Rumor Markets

In this issue:

• Rumor Markets—Prediction markets will get bigger and more socially desirable if they help investors in existing asset classes isolate particular risks that they either have a strong view on or want to hedge. And this creates a new way to filter news, which is valuable enough that it can subsidize trading in prediction markets that wouldn't otherwise be viable. There will be plenty of legal excitement to work through before this really gets going, but it's too interesting to ignore.
• Legitimacy—Cloudflare is big enough that it has both the ability and obligation to rewrite the rules.
• Tariffs—The cost hasn't gone away, but the uncertainty has gone up.
• AI and Complements—Meta would like to be the company that builds AGI, but META, the stock, is more tied to the use of AI than to who owns it.
• Tariffs and Elasticity—The US manages to punish Russia by way of India.
• Economic Nationalism—Tariffs only make "Made in the USA" competitive if the US is actually competing in the relevant product category.

The Diff September 2nd 2025 14:49 • Click to play audio

Rumor Markets

Hayek has a wonderful essay on prices where he articulates the big positive externality of a price system:

Assume that somewhere in the world a new opportunity for the use of some raw material, say, tin, has arisen, or that one of the sources of supply of tin has been eliminated. It does not matter for our purpose—and it is very significant that it does not matter—which of these two causes has made tin more scarce. All that the users of tin need to know is that some of the tin they used to consume is now more profitably employed elsewhere and that, in consequence, they must economize tin.

Beautiful!

But, also, kind of ugly: it means that the price system is a way to act on information while being completely indifferent to the details behind it. Is tin demand up because economic growth has increased the demand for solder? Because a developing country that used to have a more uncertain food supply has started canning food to reduce waste and increase storage? Or because of some horrific disaster at a tin mine? Thanks to the price system, you can gracefully adjust your life to the material consequences of that last tragic outcome without ever being aware of it.

Of course, you don't have to feel especially guilty about this; in a pre-globalized world, maybe the same disaster in the same part of the world, with the same human toll, wouldn't register at all. But the fact that you can compress many dimensions of information into a single number you have to react to remains both helpful and a little disconcerting.

While you can't truly unbundle or "decompress" a price into all of the inputs that went into it, the more complex the financial system, the more room you have to correctly interpret prices. For example, if current-month tin futures spiked but future months are unaffected, that's probably a supply story; if future months' delivery prices creep higher faster than current months', that makes sense in light of a change in demand. And once you have options, and other prices, you can infer still more; if options imply a bimodal distribution, there's some specific piece of uncertain information that market's waiting on, and if prices of other metals happen to move alongside tin, and are found in the same deposits, that can help you identify which specific location is driving prices.

But all of this is specific to commodities, and in fact a bit fanciful; there might be cases where having comprehensive knowledge of global resource extraction, storage, and end uses might let you look at all this data and say "Huh, looks like things are a little backed up at Busan, but it looks like everyone's convinced that'll be resolved by the weekend."

What you really want is a full spectrum of prop bets that will let you either express or measure a consensus view on as many topics as possible. There was a fun talk at Manifest a few years ago about doing this: you might create markets for individual stages of FDA trials, or a conditional market on how many H20s will be purchased in China next quarter if they're legal to purchase, and a separate market on whether or not they'll be legal. This is a very fun idea, especially for anyone who's always dreamed of their name coming after the v. in a precedent-setting insider trading case.^[1]

But let's suppose we find a set of predictions it's legally safe to make. Maybe you focus on commodities markets, where it's hard to have any hedging if you don't let companies let companies trade on information like "we just found a lot of oil and want to hedge the risk that it goes down in value before we sell it." Or maybe you let people make predictions about material developments for specific equities, but you end that market at the point where someone would potentially have sensitive information. So if Disney is releasing an original movie, you can let people bet on its box office results only until the premier.^[2]

What you get if you can pull this off is a solution to three distinct problems:

1. It makes markets higher-resolution because you can isolate the impact of different variables. Instead of having factors that cover multiple companies, you can also look at multiple, company-specific factors—how much of a given company's market cap is attributed to specific divisions, how much value investors think the CEO adds, what they think of the capital allocation, etc.^[3]
2. It makes prediction markets in general more useful, because suddenly there's a way for market-makers to hedge—if you're long a contract that bets on hail somewhere, you can hedge by going long whichever insurer is most exposed to correlated losses from hail damage in that area, for example. Or you can go in the other direction, and own or short the underlying while using a prop bet to hedge some specific risk where you don't have an edge. This adds something critically important to prediction markets: noise traders who are not trying to make a profit on every trade, but are accepting negative expected value on average in exchange for lower risk on other trades they make.
3. It creates raw material for a continuous, rather than discrete, news service.

This last one is particularly fun. Services like StreetAccount are great at providing a steady stream of emails about basically every development in a given equity market—upgrades and downgrades, product releases, conference presentations, earnings, etc. They'll even send along rumors. But deciding which rumors to share is a tricky, qualitative question. Is a 1% chance of something happening this year worthy of clogging someone's inbox? What about 10%? And how does it vary by company, and by job? A 20% chance that the CMO of a restaurant chain is suddenly resigning isn't very interesting; a 20% chance that the CFO of a company whose revenue and net income have accelerated but whose operating cash flow is flat gets a lot more interesting.

And this actually creates a mechanism to subsidize these prediction markets. There are many traders, systematic and discretionary, who care deeply about rumors, whether they're trying to front-run the rumor turning into a fact—once a company is a likely M&A target, other potential bidders will take a look to make sure they don't miss it, while the shareholder base will shift away from longer-term holders and towards arbitrageurs who favor a deal. Other traders will sometimes know a company or its industry well enough to underwrite a rumor, and will be a lot more confident making a trade if they know precisely what they're betting on or against. In fully systematic strategies, you might not even make a directional bet at all; an options market-maker may want to widen the spreads that they quote, without choosing a directional view, if they see that the outcomes for a given company are looking more binary than usual.

This is especially useful for long-tail macro risks that are easy to ignore early on. If you've invested in some company that sources its cotton from Uzbekistan, or that does 10% of its revenue in Brazil, or whose biggest single-source supplier's biggest single-source supplier is in Indonesia, there will occasionally be some episode of unrest that just completely ruins your day. You might see it coming, with the right mix of Google alerts and skimming The Economist. But you'd be a lot more likely to see it if you set some very sensitive alert that triggered when the odds of disorder hit, say, 5%—below the point at which you'd expect global news outlets to cover it, but at the point where you could spend a few minutes skimming local news sources and figure out some combination of how big a deal it is and how you might monitor it more closely.

All of this would be nice, albeit somewhat expensive. It's a little cheaper if you can use LLMs to crank out ideas for prop bets, or use them to scan a huge number of bets to look for likely opportunities. And even if it takes off, it'll remain a niche business in comparison to the business of trading the underlying assets. This combination of gambling and financial services naturally lends itself to crypto, either as a means of payment to accept and a term to be repeated ad nauseum in the S-1, or as an actual means to monetize the hype behind this idea in the form of an ICO, so there's an upside piece to the business model as well.

A paradox of the growth of markets and the financialization of the economy is that we should expect two things to be true at once. First, the biggest companies will be much bigger, in part because big companies are selling inputs to other companies and those other companies will be able to scale their spending faster with efficient capital markets—it's a lot easier to move GPUs if every plausible GPU-buyer (or renter) can raise a ton of debt and equity. But at the same time, liquidity for the big asset classes creates hedging opportunities for smaller ones, so that concentration is a complement to diversification in the other direction. The US market is more concentrated than it's historically been, but the menu of ETFs is also broader than ever. An ETF is, in part, a way to construct a niche bet out of a series of more general bets. Prediction markets offer another way to do that, with less liquidity because they aren't entirely composed of existing liquid assets, but with more interesting use cases for basically the same reason.

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• 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.
• A transformative company that’s bringing AI-powered, personalized education to a billion+ students is looking for elite, AI-native generalists to build and scale the operational systems that will enable 100 schools next year and a 1000 schools the year after that. If you want to design and deploy AI-first operational systems that eliminate manual effort, compress complexity, and drive scalable execution, please reach out. Experience in product, operational, or commercially-oriented roles in the software industry preferred. (Remote)
• Fast-growing, General Catalyst backed startup building the platform and primitives that power business transformation, starting with an AI-native ERP, is looking for expert generalists to identify critical directives, parachute into the part of the business that needs help and drive results with scalable processes. If you have exceptional judgement across contexts, a taste for high leverage problems and people, and the agency to drive solutions to completion, this is for you. (SF)
• Thiel fellow founder (series A) building full-stack software, hardware, and chemistry to end water scarcity, is looking for an experienced software engineer to help build the core abstractions that enable global cloud seeding operations - from mission planning to post-flight analysis. If you have 5+ years building production systems with complex integration requirements, please reach out. (Los Angeles)
• A leading AI transformation & PE investment firm (think private equity meets Palantir) that’s been focused on investing in and transforming businesses with AI long before ChatGPT (100+ successful portfolio company AI transformations since 2019) is hiring experienced forward deployed AI engineers to design, implement, test, and maintain cutting edge AI products that solve complex problems in a variety of sector areas. If you have 3+ years of experience across the development lifecycle and enjoy working with clients to solve concrete problems please reach out. Experience managing engineering teams is a plus. (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.

Elsewhere

Legitimacy

A long-running theme in The Diff is that big tech companies have government-like incentives and government-like problems. But just as every actual government has to deal with a different set of inherited problems, every tech platform has a different set of tools and constraints, but also a different set of advantages. This interview with Cloudflare CEO Matthew Prince showcases both. In a way, Cloudflare's parallels to governments are the most abstract, because what they do is so invisible to the average person who benefits from their existence, but also incredibly concrete, since Cloudflare is basically in the businsess of deciding who crosses which border, and on what terms. These are digital borders between someone using a browser and someone operating a website, but they face the same general shape of the problem—the more breadth there is to commerce, the more upside there is from classic Ricardian comparative advantage, but also the more negative externalities there can be.

The irony in this analysis is that Prince is doing a very governance-realist thing, telling AI companies that Cloudflare's customers will decide who uses their data, and how, and that Cloudflare will determine the details of facilitating this. But the mechanism involves free-riding on existing governments! "[T]he magic of Cloudflare (is, depending on where we put server farms to host content), we can make you and everybody else a UK publisher, or a Michigan publisher or an Australian publisher for our good friends at News Corp. Wherever it is that the laws are the most favorable, we can say, 'Content now lives there.'" So if there's some jurisdiction with a uniquely savvy view of IP rights in the AI age, then everyone with valuable IP is going to use Cloudflare to teleport theirs to that jurisdiction, in the same way that so many transactions between New Yorkers and Californians are ultimately governed by the laws of Delaware. For Cloudflare, the diversity of jurisdictions means that any given government can't so much write rules as propose them, while Cloudflare can't quite invent its own rules but can opt in to the best available option. That actually creates a governmental niche, of being the Dubai of AI, i.e. a jurisdiction that's heavily-regulated enough that it feels legitimate to do business there, but loosely-regulated enough that the business you'd want to do is legal there.

Tariffs

Late on Friday, a large set of Trump's tariffs were ruled illegal by a federal appeals court, though they'll be in place through mid-October to give time to appeal. From a long-term pragmatic, rule-of-law and comparative-advantage perspective, it's ideal for the judicial branch to apply a gentle corrective to the executive, with a lot of time to resolve things according to the normal procedures before anyone else has to change their behavior. Of course, to anyone who actually relies on imported goods, all of that is a lot worse than just swiftly and unilaterally striking down tariffs. Importers have to deal with all of the cost and inconvenience of both paying for tariffs and figuring out exactly what they owe, all while knowing that any effort to streamline or automate that process is potentially wasted. There is still plenty of room to renegotiate the global trade system and the US's role in it, but at a minimum that requires knowing what the rules are, how much they can change, and who gets to write them, all of which are apparently open questions right now.

AI and Complements

Meta may use AI tools developed by third parties, like Google's Gemini models, to answer some queries ($, The Information). This is, in part, the kind of story you leak in order to get OpenAI and Anthropic to swoop in and offer a similar service at a lower price, but it does illustrate Meta's unique economic exposure to AI. If generative tools mean that there's more stuff out there that the average person finds appealing, then the biggest platform is likely to reap the biggest benefit, since it can identify winning slop and then ladle it out to the biggest possible audience at the fastest imaginable speed. It would certainly be convenient for Meta if they had the best models, but it's ideal for their business to have the best distribution for good-enough models, and then find a way to capture the upside from upselling their users on even better ones, whether or not those were built in-house.

Disclosure: Long META, GOOGL.

Tariffs and Elasticity

One of the fun elements of historical tariff fights is that retaliatory tariffs target very country-specific products—Harleys, Kentucky bourbon, Wisconsin cheese, etc. Some of this just to make a splash, and some of the motivation is that if you put a tariff on some bulk commodity, all you're really doing is making everyone reroute shipments—they'll buy the same sorts of soybeans from Brazil instead of the US, and someone who would have bought Brazilian soybeans will buy American ones, instead. These products still do get tariffed from time to time, but the impact is limited and hard to predict.

And sometimes, it's completely different from what you'd expect: US tariffs on India actually decreased India's after-tariff cost of Russian oil slightly ($, WSJ). US tariff policy today doesn't exactly seem to be the result of sophisticated economic modeling, but at least in this case, the US managed to punish Russia with lower oil revenue, and to mitigate the extent to which that flowed through to India.

Economic Nationalism

For a brief period, when tariffs dominated more headlines, online shoppers would deliberately search for "Made in the USA" products to get the best deals. But global supply chains are pretty efficient, and manufacturers tend to operate at whatever scale was optimal; there are some industries where American and Chinese companies sell the same product and their cost gap is just small enough that tariffs make the American-made version instantly competitive with the Chinese one, but there are plenty more cases where America stopped making the product, because we have a comparative advantage in things like writing the ad copy and managing the working capital, while other countries have a comparative advantage at converting steel and computer chips into appliances. So, Amazon users have mostly reverted to their pre-tariff behavior, and aren't actively seeking American goods. The more credible tariffs are, the more quickly capital and marketing would adapt to a new normal where the US is a competitive manufacturer in many categories. But the first round of tariffs was sloppy enough, both in terms of who paid what and in terms of the legal architecture around them, that most people correctly assumed that the best option was to ride out a temporary cost increase and expect the status quo to return soon.

Disclosure: Long AMZN