Back to Discover

Apple, Occasionally a Commodities Trading House

March 24, 2026
•The diff

Here's something that might surprise you — Apple has occasionally acted like a commodities trader, locking in long-term supply deals to control costs and limit competitors. Byrne Hobart points out that Apple’s scale lets it make strategic moves, like signing multi-year contracts for memory chips when prices are high. This isn’t just about saving a buck; it’s a power play — ensuring supply when others are scrambling, and even shaping the market. Now, here’s where it gets fascinating — Apple’s ability to wield this kind of influence isn’t typical for most companies. They’re big enough to be their suppliers’ biggest customer, giving them leverage to secure favorable terms and even move production to politically advantageous regions. Hobart notes that this kind of scale allows Apple to make decisions not solely for profit, but to shape competitive access to key resources. Think of it as a hybrid of tech giant and commodities trader, quietly shaping the supply chain to keep rivals at bay. That’s the hidden playbook of scale, and Apple’s mastery of it could redefine what we expect from tech companies in the future.

Screenshot-2022-02-17-at-12.51.50-1.png

In this issue:

  • Apple, Occasionally a Commodities Trading House—Apple has enough scale and predictability that, from time to time, it can make strategic commitments partly for its own use and partly to lock out competitors. These aren't a big part of the company's business model, but they're a nice addition from time to time.
  • How Inflation Spreads—When oil prices go up, you lose counterfactual excess capacity somewhere else.
  • Reaching Your Target Audience—Misinformed traders are a critical ingredient in well-informed prices.
  • Labor Substitution—Companies that don't compete primarily on price don't like to pass through cost savings.
  • Agents—Email jobs provide convenient training data.
  • Financing—There's more than one way for OpenAI to get a return from AI joint ventures.

Talk to this post on Read.Haus.

audio-thumbnail
The Diff March 23rd 2026
0:00
/792.267755
1×

Apple, Occasionally a Commodities Trading House

It's lucrative to sit at the intersection of commodities as abstract financial products and commodities as physical products that get extracted in certain places and processed elsewhere. Figuring out the right destination for every one of the hundred million-odd barrels of oil extracted every day is hard enough when everything is working normally, and both harder and more lucrative when Hormuz is impassable and refining and production are under attack. The companies that master this, like Vitol, Trafigura, Glencore, etc., produce occasionally-staggering profits, and the occasional settlement or guilty plea. (There is often a big gap between legal norms in the countries that import commodities and the ones that produce them. It's tricky to navigate.)

This is a business that can start very small, given the right opportunity, but it benefits from scale: if you connect with more endpoints, you collect more information, and if you're a repeat customer, you can probably take advantage of that information more. There's a cycle, where the biggest companies get risk-averse—either having zero tolerance for legal risk or having a carefully-calibrated sense of when they should and shouldn't ask prying questions. If the gap between what a big company wants to do and what a smaller one can get away with, someone will get busy turning legal gray areas into green ($, FT).

For this model to beat vertical integration or direct supplier-processor relationships, the product in question needs to be commoditized enough that there are multiple buyers and sellers, but not so commoditized that anyone can find a good source. It helps if production is high relative to storage costs, so losing one producer or being hit with a demand shock means scrambling for new supply.

So plenty of businesses don't fit this model. There just isn't a deal to be made, so it doesn't make sense to specialize in dealmaking. And someone who operates with enough scale to have the information advantage of a big trader also runs the risk of getting into antitrust trouble.

But it does happen occasionally, and Apple happens to be unusually good at it. They recently released the Macbook Neo, for as little as $600 compared to their entry-level Macbook Air at $1,100. Apple has been selling a premium-priced product for a long, long time; even the Apple I was more expensive than peer products (other than the SOL-20, but that was an all-in-one device, whereas the Apple I required the user to supply a keyboard, power supply, and display). As with many other industries, the lowest-priced provider tends to be the most sensitive to commodity price swings. Since memory prices have roughly tripled since the middle of last year; on their last earnings call, HP said "memory and storage costs made up roughly 15% to 18% of our PC bill of materials [last quarter], and we now currently estimate this to be roughly 35% for the year." Apple tends to sign longer-term contracts for inputs, and when supply is scarce and prices don't fully react, they're the customer their suppliers would least like to disappoint.

Apple has actually used this as a strategic tool in the past: in 2005, they signed long-term agreements for flash memory through 2010, citing the need to make as many iPods as the world wanted. Which did sound realistic: the first iPod had used a hard drive, but they'd released the Nano and Shuffle which used flash memory, and they'd double storage per device for each of the first few generations of the original iPod.

Apple had bigger plans than that: more iPods meant more people using iTunes to manage or perhaps even buy their music, and Apple was cooking up another pocket-sized device that would benefit from this mild network effect and probably add one of its own.

It's important to note that as a trade, this was actually a pretty bad one. When Apple made their initial commitment, they were making a long-term commitment in a tight market, where they and their competitors were worried about a persistent flash memory shortage. Flash memory prices were plummeting by early 2008. They did better a few years later, buying Authentec for its fingerprint reader technology and patents, then promptly shipping the iPhone 5s with a fingerprint reader as a key feature. This, too, was a move that makes sense in terms of network effects:

  1. A slight decrease in the friction of using the phone means marginally more photos, iMessage texts sent, mobile gaming sessions, etc., so it gave Apple a tiny bit more lock-in and a little extra services revenue. We know from other consumer tech companies that even barely perceptible decreases in latency can increase engagement substantially. It's also very hard to switch to a device when your first interaction with it is that it isn't as convenient as the alternative, so that improvement probably translated into lower churn, too.
  2. There are many use cases for which phones can compete with something else in your pocket: keys, boarding pass, flashlight. A faster unlock means that wallets can be added to the list.

Apple is also a big enough supplier that they can use another technique from the commodity-trading playbook: locking down supply in a favorable jurisdiction. When Apple committed to spending $2.5bn on US-manufactured glass from Corning last year, they were outsourcing the annoying work of operating a factory in a highly-regulated, high-wage economy like the US to Corning, while getting the political and PR upside.

For most companies that don’t have ~2,000 Global Supply Managers at their disposal, it's hard to craft customized terms like this, so they just don't have the scope to do big, commodity trade-like deals. The closest they get is vertical integration or some nebulously good relationship with their suppliers. Apple is unique in that it's big enough to be the most important customer to a large number of companies, it's good at actually producing steady enough sales that a multi-year commitment makes sense, and because Apple is a notoriously demanding buyer, it makes their suppliers look good to those suppliers' other customers.[1] This mix of traits exists on a continuum, and so in a sense Apple is not so much unique as it is the biggest and most visible company in this general category. But it's another way scale pays off well: a big enough company can, from time to time, make decisions not just based on getting what it needs to maximize profits, but to minimize competitors' access to the resources they'd use to compete.

  1. That's true in a tricky sense, and forces people to make a bet. One way to do a great job for Apple is, when you're overcommitted, to force everyone else to accept delivery delays. On the other hand, that's a martingale bet, and Apple probably doesn't want a supplier to lose all of its non-Apple customers—at that point, it's Apple's job to keep that company alive! The better a company gets at predicting its own output and scaling it up or down efficiently, the smaller a share of its economic bankroll it's committing to that martingale bet dynamic. ↩︎


Longreads

Books

Cool: How Air Conditioning Changed Everything: Microhistories around a single product or invention can be hit or miss; The Prize is great for reframing the twentieth century as a history of oil, Paper had enough general-history errors to make its paper-specific claims questionable. Cool is on the very good end of that spectrum: it's the story of efforts to control air temperature, mostly from the late 19th century through the widespread deployment of modern AC.

The author appears to have read basically every primary source that either mentioned attempts to cool the air or the consequences of not doing so. This leads to some evocative images: many important events in the history of statecraft and the arts took place in poorly-ventilated rooms with lots of people; the Globe theater and the House of Representatives were both incredibly unpleasant places to be when the weather got warm. Sometimes, needlessly so: the House of Lords once installed a cooling system based on very hazy notions of hot air rising, causing air to circulate, which they implemented by installing ovens under the floor. It must have seemed like a good idea at the time, but it makes you feel for the unfortunate Lords who were being rotisseried in the service of cooling their building.

The book also has a story that's sure to warm the hearts of anarcho-capitalists everywhere: for a while in the late 19th century, the way newspapers reported on weather was that they checked a giant thermometer in front of a soda shop downtown. Using intellectual property without paying for it and a business that provided a public service entirely so it could sell more soda? It's like something out of a David Friedman thought experiment.

There were two related reasons that climate control took a while to get going. First, people didn't seriously consider the possibility that they could artificially lower the temperature, even though they were used to raising it during winter. Or, rather, different people in different places had come up with ways to make heat more tolerable, but couldn't imagine that you could impose mild autumn weather at will. And second, the richest and most technologically advanced countries in the early days of AC were also countries that had strong social norms against complaining about discomfort. If you weren't miserably boiling in your suit all summer, where would you draw the line? (In fact, many of the early ads for air conditioning and its predecessor technologies frame them as a solution to a different Victorian buggaboo: they were tools for getting pure air, with the cooling only incidental.)

So adoption of good AC was halting, and weird: it started being applied to some specialized industry use cases, like controlling the temperature of beer during the brewing process, or keeping cadavers nice and cool in medical schools (Cornell Medical College used to hold its graduation ceremony in the same room where students dissected corpses). It was big in theaters, where it made a play called Hazel Kirke one of the most popular in America in the 1880s (the plot of that play was so thin that when it was turned into a movie twice, one of the movies decided to change which love interest the heroine chose). It expanded into movie theaters and spread rapidly through trains in the 1930s. These public-facing early adopters seem a little odd, but it all makes sense when you realize that these are businesses that charge for admission rather than charging for a product as customers are leaving. They're the ones who can measure whether or not customers value air conditioning by seeing if they'll pay for it.

The rest of the deployment story is a classic general-purpose technology story: it started out being bought by specialized industrial users and rich, decadent hobbyists, but every iteration got cheaper, especially relative to other purchases it was associated with—in the postwar period, an AC system could be 20% of the cost of a new home, but systems got smaller and cheaper.

The air conditioner is a surprisingly consequential invention. It enhances the convexity of expertise if you can practice your craft year-round; it globalizes labor markets if every big city has the same climate once you're indoors. Cities like Dubai and Singapore are inconceivable without AC, and Austin and Atlanta would be a lot smaller in an AC-free world. Many people who would have died during heat waves got a few more years. All this should be a reminder of how much unmeasured wealth we have today. History was a sweatier, smellier, less comfortable environment than the one in which we live. Lucky us!

Open Thread

  • Drop in any links or comments of interest to Diff readers.
  • What are some good histories of particular product categories like this? This kind of book is very helpful because you'll go into it already having some preconceptions about the time period in question, and may well get the answer to a mystery that's been slightly bugging you since you read some historical anecdote.

Diff Jobs

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

  • Series A startup that powers 2 of the 3 frontier labs’ coding agents with the highest quality SFT and RLVR data pipelines is looking for growth/ops folks to help customers improve the underlying intelligence and usefulness of their models by scaling data quality and quantity. If you read axRiv, but also love playing strategy games, this one is for you. (SF)
  • A series D, next-generation chemicals company that’s manufacturing the mission-critical inputs for a sustainable American reindustrialization and next-generation defense applications is looking for a VP of Finance to own the operating model, forecasting, and fundraising prep. Demonstrated interest or experience in real economy sectors (energy, industrials, chemicals, etc.) preferred. (Remote, Houston)
  • 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)
  • High-growth startup building dev tools that help highly technical organizations autonomously test and debug complex codebases is looking for forward deployed engineers who want  to dive into customers’ complex software systems, find pressing business needs and deploy a cutting edge platform to help thoroughly test mission-critical applications. Experience with fuzzing or property-based testing a plus! (SF, London, D.C.)
  • 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

How Inflation Spreads

Gasoline is about 2% of consumer expenditures, and spending on oil overall is about 3.3% of GDP. But its effect on prices and growth sometimes follows a strange path: United Airlines will temporarily cut about 5% of its current or planned capacity, at least through fall, and Qatar Airways is stashing some unused planes in Spain ($, FT). This is an example of why airlines don't hedge fuel costs as much as they used to: as long as they aren't too levered to ride out temporary fluctuations in their margins, and as long as competitors are economically rational, higher oil prices now mean lower capacity growth in the near future, since fewer routes hit the profitability hurdle rate. So that current price increase shows up in a slightly tighter supply of plane seats for months in the future. But that has knock-on effects, too: airline maintenance gets a little cheaper as fleets get bigger, because they need fewer backups and can better smooth out big maintenance projects. And growing the number of planes an airline flies means hiring new staff, who start out less senior and thus cheaper. If the whole Iran situation gets resolved in a few weeks, and crude drifts down to roughly where it was before, most people probably won't notice that their flights are a percent or two cheaper than they counterfactually would have been, but this process, multiplied across all of the supply chains that use oil, is how even temporary energy price spikes can create a persistent inflationary effect.

Reaching Your Target Audience

One of Polymarket's marketing strategies is to post breaking news stories, implicitly suggesting that readers bet on the outcome. These stories go fairly viral, often because, as it turns out, Polymarket is either fibbing or omitting key details. The usual paradox of prediction markets, and the reason professional investors were so reluctant to use them, is that the more prices fulfill their social function of telling people what the informed probability of an outcome is, the fewer people there are to trade with. You basically need an endless supply of degenerate gamblers, the way steam ships needed a sweaty crew in the basement constantly shoveling coal or they'd be dead in the water. In that sense, Polymarket is doing a great public service: by ensuring a steady supply of gamblers with completely uninformed or deliberately misinformed opinions about Iran, the Oscars, future AI model releases, etc., they make it profitable for informed people to trade against them. But it also means that a prediction market has two steady states:

  1. They get enough volume that people use them for hedging, not just gambling, and those hedgers provide uninformed flow that it's profitable to trade against, or
  2. The prediction markets all compete to find uninformed gamblers and onboard them onto their platforms, and it's a race between rising CAC and legislation to see which one kills the model first.

Note that normalizing prediction markets as something risk-avoidant people use as a business planning tool leads to exactly the opposite incentive as cultivating bad traders. Right now, at least some of the utility from prediction markets comes from what you might call meta-uninformed traders, who do know useful facts about the world, perhaps because they got invited to the right conference room or got a good table at Mar-a-Lago, but who don't monetize this information in much more liquid financial markets. Those traders do provide some useful information, but they only exist because norms against leaking information via prediction markets are weak right now. Prediction markets are still an incredibly useful technology, but maybe in the category of opiates and amphetamines: helpful for some, too easy to abuse for others, and inevitably a target of regulation that's going to snap back and make up for its prior inactivity.

Labor Substitution

The Economist has an overview of how Zara has held its own against fast-fashion competitors, online and off ($). Retail is a generally tough business, especially because companies are trying to maintain a consistent brand even though they only control the product experience for as long as customers are in the store (or requesting a refund later). But one tidbit stood out: "His stores have introduced self-service checkouts to free staff to act as personal shoppers." Business models are nominally sticky; when some exogenous force causes prices to drop, companies that compete on price will respond immediately, but companies that compete on quality will find some way to reinvest the proceeds in quality so they don't have to lower their prices. Since most workers whose jobs are threatened by AI aren't working for companies that compete primarily on price, the average experience will look something like that: if output per worker goes up 10x, renewing a contract at the same price with 10x the deliverables is a lot more exciting than renewing last year's contract with terms unchanged except a 90% price cut.

Agents

Meta is experimenting with internal AI tools, including an AI chief-of-staff for Mark Zuckerberg ($, WSJ). When you look at which jobs are most automatable through LLMs, what you're really asking is which jobs are best-documented. An inbox and outbox are pretty good inputs for training a model on what output to produce from a given input, which is one reason The Diff has argued that short-term discretionary traders are an unusually easy to automate job ($), with lots of nicely-paired inputs and outputs. For a CEO, some of the work they do won't show up in plain text in their email, particularly if they're trying to avoid bad PR or regulatory consequences. Which means that the work that's easiest to automate is also the kind with the lowest stakes. The article mentions that one of the use cases for this bot is skipping levels in the org chart when asking for information, which is a scenario where it saves time and the answer is usually easy to validate, and also where it can't cause too much trouble.

Disclosure: long META.

Financing

Both Anthropic ($, The Information) and OpenAI have been in talks with private equity firms to create joint ventures. PE is a natural early adopter for AI tools, because PE firms want to have a consistent playbook they apply to different acquisitions, and a more flexible set of tools can be applied in more places. One strange new detail here is that OpenAI is apparently guaranteeing a 17.5% return for AI investors in their JV. There are some interest rates for which the market doesn't clear—anyone who accepts a sufficiently bad deal implies that they're even more desperate for cash than they appear to be. But some of this comes down to structure: if OpenAI is guaranteeing a return from some investment vehicle, and that vehicle also gets them more distribution and revenue elsewhere, then it's just aother case where they're subsidizing adoption of their products with the expectation that they'll monetize it later.

Audio Transcript

Screenshot-2022-02-17-at-12.51.50-1.png

In this issue:

  • Apple, Occasionally a Commodities Trading House—Apple has enough scale and predictability that, from time to time, it can make strategic commitments partly for its own use and partly to lock out competitors. These aren't a big part of the company's business model, but they're a nice addition from time to time.
  • How Inflation Spreads—When oil prices go up, you lose counterfactual excess capacity somewhere else.
  • Reaching Your Target Audience—Misinformed traders are a critical ingredient in well-informed prices.
  • Labor Substitution—Companies that don't compete primarily on price don't like to pass through cost savings.
  • Agents—Email jobs provide convenient training data.
  • Financing—There's more than one way for OpenAI to get a return from AI joint ventures.

Talk to this post on Read.Haus.

audio-thumbnail
The Diff March 23rd 2026
0:00
/792.267755
1×

Apple, Occasionally a Commodities Trading House

It's lucrative to sit at the intersection of commodities as abstract financial products and commodities as physical products that get extracted in certain places and processed elsewhere. Figuring out the right destination for every one of the hundred million-odd barrels of oil extracted every day is hard enough when everything is working normally, and both harder and more lucrative when Hormuz is impassable and refining and production are under attack. The companies that master this, like Vitol, Trafigura, Glencore, etc., produce occasionally-staggering profits, and the occasional settlement or guilty plea. (There is often a big gap between legal norms in the countries that import commodities and the ones that produce them. It's tricky to navigate.)

This is a business that can start very small, given the right opportunity, but it benefits from scale: if you connect with more endpoints, you collect more information, and if you're a repeat customer, you can probably take advantage of that information more. There's a cycle, where the biggest companies get risk-averse—either having zero tolerance for legal risk or having a carefully-calibrated sense of when they should and shouldn't ask prying questions. If the gap between what a big company wants to do and what a smaller one can get away with, someone will get busy turning legal gray areas into green ($, FT).

For this model to beat vertical integration or direct supplier-processor relationships, the product in question needs to be commoditized enough that there are multiple buyers and sellers, but not so commoditized that anyone can find a good source. It helps if production is high relative to storage costs, so losing one producer or being hit with a demand shock means scrambling for new supply.

So plenty of businesses don't fit this model. There just isn't a deal to be made, so it doesn't make sense to specialize in dealmaking. And someone who operates with enough scale to have the information advantage of a big trader also runs the risk of getting into antitrust trouble.

But it does happen occasionally, and Apple happens to be unusually good at it. They recently released the Macbook Neo, for as little as $600 compared to their entry-level Macbook Air at $1,100. Apple has been selling a premium-priced product for a long, long time; even the Apple I was more expensive than peer products (other than the SOL-20, but that was an all-in-one device, whereas the Apple I required the user to supply a keyboard, power supply, and display). As with many other industries, the lowest-priced provider tends to be the most sensitive to commodity price swings. Since memory prices have roughly tripled since the middle of last year; on their last earnings call, HP said "memory and storage costs made up roughly 15% to 18% of our PC bill of materials [last quarter], and we now currently estimate this to be roughly 35% for the year." Apple tends to sign longer-term contracts for inputs, and when supply is scarce and prices don't fully react, they're the customer their suppliers would least like to disappoint.

Apple has actually used this as a strategic tool in the past: in 2005, they signed long-term agreements for flash memory through 2010, citing the need to make as many iPods as the world wanted. Which did sound realistic: the first iPod had used a hard drive, but they'd released the Nano and Shuffle which used flash memory, and they'd double storage per device for each of the first few generations of the original iPod.

Apple had bigger plans than that: more iPods meant more people using iTunes to manage or perhaps even buy their music, and Apple was cooking up another pocket-sized device that would benefit from this mild network effect and probably add one of its own.

It's important to note that as a trade, this was actually a pretty bad one. When Apple made their initial commitment, they were making a long-term commitment in a tight market, where they and their competitors were worried about a persistent flash memory shortage. Flash memory prices were plummeting by early 2008. They did better a few years later, buying Authentec for its fingerprint reader technology and patents, then promptly shipping the iPhone 5s with a fingerprint reader as a key feature. This, too, was a move that makes sense in terms of network effects:

  1. A slight decrease in the friction of using the phone means marginally more photos, iMessage texts sent, mobile gaming sessions, etc., so it gave Apple a tiny bit more lock-in and a little extra services revenue. We know from other consumer tech companies that even barely perceptible decreases in latency can increase engagement substantially. It's also very hard to switch to a device when your first interaction with it is that it isn't as convenient as the alternative, so that improvement probably translated into lower churn, too.
  2. There are many use cases for which phones can compete with something else in your pocket: keys, boarding pass, flashlight. A faster unlock means that wallets can be added to the list.

Apple is also a big enough supplier that they can use another technique from the commodity-trading playbook: locking down supply in a favorable jurisdiction. When Apple committed to spending $2.5bn on US-manufactured glass from Corning last year, they were outsourcing the annoying work of operating a factory in a highly-regulated, high-wage economy like the US to Corning, while getting the political and PR upside.

For most companies that don’t have ~2,000 Global Supply Managers at their disposal, it's hard to craft customized terms like this, so they just don't have the scope to do big, commodity trade-like deals. The closest they get is vertical integration or some nebulously good relationship with their suppliers. Apple is unique in that it's big enough to be the most important customer to a large number of companies, it's good at actually producing steady enough sales that a multi-year commitment makes sense, and because Apple is a notoriously demanding buyer, it makes their suppliers look good to those suppliers' other customers.[1] This mix of traits exists on a continuum, and so in a sense Apple is not so much unique as it is the biggest and most visible company in this general category. But it's another way scale pays off well: a big enough company can, from time to time, make decisions not just based on getting what it needs to maximize profits, but to minimize competitors' access to the resources they'd use to compete.

  1. That's true in a tricky sense, and forces people to make a bet. One way to do a great job for Apple is, when you're overcommitted, to force everyone else to accept delivery delays. On the other hand, that's a martingale bet, and Apple probably doesn't want a supplier to lose all of its non-Apple customers—at that point, it's Apple's job to keep that company alive! The better a company gets at predicting its own output and scaling it up or down efficiently, the smaller a share of its economic bankroll it's committing to that martingale bet dynamic. ↩︎


Longreads

Books

Cool: How Air Conditioning Changed Everything: Microhistories around a single product or invention can be hit or miss; The Prize is great for reframing the twentieth century as a history of oil, Paper had enough general-history errors to make its paper-specific claims questionable. Cool is on the very good end of that spectrum: it's the story of efforts to control air temperature, mostly from the late 19th century through the widespread deployment of modern AC.

The author appears to have read basically every primary source that either mentioned attempts to cool the air or the consequences of not doing so. This leads to some evocative images: many important events in the history of statecraft and the arts took place in poorly-ventilated rooms with lots of people; the Globe theater and the House of Representatives were both incredibly unpleasant places to be when the weather got warm. Sometimes, needlessly so: the House of Lords once installed a cooling system based on very hazy notions of hot air rising, causing air to circulate, which they implemented by installing ovens under the floor. It must have seemed like a good idea at the time, but it makes you feel for the unfortunate Lords who were being rotisseried in the service of cooling their building.

The book also has a story that's sure to warm the hearts of anarcho-capitalists everywhere: for a while in the late 19th century, the way newspapers reported on weather was that they checked a giant thermometer in front of a soda shop downtown. Using intellectual property without paying for it and a business that provided a public service entirely so it could sell more soda? It's like something out of a David Friedman thought experiment.

There were two related reasons that climate control took a while to get going. First, people didn't seriously consider the possibility that they could artificially lower the temperature, even though they were used to raising it during winter. Or, rather, different people in different places had come up with ways to make heat more tolerable, but couldn't imagine that you could impose mild autumn weather at will. And second, the richest and most technologically advanced countries in the early days of AC were also countries that had strong social norms against complaining about discomfort. If you weren't miserably boiling in your suit all summer, where would you draw the line? (In fact, many of the early ads for air conditioning and its predecessor technologies frame them as a solution to a different Victorian buggaboo: they were tools for getting pure air, with the cooling only incidental.)

So adoption of good AC was halting, and weird: it started being applied to some specialized industry use cases, like controlling the temperature of beer during the brewing process, or keeping cadavers nice and cool in medical schools (Cornell Medical College used to hold its graduation ceremony in the same room where students dissected corpses). It was big in theaters, where it made a play called Hazel Kirke one of the most popular in America in the 1880s (the plot of that play was so thin that when it was turned into a movie twice, one of the movies decided to change which love interest the heroine chose). It expanded into movie theaters and spread rapidly through trains in the 1930s. These public-facing early adopters seem a little odd, but it all makes sense when you realize that these are businesses that charge for admission rather than charging for a product as customers are leaving. They're the ones who can measure whether or not customers value air conditioning by seeing if they'll pay for it.

The rest of the deployment story is a classic general-purpose technology story: it started out being bought by specialized industrial users and rich, decadent hobbyists, but every iteration got cheaper, especially relative to other purchases it was associated with—in the postwar period, an AC system could be 20% of the cost of a new home, but systems got smaller and cheaper.

The air conditioner is a surprisingly consequential invention. It enhances the convexity of expertise if you can practice your craft year-round; it globalizes labor markets if every big city has the same climate once you're indoors. Cities like Dubai and Singapore are inconceivable without AC, and Austin and Atlanta would be a lot smaller in an AC-free world. Many people who would have died during heat waves got a few more years. All this should be a reminder of how much unmeasured wealth we have today. History was a sweatier, smellier, less comfortable environment than the one in which we live. Lucky us!

Open Thread

  • Drop in any links or comments of interest to Diff readers.
  • What are some good histories of particular product categories like this? This kind of book is very helpful because you'll go into it already having some preconceptions about the time period in question, and may well get the answer to a mystery that's been slightly bugging you since you read some historical anecdote.

Diff Jobs

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

  • Series A startup that powers 2 of the 3 frontier labs’ coding agents with the highest quality SFT and RLVR data pipelines is looking for growth/ops folks to help customers improve the underlying intelligence and usefulness of their models by scaling data quality and quantity. If you read axRiv, but also love playing strategy games, this one is for you. (SF)
  • A series D, next-generation chemicals company that’s manufacturing the mission-critical inputs for a sustainable American reindustrialization and next-generation defense applications is looking for a VP of Finance to own the operating model, forecasting, and fundraising prep. Demonstrated interest or experience in real economy sectors (energy, industrials, chemicals, etc.) preferred. (Remote, Houston)
  • 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)
  • High-growth startup building dev tools that help highly technical organizations autonomously test and debug complex codebases is looking for forward deployed engineers who want  to dive into customers’ complex software systems, find pressing business needs and deploy a cutting edge platform to help thoroughly test mission-critical applications. Experience with fuzzing or property-based testing a plus! (SF, London, D.C.)
  • 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

How Inflation Spreads

Gasoline is about 2% of consumer expenditures, and spending on oil overall is about 3.3% of GDP. But its effect on prices and growth sometimes follows a strange path: United Airlines will temporarily cut about 5% of its current or planned capacity, at least through fall, and Qatar Airways is stashing some unused planes in Spain ($, FT). This is an example of why airlines don't hedge fuel costs as much as they used to: as long as they aren't too levered to ride out temporary fluctuations in their margins, and as long as competitors are economically rational, higher oil prices now mean lower capacity growth in the near future, since fewer routes hit the profitability hurdle rate. So that current price increase shows up in a slightly tighter supply of plane seats for months in the future. But that has knock-on effects, too: airline maintenance gets a little cheaper as fleets get bigger, because they need fewer backups and can better smooth out big maintenance projects. And growing the number of planes an airline flies means hiring new staff, who start out less senior and thus cheaper. If the whole Iran situation gets resolved in a few weeks, and crude drifts down to roughly where it was before, most people probably won't notice that their flights are a percent or two cheaper than they counterfactually would have been, but this process, multiplied across all of the supply chains that use oil, is how even temporary energy price spikes can create a persistent inflationary effect.

Reaching Your Target Audience

One of Polymarket's marketing strategies is to post breaking news stories, implicitly suggesting that readers bet on the outcome. These stories go fairly viral, often because, as it turns out, Polymarket is either fibbing or omitting key details. The usual paradox of prediction markets, and the reason professional investors were so reluctant to use them, is that the more prices fulfill their social function of telling people what the informed probability of an outcome is, the fewer people there are to trade with. You basically need an endless supply of degenerate gamblers, the way steam ships needed a sweaty crew in the basement constantly shoveling coal or they'd be dead in the water. In that sense, Polymarket is doing a great public service: by ensuring a steady supply of gamblers with completely uninformed or deliberately misinformed opinions about Iran, the Oscars, future AI model releases, etc., they make it profitable for informed people to trade against them. But it also means that a prediction market has two steady states:

  1. They get enough volume that people use them for hedging, not just gambling, and those hedgers provide uninformed flow that it's profitable to trade against, or
  2. The prediction markets all compete to find uninformed gamblers and onboard them onto their platforms, and it's a race between rising CAC and legislation to see which one kills the model first.

Note that normalizing prediction markets as something risk-avoidant people use as a business planning tool leads to exactly the opposite incentive as cultivating bad traders. Right now, at least some of the utility from prediction markets comes from what you might call meta-uninformed traders, who do know useful facts about the world, perhaps because they got invited to the right conference room or got a good table at Mar-a-Lago, but who don't monetize this information in much more liquid financial markets. Those traders do provide some useful information, but they only exist because norms against leaking information via prediction markets are weak right now. Prediction markets are still an incredibly useful technology, but maybe in the category of opiates and amphetamines: helpful for some, too easy to abuse for others, and inevitably a target of regulation that's going to snap back and make up for its prior inactivity.

Labor Substitution

The Economist has an overview of how Zara has held its own against fast-fashion competitors, online and off ($). Retail is a generally tough business, especially because companies are trying to maintain a consistent brand even though they only control the product experience for as long as customers are in the store (or requesting a refund later). But one tidbit stood out: "His stores have introduced self-service checkouts to free staff to act as personal shoppers." Business models are nominally sticky; when some exogenous force causes prices to drop, companies that compete on price will respond immediately, but companies that compete on quality will find some way to reinvest the proceeds in quality so they don't have to lower their prices. Since most workers whose jobs are threatened by AI aren't working for companies that compete primarily on price, the average experience will look something like that: if output per worker goes up 10x, renewing a contract at the same price with 10x the deliverables is a lot more exciting than renewing last year's contract with terms unchanged except a 90% price cut.

Agents

Meta is experimenting with internal AI tools, including an AI chief-of-staff for Mark Zuckerberg ($, WSJ). When you look at which jobs are most automatable through LLMs, what you're really asking is which jobs are best-documented. An inbox and outbox are pretty good inputs for training a model on what output to produce from a given input, which is one reason The Diff has argued that short-term discretionary traders are an unusually easy to automate job ($), with lots of nicely-paired inputs and outputs. For a CEO, some of the work they do won't show up in plain text in their email, particularly if they're trying to avoid bad PR or regulatory consequences. Which means that the work that's easiest to automate is also the kind with the lowest stakes. The article mentions that one of the use cases for this bot is skipping levels in the org chart when asking for information, which is a scenario where it saves time and the answer is usually easy to validate, and also where it can't cause too much trouble.

Disclosure: long META.

Financing

Both Anthropic ($, The Information) and OpenAI have been in talks with private equity firms to create joint ventures. PE is a natural early adopter for AI tools, because PE firms want to have a consistent playbook they apply to different acquisitions, and a more flexible set of tools can be applied in more places. One strange new detail here is that OpenAI is apparently guaranteeing a 17.5% return for AI investors in their JV. There are some interest rates for which the market doesn't clear—anyone who accepts a sufficiently bad deal implies that they're even more desperate for cash than they appear to be. But some of this comes down to structure: if OpenAI is guaranteeing a return from some investment vehicle, and that vehicle also gets them more distribution and revenue elsewhere, then it's just aother case where they're subsidizing adoption of their products with the expectation that they'll monetize it later.

View original article
0:00/0:00
Apple, Occasionally a Commodities Trading House | Speasy