The Lots-of-Low-Satiety-Calories Business Isn't What it Used to Be

In this issue:

• The Lots-of-Low-Satiety-Calories Business Isn't What it Used to Be—Consumer packaged goods stocks have done well for investors over many decades. Just not the last one. It's interesting to explore why.
• The Other Net Present Value—A company is worth whatever change in cash flows it produces to its owner, which is not the same thing as the cash flows it produces on its own.
• Capitulation—It's tough out there for short-focused investors; it's a good strategy but not an easy to sell financial product.
• Insurgent Advertising—Why Lyft wants you to think about how Uber is more popular.
• Alignment—The AI tools we're allowed to use need to be as powerful as the ones adversaries choose to use.
• Math—Thoughts on conjectures.

Talk to this post with Read.Haus.

The Lots-of-Low-Satiety-Calories Business Isn't What it Used to Be

The single best-performing stock over 1926-present, the period covered by the CRSP database, is Altria, formerly Philip Morris. It grew an investor's wealth 2.65m-fold over that period. If you're trying to reverse-engineer durable long-term returns, you might look at what drove that:

1. It's a consumable product. Or, put in terms more relevant to analyzing growth stocks today, Philip Morris applied a usage-based pricing model in which customers who liked it more spent more on it. In fact, addiction makes willingness to pay rise (or at least stay inelastic) as consumption increases—it’s easier to kick a “one bummed cigarette every few months” habit than a two-packs-a-day one! So, in SaaS terms, the higher a customer’s “ACV” the higher the future NDR, up until some saturation point, at which you have a fairly long duration annuity at a relatively high level of spend (though the higher the level, the shorter the duration.)
2. There's more brand differentiation than product differentiation. The underlying product is nicotine, and while there are many delivery mechanisms, the cigarette turned out to be the logistical winner for a long time: a pipe requires bulkier equipment and some setup time, a cigar assumes that you're going to be doing little but nicotine consumption for an extended period, but cigarettes can be consumed quickly, or continuously. It helps those 1926-and-onward returns that cigarettes got incredible product placement a decade earlier, when they were included in soldiers' rations, and those soldiers later came back to countries where the economy was booming and the labor force of young men was a bit smaller than it had been a few years back, i.e. they were primed to enjoy some new kind of consumption.
3. They're addictive, though this has an ambiguous effect: great for business in the short term, but, bad news on many dimensions if customers get addicted to a product that kills them. This made tobacco a target for regulators. Which, as those long-term returns show, is not the worst thing that can happen to a business; what the US did, in effect, is what plenty of other countries do: nationalize the tobacco business and have it operate as a monopoly, because monopolistic pricing discourages consumption and also raises more revenue. In the US case, they let the private sector keep running it. under constraints. But if you look at the price of a $10 pack of cigarettes (the national median), roughly $3 of that will be state and federal sales/excise taxes, another dollar will be for funding settlements with the states, and the manufacturer and retailer together earn enough contribution margin to add another dollar or so in tax liability. Or, put another way, cigarettes are an 80% net profit margin business in which the government is a 62.5% shareholder.

Warren Buffett, fond of one- to two-page offers for sometimes-complex businesses, naturally had a pithier explanation: I'll tell you why I like the cigarette business... It costs a penny to make. Sell it for a dollar. It's addictive. And there's fantastic brand loyalty." Buffett doesn't mind telling jokes, sometimes on fairly dark topics. But his usual public-facing persona is a little less willing to talk about a direct link between profits and negative social effects like this. To be fair, he wasn't speaking publicly; this quote comes from Barbarians at the Gate, and he's no doubt steamed that somebody shared that memorable zinger with the book's authors. To be fair, the context was his refusing to invest in the buyout of RJR Nabisco on the grounds that he was rich enough that it wasn't worth the moral culpability to own a tobacco company.^[1] RJR itself was another monster long-term winner, though the time series gets interrupted because of the buyout described in Barbarians. One of the reasons that deal closed at the price it did was that the Nabisco part had so much growth potential. Nabisco shares just enough of the economic characteristics of the cigarette business that there was a plausible synergy between the two: consumable products, more brand loyalty than brand differentiation (Oreos are not that much better than the knockoffs, but part of what you're paying for is the certainty that they're exactly what you expected).

Consumer packaged goods companies in general have been friendly to investors for many decades. Like some consolidation stories in media, that's partly a function of survivorship bias: the period when national brands got big was tough for the regional ones, so there were more losers than winners overall. But the winners kept on winning: these businesses tend to have high margins and returns on equity relative to other manufacturers, and that's been going on for a long time. (In the wonderful The Crash and Its Aftermath—wonderful specifically for anyone who's read a work of financial history and wished there were more tables, because it has a lot of tables—shows food and dairy companies earning returns on equity in the teens, with General Foods and Standard Brands in the high 20s, during the Great Depression.)

There are a few reasons that processing agricultural products into boxed, bagged, and canned foods is a better-than-average business. The biggest companies turn into a term for the entire category (if a writer says someone ate a Hershey bar, they're saying they ate chocolate, with basically no other connotation—it's a completely neutral descriptor, which is another way of saying it's the default, can't-go-wrong expectation). National distribution (through the rise of railroads) and national marketing (radically improved by TV) synergized nicely: the company that could get the biggest returns to scale from manufacturing and distribution could also amortize its media spend over more unit volume. And that played nicely with their grocery store relationships: stores charge for eye-level shelf space, endcaps, and other attractive in-store real estate. And just as digital advertisers get better results when they factor in click-through rate, the store is looking not just as who pays the most for shelf space but what shelf placement gets the best combination of slotting fee revenue and contribution profit from units moved. And these things are cheap to move! A box of Kraft Mac 'n' Cheese requires a lot less care and handling than, say, a pear. They're also cheap to handle or the manufacturer. Kraft can fill an entire truck with pallets full of master cases of blue boxes of pasta-with-quasi-cheese; they're shipping a box of boxes of boxes of boxes, which gets predictably and continuously subdivided as it gets closer to customers.

But the grocer-CPG relationship is a strained one, because it's a bad customer experience—not to mention a big logistical problem—for some brand to disappear from store shelves. This happened a few years ago between Heinz and Tesco, and it's costly to both sides; Tesco will have customers who've had the same item on their shopping list for decades, and who will feel ripped off if they don't get it. But if they try something else and decide it's an acceptable substitute, or that they can't believe they've been settling for Heinz this whole time, Heinz loses an end customer. Fights over shelf space can lead to all sorts of exciting maneuvers: if Purina finds out that Pedigree is about to launch a specialty product that competes with one of theirs, they might decide that it's a great time to sell some bulk packages, so as many customers as possible are well-stocked with 50lb-bags of dog food while the campaign is running.

CPG companies rewarded investors for many years, especially as the quality factor got identified ad hoc by discretionary investors and then more formally by quants: if you own a high-margin business, it's likely to outlast downturns that hurt its competitors, and the decision to defer some short-term profits in exchange for longer-term gains is rarely existential. High-ROE companies can rely on their own profits to finance growth, and if a company benefits from scale, and that benefit sometimes involves strong-arming its competitors, it can sustain high incremental returns on equity by having more and more room to do that. Meanwhile, every year its brands get a little more established, so there's room to keep raising prices.

Modest unit growth, modest pricing growth, low capital requirements, and resistance to the economic cycle—that's the kind of business where you can sketch out a ten-year discounted cash flow analysis and be reasonably confident that it will actually represent reality.

Then again, if you did that ten years ago, you have some problems. The big brands have had some trouble, and while they're still high-margin companies, investors aren't extrapolating indefinite volume and pricing growth. Stores can take that captive shelf space and redeploy it to house brands, letting their suppliers choose between high volume at low margins or the prospect of no unit volume at all. As retailers have collected more data. And the value of information has gone up—there are just a lot more things you can do with data on who buys what, how loyal they are, how responsive they are to discounts, etc., when more advertising is digital and can be very precisely targeted. The CPG companies know they're part of a valuable bundle, but the grocery stores are in a better position to know just what value they add to that bundle. It hurts even more as eating continues to shift from home-prepared foods to restaurants including takeout—a restaurant is partly in the business of obliterating any chance for Sysco to take credit for what it makes possible, and to replace ingredient brand loyalty with restaurant loyalty—and within grocery, more shopping has moved to warehouse clubs and supercenters, whose entire pitch is that customers are paying less per unit and getting more. (In the case of Costco, the other part of the pitch is that if you're buying the Kirkland brand, you're getting more, for less, and it's actually slightly better than the name-brand product.)

GLP-1s provide even more excitement here, because they suggest an end to the long bull market in calories per capita. Calorie consumption can grow for a long time, because adding weight means raising your basal metabolic rate. So there's a nice feedback loop, where the food industry's constant efforts to make things more flavorful, give them a better mouthfeel, etc., also increases the total demand for snack foods, appetizers, and larger portions.

If people were perfectly rational, it would be just as easy to divvy up a growing stream of profits as a shrinking one. You can afford to be generous and prosocial when you tell yourself that you're just biding your time to really squeeze the other side, and there is nothing easier to overestimate than your own future abilities. When the sum to be divided is shrinking all the time, people naturally start thinking of it as a finite resource, and wondering who will get the last dollar and at whose expense.^[2] When the pie is shrinking, the knives come out.

This is the kind of story that happens quietly and gradually, though GLP-1s have accelerated it. When information is scarce, a brand name is valuable, a signal straight from manufacturer to customer that this is a product that delivers on its promise (a promise whose details are fleshed out in fifteen- and thirty-second spots). But while it's very nice that you and everyone you know know exactly what a Ritz cracker is, it's also probably the case that your local grocery chain, or Instacart, could pick a better one, and the biggest online ad platforms could tell you things about your true cracker-related preferences that would shock you. It's falling to the same force that older generations of AI did: something simple given a lot of data beats something complicated and clever with a little of it.

1. He did end up participating in the deal as an arbitrageur, but he'd also previously joked that that was morally dubious (“Because my mother isn't here tonight, I'll even confess to you that I've been an arbitrageur,") and because he was betting less on a tobacco business and more on the dynamics of a bidding war. ↩︎

2. In practice, if you can put it into a discounted cash flow model, you can make it finite, and if you can't do that, either your model is broken or your business model will break the economy. (It's usually the first one.) ↩︎

Diff Jobs

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

• YC-backed, ex-prop trader founder building the travel-agent for frequent-flyers that actually works 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. (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 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, this is for you. (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.
• 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

The Other Net Present Value

One thing that's sometimes missing from discounted cash flow models is the identity of the buyer. There are times when 100% of a company is worth more than 100 1% stakes because that company can influence outcomes for a much bigger one. Last week, The Diff noted that Nous Research, the maker of Hermes, is surrounded on all sides by companies that have leverage over it and can see it as a source of margin, but AI is growing so fast that it's still in talks to raise money at a valuation of $1.5bn. But a good illustration of why thin layers can be worth a lot is that OpenRouter is in talks to sell ($, The Information). It's a very strategic asset, both as a direct way to influence the distribution of specific models and as a way to monitor the behavior of relatively savvy users who are comfortable with complexity and spending enough on AI that they care to manage their budgets. In OpenRouter's case, labs want a direct route to users and users want to commoditize whatever cost-saving layer they find. But everyone's in a bigger hurry to grow revenue than to control costs—the stories about companies watching employees' inference tend to be about slowing a growth rate, not reversing a trend—and so there are many places to capture value even if everyone's theoretically after it, and then to preserve that value by giving one of the contenders a chance to control some dangerously strategic territory.

Capitulation

One of the hard parts of running an increasingly unpopular kind of business is that the longer you stubbornly hold on, the bigger a story it is if you quit. There used to be dozens of funds that focused exclusively on short-selling, but they've had headwings: the market getting more efficient, SarBox shrinking the opportunity set, and the inconvenience of explaining repeatedly to an investment committee that, no, down 2% year-to-date is actually beating the benchmark, since the market's up 9%. So another short-focused fund, Sakhet, is winding down so the founder can focus on buying Korean value stocks ($, WSJ). Short-focused funds have an annoying marketing problem, in that stocks go up and that by convention returns get quoted in absolute terms, rather than in terms of alpha versus a benchmark. It's just easier to market the same stream of returns by wrapping it inside a structure that is either market-neutral and sold as such, or is long-biased and pitched to investors who expect that. The opportunity set for a short seller fluctuates a lot; credit cycles can create feedback loops where basically every name in a given subsector is a short, and there are occasionally big companies like Wirecard with clearly-articulate short cases. But other times, there just aren't as many big opportunities, or if there are, they're closer to shorter-term changes in sentiment—betting that something will underperform a bit in the next few months, not that it's a zero some time in the next few years. Strategies that require variable amounts of capital will tend to find themselves tucked into platforms that are otherwise steadier but can treat their balance sheet flexibility as a competitive advantage.

Insurgent Advertising

One of the most famous ad campaigns in history is Avis' "We're only #2. We try harder," which somewhat passive-aggressively noted they're a little more grateful for your business than Hertz because they get less of it. The campaign worked, or at least coincided with other things that worked (they got taken over by Lazard, which kicked out the old management, killed some money-losing divisions, and expanded the fleet. But it's a pretty great ad campaign).

So it would be hard for a company that's also in the business of providing car rides to people who don't have a car handy might emulate it: Lyft is doing a new ad campaign telling people to save money and check Lyft, i.e. its default assumption is that you're going with the competitor. Perhaps some loyal Lyft users will see the ads and start checking Uber. But, overall, that trade is to Lyft's benefit while they're below 50% market share. The time to check one more app is a tiny fraction of the wait time, so it feels less costly, and the customer already has the phone out and unlocked. Meanwhile, drivers tend to multi-home more than customers do. So the #2 site will tend to have an oversupply of drivers relative to the #1. If you're #2, what you want—and, in this context, what you can try to engineer—is pushing the market towards being a coin toss.

Alignment

Hugging Face was recently hacked by an autonomous AI tool. The attack was also identified by one, an anomaly detection system that uses LLMs to identify and prioritize the investigation of anomalies in their logs. But, when they tried to use frontier models to analyze the attack, their requests got rejected, because of course those requests take the form of sharing a malicious payload and asking for help. In fairness to the model providers, this is part of a hacker's workflow, too: they want to know about countermeasures so they can counter them. Huggingface was eventually able to run GLM 5.2 instead. What this illustrates is that when the best models are more locked down, it's the best open weight models that ultimately determine which capabilities will be available; if someone can just use GLM-5.2 to do something, it's hard to argue that Fable or GPT-5.6 should refuse.

Math

Sometimes, you will happen to have some kind of unique-to-you information that you recognize is a big deal to others, and you'll have enough time to formulate the best way to present it to people. In some domains, this means putting a lot of effort into the superficial signals: you know people won't necessarily believe you, and they can't necessarily verify the results. But in the case of math, particularly parts of it where name recognition implies some mathematical skill, you can be as informal as you like. So: "hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final" is how we learned that a conjecture that had resisted efforts to prove it for 87 years turns out to be false.

It's deflating, in a way: someone came up with some clever formulation, many brilliant people can try to prove it, and a computer program (albeit one that also required some impressive cleverness) can come along and bluntly shoot it down. But! There are some cases in mathematical history that follow a pattern like this:

1. X always implies Y
2. Actually, X usually implies Y, but there are some counterexamples.
3. As it turns out, there is some deeper underlying pattern that yields a more comprehensive theory.

This is more common in less pure fields where we're working with simplified models of the real world rather than the real world itself (here's one from biology, and Mendeleev's periodic table broke the assumption that the sensible ordering of atoms was by weight). But it has happened in math: integers have prime factors, but algebraic integers don't necessarily have them, which was an observation that led to later advances in number theory that are mercifully beyond the scope of this newsletter, but that do seem pretty cool. If something is almost always true, and you find a counterexample, it means there's more thinking and research to do, not less. And, fortunately for us, the tools available for that kind of research keep getting better.