In this issue:
- The Super-Rich as Financial Sin-Eaters—An institutional framework is a good way to manage known problems, and markets have gotten pretty good at matching the right kind of capital to the right investment vehicle. But what's left over is exploring the as-yet-unoptimized areas. It's a useful social technology that this risky, sometimes negative-EV practice mostly hits the balance sheets of the very rich.
- Adversarial Adoption—People who don't much like AI are unusually vulnerable to being exploited by it.
- Rates—Countries can't grow the promises they make faster than their economic output indefinitely, but only one kind of promise gets regularly marked to market.
- Scale—It keeps getting harder to launch a new hedge fund—or to justify doing that instead of working at an existing one.
- Talking Your Book—If there's going to be growth, some people would rather it be more predictable than as fast as possible.
- Revenue—A game of accounting-adjustment telephone wiped out more value than strictly necessary, but when Anthropic's S-1 comes out, we'll have a standardized way to think about labs' financials.
This issue of The Diff is brought to you by our sponsor, Perk.
Talk to this post on Read.Haus.
The Super-Rich as Financial Sin-Eaters
Modern financial systems offer everyone a simple social contract. You have two choices:
- There is abundant information about index funds, retirement accounts, what fraction of your income you need to save at a given age in order to have some level of annual spending power from age 65 onward. The weekend after your first week in the workforce, you could spend an afternoon reading up on this stuff, an hour setting things up, and then give it minimal thought from then until retirement.
- On the other hand, as Heinlein says: "Come on, you apes! You wanna live forever!?" You could also ignore the math on diversification, volatility drag, and tax efficiency, disregard how much effort goes into keeping prices roughly efficient, and take some big bets.
Almost everyone should take the first option. But there's a calibration problem, which you can phrase in two ways. The first is: if everybody's just investing passively, then the only people in the world trying to figure out which companies are better or worse than one another are either a) management, when they decide whether to raise or return capital, and b) the committee deciding which companies to add to or remove from indices. The other way to put it is that passive investors free-ride on the efforts of active investors, and that the fewer active investors there are, the more rewarding it is to switch.
Markets are continuously working towards a perfect balance between some investors passively harvesting various simple factors—equity risk, duration, etc.—while others do a mix of turning strategies into cheaply-replicable factor products or actually finding interesting sources of returns.[1] This balance gets upset from time to time, by a few different forces:
- The supply of companies isn't fixed. When markets are more efficient, they also tend to trade at higher multiples, which encourages more companies to go public—meaning that there are new business models for analysts to compete to figure out, and also that there's a new supply of less price-sensitive sellers in the form of founders, early investors, and employees.
- New kinds of buyers entering or exiting a market can throw prices out of whack or make them more efficient. There was wide dispersion in Japanese equity valuations a few years ago, partly because bad corporate governance creates a lot of value traps.[2]
- Some collision of technology, business models, and zeitgeist that leads more retail investors to make their own trading decisions. There was some of this in the 1960s, when mutual funds exposed more people to the possibility of making money in the stock market, and when celebrity fund managers like Gerry Tsai, Jack Dreyfus, and Freds Alger, Carr, and Mates were getting famous. On one hand, the implication was: they're really good at this. On the other hand, the implication was: you might be able to make money, too.
- Some changes in industry fundamentals happen slowly enough that they won't be accurately captured by traders who are looking for repeatable signals. Some industries evolve away from being cyclical, or collapse from differentiated businesses to commodities. Sometimes, they find the right way to monetize their products—subscriptions instead of one-time sales, advertising instead of subscriptions, etc.—which makes their fundamentals look temporarily worse but also makes them more valuable in the fullness of time. These changes happen in increments, so shorter-term fundamental trading captures some of them. But a well-tested, repeatable model for alpha generation persistently puts you on the wrong side of the trade.[3] This is one reason public equities aren't as diverse as they used to be: venture and growth investors specialize in manufacturing companies that break some of the rules, and companies going through a complex multi-year transition often do it under the ownership of PE. Which, naturally, means that the investors who specialize in underwriting this kind of bet gravitate away from public equities, making it harder for those companies to go public, in a cycle that's virtuous or vicious depending on whether or not you have carry.
This creates a kind of market niche for handling complicated, hard-to-diversify bets that also require full attention. It's just hard to fit that kind of thing into a typical investment vehicle's framework. But this is one of the functions of rich people with idiosyncratic views about the world: they take on a disproportionate share of that risk, mostly by continuing to own big chunks of the companies they found.
One great example of this is Larry Ellison. It may or may not be a great idea for Americans' retirement savings to be partly contingent on whether an octogenarian executive chairman can successfully pivot a database company into being a levered-at-many-levels AI infrastructure provider.[4] It may work, but it's not the decision most people would consider best on a risk-adjusted basis. Fortunately, someone investing in a float-weighted index fund is exposed to only about 62% of their share of the world's total Oracle-risk, because Larry owns the rest. Other big AI plays tend to have high insider and employee ownership, too: they're often newish, founder-led companies that make aggressive capex decisions.
Risk is only a meaningful concept when it plays out unpleasantly, so it's worth mentioning examples of that: Eike Batista loved oil, mining, infrastructure, and leverage; his net worth peaked at $35bn and dropped 99% by the next year, before going negative. Brazil is a huge country with abundant natural resources, and institutions tend to improve over time, or at least mean-revert. His bets weren't crazy other than in comparison to his bankroll, but if he hadn't been the one to lever up this way, someone else would have done something similar. Maybe they would have peaked at $10bn, maybe at $50bn, but when there's the right confluence between opportunities and capital to support them, someone will show up, and if the capital is a little miscalibrated, they'll put a lot of money to work before the law of large numbers catches up with them. And it often does! Many of these bets do not, on their own, have an expected value that justifies the risk in question. Competent institutions will consider them, and turn them down, because the numbers don't pencil out. But in the end, that means that these institutions are optimizing, rather than creating. Sometimes, you just have to blow a lot of money to know whether or not there's anything there.
It's basically impossible to design an institution to take these kinds of risks, because it will inevitably be captured by people who are good at performative risk-taking that truncates both tails of the bell curve. The Gates Foundation and some EA efforts come close, but these are also pretty low-bus factor situations, where it really is about particular people with distinctive views. Institutions are good at scaling something with measurable goals, but they're terrible at dumping ridiculous sums into something that might be an abject failure. When state capacity is high, governments can pull it off, but right now you're basically counting on the billionaires.
Regular investors don't think of themselves as "harvesting risk premia" when they put money in an index fund, which is fine. It's a counterintuitive way to think about what they're doing, and it doesn't add much to the analytical process. One reason it is useful is as a reminder that part of the role of household balance sheets is to take temporary losses during market downturns. If you're saving for retirement, the reason your wealth compounds at something like 8-9% rather than 4-5% is that it will periodically have 30%+ drawdowns. But a retirement saver is positioned ideally for this; when you're 65, it doesn't really matter if you were poorer at 45 than you were at 40 because of a big market decline. That's water under the bridge! Older savers do tend to shift their savings away from equities and toward more predictable products, though this means they're taking some inflation risk. But older people are also less exposed to inflation in absolute terms—less driving, less eating, fewer big purchases, etc. ↩︎
If you do a screen for companies that have a low EV/EBITDA and rarely or never lose money, you get a list of companies that have a lot of cash on hand, which will typically stay on-hand. ↩︎
For example, if an industry's sensitivity to macro has stabilized, as has happened with airlines, they'll all look persistently expensive, and whichever one has done the best job of cleaning up its capital structure and investment approach will tend to look most expensive. In the other direction, switching from upfront sales to subscriptions kills current-year revenue for any product that customers don't stop using after a couple months. But if a company makes $1 from software that customers upgrade every five years, and switches to making $0.40/year from selling continuous access, it gets optically more expensive but is actually a more valuable company. ↩︎
Obviously easy to make fun of, though Larry Ellison has presumably learned a thing or two in his half-century with Oracle, including surviving multiple corporate near-death experiences (and the literal kind, too). But there's also a precedent: the single American who made the most money from the last buildout to hit the same neighborhood of capex as a percentage of GDP, Cornelius Vanderbilt, spent his early career in the steamship business and didn't shift to being mostly a train guy until his late 60s. ↩︎
You're on the free list for The Diff. Last week, paying subscribers read about how AI will be increasingly effective at A/B testing compelling content ($), some unusual ways people have hedged ($), and why it's hard to roll your own AI-powered hedge fund or prop trading firm ($). Upgrade today for full access.
A Word From Our Sponsors
Your team is committing expense fraud, and it might not be their fault.
1 in 5 employees misrepresent expenses regularly. New research from Perk shows how it happens and why the receipt can no longer be trusted as proof.
When expense misrepresentation is this common, checks after spend happens aren't enough. Perk helps modern CFOs and their finance teams set card controls, limits, and approvals before spend happens.
Elsewhere
Adversarial Adoption
Yet another literary prize has been won by a story that Pangram sees as AI. (Pangram has a low enough false-positive rate that you basically need that plus a lame excuse to be certain.) It's been interesting to see who's most vulnerable to AI. Sometimes it's the usual suspects: any process that tries to throttle access by requiring people to write a few hundred words or read a few thousand is suddenly inundated with requests, for example. But there are some people who've made themselves vulnerable because they're already running a reinforcement learning process on humans, and AIs are far better at matching what they're looking for. It's always awkward to say that some piece of short fiction is really great, because there's probably something to quibble with. Which makes judges risk-averse. Someone who's mentally modeling someone else is going to have a somewhat two-dimensional view of them, so that naturally leads to more predictable criteria for selection. That wouldn't be a big deal if judges in these contests regularly used AI, and knew what AI-inflected text sounds like. But if they don't, they'll tend to love it, because it excels at good-enough-sounding prose.
Rates
The Economist looks at which countries can outgrow their debt, and which can't ($): France, Italy, and the UK are worst-positioned. Aging does strange things to a country's ability to borrow: at one level, it creates a population of willing lenders who are looking for low-risk fixed income. And it leads to slower growth, which lowers interest rates. But eventually countries reach a point where the legal obligation of bonds and the political obligation of social spending capture enough of a country's output that they slow its growth. Barring a big acceleration in GDP growth, the 2030s may be the decade where pension recipients and bondholders have some zero-sum arguments over who gets what fraction of a given country's taxes. Whoever restructures first probably has the easiest time.
Scale
It's getting harder for managers to launch new hedge funds, as opposed to working for one of the big ones. It's an interesting industry structure, because historically, single-manager hedge funds were a viable business model, and someone who wanted to construct the return profile of a multi-manager fund could do so by, well, hiring multiple managers. They'd miss a few things with that approach, though: they might not notice new strategies, and hire practitioners before those strategies got big and efficient; they wouldn't be able to amortize infrastructure and data expenses over multiple teams; they wouldn't have cross-margining, where some parts of the fund could offset one temporary loser, etc. There are just many cases where putting strategies together—the less related to one another, the better—produces a higher return for a given level of risk. Some of those factors could have gone differently, but the balance sheet part is hard to ignore: a drawdown is just easier to ride out if it affects 5% of a firm's assets rather than 100% of them. There are plenty of strategies that don't fit a pod shop model, but the pod shops are always open to adding new sources of returns that increase their capacity, and over time the scope will keep expanding. In the end, pod shops are in the business of bundling together different kids of risks and different kinds of liquidity-intensity into something with a reasonable amount of both.
Talking Your Book
AI labs, chip companies, and investors want different things from AI. The labs tend to want to build the most powerful intelligence they're sure won't kill us all; investors want the most lucrative AI that won't kill us all (but they're less worried about that risk than the labs). But hardware companies want a steadier, more predictable growth rate that allows them to make sensible capital allocation decisions, enjoy predictable growth, and never get stuck with stranded assets. This is one reason Hock Tan is cautioning investors about their AI enthusiasm: Broadcom benefits from growth, but gets penalized, both as a stock and as a business, if that growth is partly contingent on big inflows of capital that could disappear at any time. The more AI spending is supported by the cash flow from inference rather than cash from investors, the higher a multiple suppliers get.
Revenue
Last week, AI stocks sold off because OpenAI's reported revenue was lower than expected. But this seems to have been a game of telephone: OpenAI and Anthropic report their revenue differently, with Anthropic including revenue paid to third parties for distribution and OpenAI omitting it. You can make these consistent by either grossing Anthropic down or grossing OpenAI up. Either one works, and it's mostly a question of whether you want an impressive revenue number or better margins. (Anthropic caught a little flak recently for using the grossed-down number when talking about its incremental margins ($, FT).) AI labs are big, influential companies whose financial reporting doesn't match that of similarly important businesses, which can lead to situations like this. It's temporarily bearish, if it means that the marginal price-setter in AI was double-counting some of OpenAI's growth—even if you weren't personally thrown off by this, it's always a little worrisome to learn that other people in the same trade as you weren't that well-informed. But only temporarily; once Anthropic files an S-1, they'll set a standard for what disclosure looks like, and more analysis of OpenAI will be explicitly benchmarked to how Anthropic reports things rather than fuzzily compared to it.
Diff Jobs
Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:
- General Catalyst-backed founder is looking for technical generalists who are excited about creating a fully autonomous firm. If you want to own the entire process of identifying, instrumenting, steering, and monitoring post-PMF businesses that run autonomously, and study how machine intelligence actually behaves in the real economy, this one is for you. (SF)
- a16z, Sequoia, Thrive-backed next generation pharma company that instruments the best drug hunters in the world with deterministic software, data and AI to accelerate the process of acquiring, developing, and monetizing promising therapeutics (three active immunology programs) is looking for senior software engineers. If you are excited about sublinear headcount scaling and building technology that drives therapeutic abundance, please reach out. (Boston, NYC, SF)
- A top prop trading firm is looking for people who combine exceptional quantitative ability with the judgment and relationship instincts to build critical financial partnerships and lead high-stakes negotiations. Candidates may come from markets, finance, or further afield; what matters is this combination of abilities, strengthened by the perspective that comes with experience. (NYC)
- A top prop trading firm is looking for people with exceptional strategic thinking and quantitative skills to help the firm model and manage its financing risk and strategy. Open to candidates from a range of analytically demanding fields. (NYC)
- Ex-Bridgewater, Worldcoin founders using LLMs to generate investment signals, systematize fundamental analysis, and power the superintelligence for investing are looking for machine learning and full-stack software engineers (Typescript/React + Python) who want to build highly-scalable infrastructure that enables previously impossible machine learning results. Experience with large scale data pipelines, applied machine learning, etc. preferred. If you’re a sharp generalist with strong technical skills, please reach out. (Remote)
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.
And: we're now actively deploying capital into early-stage companies through Anomaly. Our focus is on defense, logistics, robotics, and energy. If you'd like to chat, please reach out.