|
In this issue: - General-Purpose Technologies Revalue Natural Resources—Technological progress means getting more outputs from a given set of inputs—but it also changes which inputs have value.
- Meme Coins—Meme coins as a way to put a price on fame.
- Fees—Hedge funds have captured almost half of gross returns in the form of fees. But that's really the result of two forces pushing in opposite directions.
- Platforms—The software-plus-operating-company model has some unique growth paths.
- Opportunistic Offers—When network effect-driven businesses compete, when one of them has a stumble the other will give them a shove.
- Private Accounting Fraud—Accidentally cooking the books is a problem, even in private companies.
General-Purpose Technologies Revalue Natural Resources
One of the features of the current AI boom is that it's delivering random windfall profits to companies that didn't really expect them. Some business process outsourcing companies are hurting, but the ones that focused on providing training data are flush. Reddit's infinite long tail of content has made its stock worth more; Quora also turned out to be gathering decades' worth of well-organized training data. User-facing products for both business and consumer markets turn out to be useful entry points for AI distribution, crypto miners' power purchase agreements turn out to be worth more for AI workloads than they were for mining. Encyclopedia Britannica turns out to be less in the business of selling books full of information and more in the business of selling high-quality tokens. And, as always, people who bought houses in San Francisco in the 1970s and who've persistently voted against anyone else being able to do so on remotely similar terms are enjoying the fruits of their labors in the form of capital appreciation.
This turns out to be a persistent historical pattern: the deployment of any general-purpose technology radically alters the value of existing resources, mostly revising them upward but sometimes pushing them the other direction. In the end, economic activity is constrained by real resources in one way or another, so looking at how these get repriced is a helpful lens for understanding how new technologies get deployed and what their impact is.
Coal and iron have existed and been exploited for a long time. The Iron Age was actually a notable very early example of technology-driven de-globalization: bronze could be used to make sharper weapons, but it had a more complex supply chain since it's an alloy of copper and tin, and tin wasn't always available in the ancient world. Iron is more abundant, but requires higher temperatures. So there was a gap between when it was a known possibility and the Late Bronze Age collapse that disrupted supply chains and forced the adoption of iron. There are ancient sources and archaeological evidence indicating that coal has been mined for thousands of years, too. But both of these materials got much more important once they were energy and material inputs into, among other things, extracting more of these materials—using coal to power a steam engine that pumped water out of a flooded coal mine, and using the coal and iron to make steel to build the next round of steam engines, was an early example of an economic singularity—you could have predicted that it was a big deal, but couldn't possibly have extrapolated from this to the existence of the transcontinental railroad or the Eiffel Tower.
Some raw materials go through a cycle; for a while in the late 19th century, one of the world's most critical resources was guano, because it was a form of nitrogen-dense fertilizer. This led to a series of wars, which made sense because the world was outgrowing its food supply., That problem was fortunately solved through the Haber-Bosch process—itself one more in a long series of incidents where some complex process requiring idiosyncratic resource inputs gets replaced by one whose main input is energy.
One of the most interesting patterns at work in this is when economic growth turns literal advantages into abstract ones, and then retains those abstractions after the literal benefit has gone. Financial centers colocated with ports are a classic example. A port creates many opportunities for financial transactions: it's a good venue for buying and selling insurance, since valuable ships will be carrying expensive cargo in and out. It's also a good place to trade commodities, because it's where those commodities get graded in the process of being shuffled around—a ship might show up to 17th-century Amsterdam full of exactly one commodity, which would be transferred to a warehouse and then parceled out as a small percentage of the cargo of a series of ships going to different destinations. All of that warehousing, not to mention the business of building and repairing ships, leads to banks and even equity markets. And then those markets retain their network effect even after their original justification has gotten less important; you don't need an office with a view of a port to write options pricing software in C++, but for historical reasons you're a lot more likely to write such code if you have such a view. This happens with some property rights at the intersection of concrete and abstract. A physical communication network needs some path from any given point A to any given point B, and one thing that provides that is an existing transportation network. So railroads sometimes follow a similar path to previous roads, and those rights-of-way can be used for other purposes—Philip Anschutz made his first money in the oil business, acquired the Southern Pacific, and used its rights-of-way to build Qwest.
The Internet speedran this in a few ways: the link graph turned out to be a valuable, un-owned natural resource that was good feedstock for building a search engine. There were also niche proto-businesses that got much larger than they otherwise would have—turned from hypothetical or hobby to a livelihood—because of better ad targeting and various software and physical infrastructure providers who made it possible to host a site, accept payments, store goods, and ship them.
We're in the middle of such a revaluation, where existing corpuses of text, images, and other data suddenly turn into a different kind of asset. They've always had some value, at least in gross terms, but if they were net valuable on average then bit-rot wouldn't be such a problem; there's a lot of content out there that blinked out of existence because it wasn't quite worth the effort to pay the next hosting bill or to archive someone else's site. Now, it seems that the value of archived text and images as model inputs is higher than the cost of storing them—suddenly, bit-rot is an economic inefficiency instead of a default expectation.
Historically, the biggest revaluations have been in the form of human capital. Weighted by chronology rather than population, the historical norm is to treat people outside of the ingroup as a liability, with assimilation being a brutal, bloody process. In an agrarian economy, people are valuable to the powerful, but they have operating leverage—if serfs are able to produce a food surplus, they're a source of wealth, but if they don't produce enough to feed themselves, they're a dangerous liability. When countries get richer, there's wider differentiation in the economic value of different skills and tendencies, which takes the form of increasingly aggressive filtering. One phase of economic growth is mass migration from agricultural jobs in the countryside to manufacturing jobs in cities, but a longer one is restructuring people's entire life paths around identifying talent early and matching it with education, networks, and responsibilities. One reason that process is important is that the richer a country gets, the more its highest-value labor is dedicated to capital formation rather than to producing end goods (the classic software margin structure where a growing business might lose money but still have high incremental margins is just a more visible instance of this). AI represents yet another way to shift to high-value capital formation; models are great at predicting future tokens from a given distribution, and are getting better at extending that in creative directions by chaining outputs together, but they still need more training data. And, of course, they still need good questions.
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - An AI startup building tools to help automate compliance for companies in highly regulated industries is looking for a General Counsel. 4+ years of transactional law experience required; experience working with AI technologies and/or in-house at a software company a plus. (NYC)
- A premier proprietary trading firm is looking for smart generalists to join their investor relations team, working with external investors, rating agencies, and the internal finance team. Investment banking and/or investor relations experience preferred. Quantitative background and technical aptitude a plus. (NYC)
- A well-funded startup that’s building the universal electronic cash system by taking stablecoin adoption from edge cases to the mainstream is looking for a senior full-stack engineer. If you’re interested in using blockchain technology to solve real business problems, this role is for you. (Remote, Singapore)
- Ex-Ramp founder and team are hiring a high energy full-stack engineer to help build the automation layer for the US healthcare payor-provider eco-system. (NYC)
- YC-backed AI company that’s turning body cam footage into complete police reports is looking for a tech lead/CTO who can build scalable backend systems and maintain best practices for the engineering org. (SF)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up. If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority. Elsewhere
Meme Coins
As some readers may be aware, there was a bit of political news over the last twenty four hours, but it's the kind of news where there's a wide range of realities that could be implied by a narrower set of headlines. Rather than analyze the impact of hypotheticals, we can focus on something concrete and tangible: the launch of an official Trump memecoin that almost instantly represented the majority of Trump's net worth. The Diff coincidentally covered meme coins on Friday ($), noting that it's a way to literally capitalize on fame—probably not to raise funds to invest in more fame, but to capture some of the hype before it fades.
The entertainment ecosystem usually shies away from directly capitalizing any one person's celebrity in this way, probably because it's hard to maintain an all-time high without a constantly increasing number of fans, and it's hard to maintain fans when they're all underwater on their financial bets. Instead, what gets capitalized is bundles of transactions (movie studio, for example) or existing intellectual property from existing content. Speedrunning financial history doesn't just mean discovering what the rules are for, but discovering what the norms are for, too.
Fees
Hedge funds have captured 49% of investors' gross returns in fees since 1969, according to a study from a fund-of-funds ($, FT). It's an interesting number, but one that is almost meaningless as an analytical tool because it captures so many disparate features. These include:
- As long as there's such a thing as sales talent, there will always be some fraction of LP capital allocated on that basis rather than on the basis of pure investment skill. And one of the best ways to do finance sales is to take a backtest that looks good and find a story that makes it sound great. If there's a bit of mean reversion to strategies, perhaps driven by investors chasing historical returns and underestimating the impact of luck, then fees will be higher than they look because LPs will tend to overpay after a lucky run.
- Meanwhile, when funds get better at isolating the impact of skill rather than luck, they can charge higher fees because they're selling a more differentiated product. A 12% return that correlates with the S&P is worth a lot less than a 10% return that doesn't correlate with anything.
So what this report is really measuring is, oddly enough, two equally opposed forces that both increase fees. One of those is that getting paid 2 and 20 for luck is just as lucrative as getting paid 2 and 20 for skill, at least in a good year, with the difference being that luck is more likely to mean-revert. If the firms that have a good run raise more money after, then their highest-return years will be on a small asset base and their biggest losses will be on a larger one. Absent a clawback mechanism, that means that total fees divided by total gains go up. But that exact phenomenon is what pushed more investors to demand that funds either control for hedgeable risks or stop charging for them. If you want to invest in a vehicle that produces excess return and doesn't tank alongside the market, you'll pay for it—what's changed over time is that you're increasingly likely to get what you paid for.
Occasionally in public markets there's a vogue for "platform" businesses, where the idea is that the company has operating businesses in some sector and can bolt on small acquisitions, getting economies of scale and using its own exposure to the industry as a way to identify potential winners and dodge some losers. It's a great story for public markets, but it's worked more broadly than that. This intersects in a very interesting way with the recent model of tying together vertical-specific software with operating companies in that same verticle: Metropolis, a parking garage software company turned parking garage operator, is acquiring AI vision startup Oosto at a discount to Oosto's cumulative funds raised. This is a notable advantage to the software-plus-operating-company combination: there are some deals that wouldn't quite make sense for a pure software business, because it's hard to charge for the acquired features, and that wouldn't make sense to an operating business, because it's hard for non-software companies to integrate software acquisitions. But a buyer that's both can mitigate many different risks to doing a deal.
Opportunistic Offers
The more a business is driven by network effects, the less valuable it is to build a perfect clone of all of its features. But that clone ends up being an option: it's an opportunity to either build the network effect slowly and organically, or to suddenly accelerate it when the incumbent stumbles. Meta has used this model with Threads—they were perfectly capable of copying Twitter's features, but the time to do that was when Twitter users were looking for alternatives. Reels had a more organic path, with Meta persistently paying the opportunity cost of missed revenue in order to bootstrap its network effect, but they're also targeting TikTok creators directly, offering them cash to post on Reels first ($, The Information). Network effects are delicate things, especially when some participants in the network are deeply aware of how dependent they are on it—once they get worried, it takes a comparatively small sum of money to get them to switch to a different network, and if enough of them do that, there's suddenly a new incumbent.
Disclosure: long META.
Private Accounting Fraud
Startup investing is more loosely regulated than public markets, in part because the default assumption is that participants on both sides are responsible adults who will take responsibility for their mistakes. There's an economic argument for this, too: if the fixed cost of making an investment is higher (there is no Robinhood of private markets where you can YOLO your money into Databricks because you're bored), the marginal cost of doing a bit more research is lower. But there are limits to this; Grubmarket just paid an $8m fine to settle allegations that it presented investors with inflated revenue numbers. The complaint makes this look more sloppy than malicious: they had two different people preparing financial statements, one of whom had many other responsibilities and the other of whom was the CFO. The numbers didn't match, and when the company realized this, they didn't go back and inform investors who had relied on these numbers. That is the kind of thing that can happen when a company is growing fast and has different segments of its business mature at different rates—basically every company starts out with wonky accounting, and cleans it up before it leads to serious problems. But there's a point on that continuum where it stops being a hack from a scrappy startup and turns into a material accounting misstatement from a company that ought to know better.
|
