than the radio, and the $500 iPhone is more successful than the $99 Motorola Q phone and its 35 physical keys. But it probably means the bet is overvalued—or, maybe more importantly, the other side of the bet is undervalued. 

When the dust on this gold rush settles, it’s hard to imagine that there won’t be some companies that pivoted too far, and left too much of themselves behind in their scramble to reinvent themselves for the “era of artificial intelligence.” And, with everyone throwing money and attention at people who zig towards AI, it’s hard to imagine there won’t be some companies—and VCs—who find undervalued talents and great investments in the people and ideas that are great at zagging.

---

Whatever, do the secondary sales 

We watched all those FanDuel ads for nothing:

In July 2018, Paddy Power Betfair (now known as Flutter) acquired FanDuel for $465M in cash. On the surface, this looks like a great win for the FanDuel founders and employees. However, because the two lead investors held strong liquidation preference rights, the FanDuel founders and most employees received nothing in this massive deal.

You could interpret this story in a few ways. One version is about greed and power: It’s about selfish venture capitalists cheating founders of their rightfully deserved fortunes. From the same post:

When the FanDuel founders raised funds, two key investors received a liquidation preference that entitled them to the first $559M in an acquisition. Founders and employees would be paid only if the acquisition exceeded $559M. Because the Paddy Power Betfair was for just $465M, the founders received nothing.

Liquidation preferences guarantee investors their return before anyone else gets paid. For example, suppose a VC buys 10 percent of a startup for $10 million, valuing the company at $100 million. Now, suppose the startup eventually sells for $20 million:

• If the VC has no liquidation preference (this is uncommon), they will get 10 percent of the proceeds, or $2 million.

• If the VC has a 1x liquidation preference (this is common), they are entitled to either 10 percent of the proceeds or 1x their original investment. They will take the $10 million—i.e., 1x their investment—and founders, employees, and other investors will split the other $10 million.

• If they have a 3x liquidation preference (this is what happened to FanDuel), the VC will be entitled to either 10 percent of the proceeds or three times their original investment. Because the latter amount is $30 million, and there are only $20 million of proceeds to distribute, the investor is entitled to everything. And everyone else gets nothing.⁸ 

The FanDuel story could be about that: A warning to startup founders to watch out for VCs who load up their investments with punitive terms, and then rob founders blind. 

But another version of this story is simply about supply and demand. As one VC described it, liquidation preferences are an elegant solution for clearing a mismatched market:

For example, let's say a founder raised a round at $500m, then the company didn't do as well as hoped, and now realistically the company is worth $250m. The founder wants to raise more to try to regain momentum.

A VC comes and says "ok, company is worth $250m, how about I put in $50m at a $250m valuation?"

Founder says "you know, I really don't want a down round. I think it would hurt morale, upset previous investors, be bad press, etc. What would it take for you to invest at a $500m+ valuation like last time?"

VC thinks and says "ok, how about $500m valuation, 3x liquidation preference?"

The founder can now pick between a $250m and a 1x pref, or $500m and a 3x pref. Many will pick #1, but many others will pick #2.

It's a rational VC offer -- if the company is worth $250m but wants to raise at $500m, then a liquidation preference can bridge that gap. The solution is kind of elegant, IMHO. But it can also lead to situations like the one described in the article above where a company has a good exit that gets swallowed up by the liquidation preference.

FanDuel raised $416 million, including a round that valued the company at over a billion dollars. It makes sense that founders wouldn’t make anything for selling a business for barely more than the amount of money they spent to build it. It also makes sense that, in 2015, when FanDuel was in a massive marketing war with DraftKings to become the dominant mobile betting app, FanDuel might’ve cared more about the positive optics of a huge funding round than they did about the negative terms that were attached to that round. It was an arms race, and FanDuel needed to keep up.

Still, there’s a third version of this story, about incentives. For FanDuel’s founders and employees, the problem wasn’t the liquidation preference per se; the problem was their incentives were no longer aligned with those of their investors. For some investors, a $465 million sale would net a decent return—and it would be especially appealing if the company was at some risk of unraveling further. For founders and employees, a $465 million sale is barely any better than going bankrupt. So investors voted to sell—“it’s just business, you understand”—and founders couldn’t stop them.

That’s the real issue with liquidation preferences. If a company raises money on clean terms, its investors and founders are generally aligned. But for companies that have messy terms, there are often ranges of outcomes where the incentives get very distorted. For example, FanDuel’s founders had no incentive to try to push the $465 million sale into a $500 million sale; they get paid nothing either way. But FanDuel’s investors had every incentive to do it, because for every extra dollar that got added to sale price, they’d get 100 cents of it. And similarly, FanDuel’s investors would be more or less indifferent to a $560 million sale compared to a $600 million sale, because most of those returns would go to founders and employees. Founders and employees, by contrast, would have the opposite incentive—that $40 million would be the only $40 million they get.

That’s because once the sale price clears the liquidation preferences, everyone—founders and investors alike—get paid out relative to their ownership share. Given that FanDuel’s series E investors put in $275 million at a valuation of over $1 billion, let’s assume they owned 25 percent of the company, and would therefore get paid 25 percent of any returns over $559 million. That implies roughly the following returns for those investors:

• $300 million sale: Returns $300 million, or 9 percent.

• $400 million sale: Returns $400 million, or 46 percent.

• $500 million sale: Returns $500 million, or 81 percent.

• $559 million sale: Returns $559 million, or 103 percent.

• $600 million sale: Returns $569 million, or 107 percent.

• $700 million sale: Returns $594 million, or 116 percent.

All of their returns come from increasing the sale price up to $559 million, and not much comes from increasing it above $559 million. For founders, it’s the opposite: They would make $0 in the first four cases, about $6 million in the $600 million sale (assuming they own 20 percent of the company), and about $26 million in the $700 million. 

These incentives dramatically change how people would run a company. Once they took the term sheets with the 3x liquidation preferences, FanDuel’s founders needed to swing for the fences; sell for at least $600 million or bust. For investors, the opposite was true—so long as the company didn’t sink too much, they were probably ok. 

For all of the histrionics about secondary sales distorting incentives, liquidation preferences are far worse. Even if a founder does a large secondary sale, everyone is still aligned: Exiting the company for more is better, and every extra dollar is as roughly beneficial as the last. With liquidation preferences, that’s no longer true. The investors become overwhelmingly incentivized to sell for up to a certain price and then it doesn’t matter that much; founders become overwhelmingly incentivized to sell for at least that price, and everything above it is all that matters. 

1. There are so many potential tie-ins here.“Shut up and drive,” Slootman says to his team.* “Work, work, work, work, work, work. He said me have to work, work, work, work, work, work,” Slootman’s team says to him. “Please don't stop the, please don't stop the, please don't stop the music,” the modern data stack says to the market. “You put on quite a show, really had me going, but now it's time to go, curtain's finally closing,” the market says to the modern data stack. “I don't wanna do this anymore,” the readers of this blog say to me. “I just can't apologize, I hope you can understand,” I say to the readers of this blog.

* Shut Up and Drive is Rihanna’s best song, and I will die on that hill.

2. But what is this suspicious subtweet?

3. To be fair, Ramaswamy is probably a ruthless businessman in his own right. Prior to founding Neeva, he ran Advertising products at Google for 15 years, which grew from $1.5 billion in revenue to $100 billion. For comparison, Nvidia made $1.5 billion in 2002; in 2023, it made $60 billion; over that 21-year period, its stock price is up 57,000 percent. And Ramaswamy built a business that both grew faster and got bigger than that. 

4. He’s not going to raise $7 trillion dollars from the United Arab Emirates, because the United Arab Emirates has an estimated 111 billion barrels of proven crude oil reserves, and oil is currently trading at about $78 a barrel. So the offer would basically be, 80 percent of the country’s oil for, what, 20 percent, non-participating preferred Series A shares, 1x liquidation preference, and one board seat? Tough deal to take. (Also, if Altman did raise $7 trillion, would that make him the first trillionaire?)  

5. This blog is also prone to irrational exuberance…about the new Griff album that drops on April 5. 

6. I wonder if Zuck is regretting renaming the company Meta. It feels like naming your kid Daenerys before the last season of Game of Thrones—he didn’t wait quite long enough to find out who the real hero was. 

7. Is AI improving our lives yet? Just ask yourself: Are you better off now…than you were…three…years…ago?

8. It’s more complicated than this, because there can be other investors who are entitled to getting paid before this investor, debt is almost always paid out before investors, and so on. But typically, founders and employees are the last to get paid.