In this issue:
- It's Maturity Transformation All The Way Down—The economy has worked out a surprising number of ways for various institutions and individuals to make short-term promises backed by longer-term assets, even if they can't technically meet all of those short-term commitments.
- Weather—A weather prediction market is good for hedging and good for informing policy.
- Adverse Selection—Most advertisers shy away from politically controversial commentators, but sometimes their audience is ideal.
- Compute, and P/S Ratios—It's hard to pump up SpaceX even more.
- Software PE—Waiting for the unlock.
- Automation and Utilization—The more deployment-hours per day you get, the faster automation pays for itself.
Talk to this issue with Read.Haus.
It's Maturity Transformation All The Way Down
Years ago, I worked in the business of "alternative data"—buying things like anonymized spending data, web scraping, web traffic, app usage, etc. and blending all of these together into fundamental predictions. (This is "alternative" in almost exactly the same sense that "alternative" music is—i.e. a recognizable subcategory of mainstream.) Once, I was commiserating with a friend about how one data delivery delay from a vendor led to a cascading series of problems that eventually meant that we missed a delivery date for a report. We’d baked in some time to get the data, clean it, analyze it, and write it up, and while we usually wound up with a little buffer, this time our run of bad luck meant that we missed the deadline. My friend sagely responded: "this industry is maturity transformation all the way down."
Maturity transformation is the alchemy by which banks and other financial institutions can own a portfolio of long-duration assets, but offer depositors (or repo counterparties, or prime brokers, or whoever) theoretically instantaneous liquidity. Calling it "alchemy" is not just poetic—at various points in human history, gold and money have been synonymous, so being able to synthesize a portfolio of mortgages, long-term bonds, business loans, etc. into something that can back a checking account or credit card is a way to literally create what is metaphorically equivalent to gold. This is the closest we'll get to alchemy unless certain approaches to fusion pan out.
Depending on the details of your economic orientation, the fact that the financial system sometimes backs demand deposits and short-term borrowings with longer-term debt is a) just a prosaic fact about the way the world works, b) a marvelous wealth-creation engine when used appropriately, or c) a multi-trillion dollar scam. But if you look beyond the literal case of banking, maturity transformation is everywhere—the modern economy is awash with arrangements that, formally or informally, promise more of something on-demand than the provider can always guarantee. And yet, the system mostly holds together.
One of the ways this shows up is in retailers' inventories and quick-service restaurants' preparation speeds. These are not guarantees; you aren't going to sue McDonald's because the drive-through took eleven minutes one time. But the value proposition is that what you get will consistently meet expectations; they're mostly selling a standard deviation of outcomes, not a mean. For mass merchants, part of their pitch is that they'll have what you're looking for, even if there's a surge in demand—candy and flowers on February 13th (or the morning through afternoon of the 14th), burger patties and hot dogs ahead of July 4th and Labor Day, etc. Given that this kind of demand is not just serially-correlated but sometimes self-fulfilling—people preemptively stock up on what they think stores will run out of—it's surprising that the system stays in balance. For quick-service restaurants, part of how they manage it is precisely the opposite of how pre-FDIC banks did; at a McDonald's seeing a big line means you're less likely to participate in a run on their burger inventory. Retailers have more limits, but they also have some tricks that fast food can't match. Seasonal products are a mix of things that basically only sell in certain weeks (e.g. Christmas wreaths, holiday cards) and things that sell continuously but have demand spikes around particular holidays. So they can avoid the empty-shelf problem by restocking seasonally-fluctuating-demand goods to replace season-specific ones—sell the last black cat costume, and you can replace its shelf spot with a jumbo bag of Snickers that has at least a little demand later on. Inventory also exists beyond the level of individual stores; fulfillment centers can keep inventory balanced across a region, and they can shuffle goods between stores if needed. And in both the restaurant and store examples, app-based omnichannel ordering makes things easier. If someone opens an app, and your guess is that they're likely to buy the last of item X, you might intercept that by giving them a big discount on a more in-stock substitute Y. Essentially, these stores get to perform a shell game where they always present as fully-stocked, by cleverly redefining what that means on the fly—but this ends up being what consumers expected in the first place.
It's easier to deliver bits than burgers, but demand swings can be bigger. It's likely that the biggest single win from the Covid era was Operation Warp Speed, but the second-biggest was that billions of stir-crazy people never lost access to endless high-quality video streaming entertainment. It probably saved us a few riots and revolutions. But Netflix and YouTube weren't designed in advance to keep that many people happily couchlocked at once, nor was the infrastructure through which they're delivered. Netflix agreed to reduce the quality of their streaming in the EU, partly to avert congestion for other parts of the network. YouTube did the same, and video game companies reduced their download speed at that time as well. There was just a lot of streaming, both of entertainment and of work-related conference calls, at a blurrier and choppier resolution than in January of 2020. And yet, the streaming continued. The relevant finance analogy is probably banks imposing withdrawal limits; you can get your pixels, just not all of your pixels all at once.
Some services have had a more pleasant, but no less stressful, demand spike. OpenAI had a big one when they released ChatGPT and created a new category, which they promptly had to throttle and paywall. In their case, revenue was an important consideration, but the other one was finding some way to ensure that users didn't feel that they were getting ripped off—paying customers still got their tokens, and free users couldn't be too annoyed that their research preview didn't come with a rock-solid SLA. In OpenAI's case, they implicitly offered users much more than they could deliver, but that's a nice problem to have. (It was an especially nice problem back then, when the supply of GPUs was a bit more elastic.)
Maturity transformation doesn't exist entirely at the level of companies, or entirely in the world of commercial transactions. In fact, one of the places it's most relevant is in the tradeoff between professional and personal obligations. It's generally possible to be on-call for one of these, but not both. Somewhere or another, either there's a Bagehotian source of emergency liquidity, or something has to give. And if that hasn't come up yet, it means that you're more reliant on credit than you think.
You're on the free list for The Diff. Last week, we wrote about two surprisingly parallel stories of capital allocation that finally met ($), the latest incentives to create volatility ($), and what it means that global trade in AI hardware is reaching levels similar to oil ($), with all the potential imbalances that entails. Upgrade today for full access!
Diff Jobs
Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:
- YC-backed, ex-prop trading founder building the travel-agent for frequent-flyers that actually works (we’ve tried it, it’s magical)0 is looking for a senior engineer to join as CTO. If you have shipped real, working applications and are passionate about using LLMs to solve for the nuanced, idiosyncratic travel preferences that current search tools can't handle, please reach out. (SF)
- A hyper-growth startup that’s turning the fastest growing unicorns’ sales and marketing data into revenue (driven $XXXM incremental customer revenue the last year alone) is looking for a senior/staff-level software engineer with a track record of building large, performant distributed systems and owning customer delivery at high velocity. Experience with AI agents, orchestration frameworks, and contributing to open source AI a plus. (SF, NYC)
- Well funded, Ex-Stripe founders are building the agentic back-office automation platform that turns business processes into self-directed, self-improving workflows which know when to ask humans for input. They are initially focused on making ERP workflows (invoice management, accounting, financial close, etc.) in the enterprise more accurate/complete and are looking for Engagement Leads, FDEs and Platform Engineers. If you enjoy working with the C-suite at some of the largest enterprises to drive operational efficiency with AI and have 3+ YOE as a SWE or MBB, this is for you. (NYC, Remote)
- 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. (SF, NYC, 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)
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.
Elsewhere
Weather
A few years ago, The Diff wrote about parametric insurance ($), particularly applied to flooding. Traditional insurance covers some specific loss (your house burned down, you got sued, etc.), but parametric insurance covers some statistical event that leads to a high probability of loss, like flooding of X inches over Y minutes at location Z. It doesn't hedge as perfectly as traditional insurance, but it can be a lot faster to administer, and it's less prone to adverse selection.
One reason for that is that it can be traded on a market, as is happening with monsoon-linked parametric insurance in India ($, Economist). Different people face different monsoon risks, and some might either be insulated against them or see the monsoon as an uncorrelated risk they can profit from insuring. One of the most interesting features of this kind of insurance is that it reveals some mix of a best guess about what future people expect and a best guess about what future they're dreading, weighted by how much they dread it. And that's a valuable signal for policymakers, if they choose to use it.
Adverse Selection
Polymarket has been running an undisclosed influencer marketing campaign using controversial influencers. This is an unfortunate market opportunity. Historically, one of the tradeoffs an online political entertainer faced was that they could be blandly mainstream and have a small audience, or they could throw some bombs, get a big audience, and lose their advertisers. You used to be able to measure just how controversial a commentator was by what kind of health or workout supplement they were shilling. But in the case of prediction markets, that kind of controversy-courting is great for business! The kinds of people who like fringe opinions are the kinds of people who will take the long side of long-shot bets and persistently bleed money to liquidity providers (or, occasionally, have a big win that they then talk about on social media). There are plenty of advertisers who deliberately target less sophisticated customers, but few are in such a good position to systematically extract money from them. All of this is particularly unfortunate given that prediction markets are such an effective tool for aggregating informed opinion; there are probably more socially-optimal ways to subsidize the price-discovery side of these services without encouraging them to part the foolish from their money, even if that's an inevitable side effect of this kind of betting venue.
Compute, and P/S Ratios
Google is temporarily buying $920m/month of compute from SpaceX. Since Google owns about 5% of SpaceX, and since SpaceX plans to go public at more than 20x revenue, this is technically accretive—but it's year-one accretive, not indefinitely so. Google is not really at a point where they're dependent on getting markups like this, and it's also not likely that SpaceX investors are going to mechanically mark up the company's valuation by x% for every x% revenue increases—the sophisticated investors are putting different multiples on different lines of business (and would put a low one on this one, given that it can be terminated by either side with 90 days' notice). And the Musk fans are betting more on a narrative—an accurate one!—rather than specific current numbers. FY26's cloud computing revenue number matters insofar as it gives SpaceX the cash flow to hit FY36's orbital computing revenue number, but is not that important on its own. Overall, this is a pretty mysterious deal, but the Occam's Razor interpretation is that there really is a compute shortage, and SpaceX is one of the rare companies that can fill in the gap.
Disclosure: long GOOGL.
Software PE
Historically, for many software companies there's a level of spending on sales and marketing, and on R&D, that makes sense if they expect to grow indefinitely. And there's a much lower level of spending that implies that they're going to keep getting more revenue from existing customers, but won't actually take over the world. Some of them can gracefully shift to the second model over time, but the natural way to do it quickly is to sell to PE. Which is rarer this year than any time since the start of Covid ($, FT). We're at a very strange time in the software industry, because AI offers one of the most legible paths to radical cost cuts for software businesses, which ought to make PE deals easier to underwrite. But no PE firm wants to acquire a company thinking they'll harvest high cash flows for a decade only to find out that the business evaporates in a few years. And this caution also means that software companies aren't spending as much on one another's products. The current setup means that there's a lot of pent-up demand, for both software and software equities, that can be unlocked as soon as deals start happening.
Automation and Utilization
One of the things that makes big consumer packaged goods companies hard to compete with is that their delivery network is denser, so they can manage inventory more efficiently and have a lower cost of delivery. This also makes them ideal early adopters for driverless trucks, which Pepsico is now using ($, WSJ). The best place for a driverless truck is somewhere that trucks are running 24/7, where the fixed cost of automation can be amortized over as many ton-miles as possible. That also means that each vehicle is picking up more training data per day of deployment, so it's a data subsidy to future use cases that can't support quite as many hours per day.