In this issue:

• When Prices Peak Before Fundamentals—Unfortunately, improving fundamentals don’t mean that the market will rise. In fact, we’re currently in the kind of cycle where asset price declines will cause worse fundamentals, but only on a lag.
• Negotiations—YouTube buys negotiating credibility by burning cash.
• Top-Down AI Deployment—Does “use it or else” work? Maybe. But it implies some serious problems.
• The Nostalgia Economy—Algorithmic feeds make brands built in a less fragmented media era more valuable. That leads to endless sequels and remakes—which are now coming for ads.
• Full-Stack AI—OpenAI looks for another lucrative consumer touchpoint.
• Tariff Dividends—Add up the impact and US trade policy increasingly seems to subsidize the services sector.

The Diff November 10th 2025

0:00

/899.73551

When Prices Peak Before Fundamentals

In one sense, it’s meaningless to talk about how bullish and bearish people are in the aggregate. 6,728.8 is the exact level of the S&P 500 at which, as of 4pm, bulls and bears agreed that stocks were fairly priced. But narratives bounce around right along with assets, and any time the market sells off after a big bullish episode, as it has so far this month, people start wondering if the party’s over, and if current prices will in retrospect be not so much “5% off the highs” as “50% above what turns out to be the low.”

One of the best bullish arguments is that equity fundamentals actually look pretty good, especially since some of the companies whose losses would be dragging down the S&P’s average profitability are still private. It’s not so much “this time is different” as it is “this time is like every other time”: take a snapshot of the US economy at any point in history and the most likely situation you’ll see is that the economy is growing, profits are growing a little bit faster, stocks are high enough to make people nervous, but that growth bails them out on average. Growth that’s priced at a premium means that you pay more for today’s earnings, but it also means you’re eventually paying a discount for future earnings. So, why sell when the outlook looks good?

What actually happens, though, is that markets tend to peak before the bad economic news really hits, and often rally earlier than recessions end, too. The details are different each cycle, but George Soros’ theory of reflexivity helps explain what’s going on. That theory, that asset prices lead fundamentals, sounds almost mystical when stated at a high level. And Soros’ specific example of this dynamic relies on a credit cycle (higher equity valuations make lenders more confident, so they lend at lower rates, which allows borrowers to expand faster and earn a bigger spread between what they borrow and what they buy, while also sometimes buying from previous borrowers and thus validating the last round of purchases).

But there’s also a macroeconomic way to apply reflexivity. The ratio of market value to replacement cost for assets is a rough measure of whether the market thinks adding more inputs to the economy will increase returns. Those inputs have different elasticities; you can’t add land, it takes a long time to add labor (in the aggregate, though immigration implicitly means pulling labor out of low-ROI economies and putting it into higher-ROI ones). But when those forces are all running at full speed, what’s left is capital. High equity valuations are a signal to every company that they need to invest more, and when they do that, the companies that make whatever they’re investing in will tend to see higher margins.

And companies are all linked together. When the market tells businesses that a dollar of cash is worth $1 of market cap, but $1 worth of GPUs or gas turbines is worth a multiple of that, capital allocators will do the prudent thing and expand.[1] This happens in other sectors, too: in 2007, the market told banks that $1 of equity was worth $3-5 of market cap, and thus that they ought to expand as fast as possible. For about a decade after the crisis, the market was telling European banks that they way to turn €1 into €0.50 was to retain it in a bank instead of returning it to shareholders.[2]

So, if equity prices rise, and companies increase their capital expenditures, what happens next? In general, what you’d see is that this shift will propagate its way through the supply chain. In some cases, this is pretty seamless: if the big growth bottleneck is unskilled labor, and your business can justify paying slightly more than competitors for a similar-quality job, then it’s straightforward for that labor to shift—Uber and DoorDash aren’t going to run out of drivers, though both companies do have to spend a lot of effort figuring out how to attract, best utilize and retain them. It’s a constraint, but not the binding one. If the core demand is for something with low or zero marginal cost—most software and media, for example—then the relevant bottleneck is even weaker.

But that propagation process is just a way for economies to, in a smooth and econ 101-compatible way, slice that demand as efficiently as possible into whatever real resource barriers limit it. Right now, those barriers include foundry capacity, power generation, and HVAC technicians, and if you roll that forward, the real barrier is convincing ASML, GE Vernova, Siemens Energy, Comfort Systems etc. that the current boom has legs and that they shouldn’t be afraid to increase capacity (either through capex or, in Comfort Systems’ case, acquisitions of operating companies) to take advantage of it.

Those companies are all big, cyclical businesses run by prudent capital allocators who, if they haven’t seen this exact movie before, have at least seen enough to know the signs. They don’t want to put themselves in a position where their splashy capital expenditure decision turns out to be the event people gesture to when they talk about how out of hand the bubble got (“Remember when we suckered that gas turbine executive into thinking that there was infinite demand for AI-generated videos of Sam Altman and Dario Amodei fighting a wizard duel? Good times!”).

If there’s enough demand, they’ll eventually expand, but in the meantime, they have no choice but to, with great reluctance, run their existing operations at 100% capacity, raise prices, and force their customers to offer more cash upfront and to commit to buying further and further in the future. The whole reason these cyclical companies don’t just stand up another factory or two is that those have fixed costs, and they need a certain amount of business to recoup those costs, and the flip side of that is that they’re very profitable indeed when there’s plenty of revenue to go around.

If they do finally decide to grow, the demand-propagation effect continues once again, and we’ll find out what the constraints-behind-the-constraints are. As this happens, though, more aggregate economic activity consists of revenue for cyclical businesses that are operating at peak-of-the-cycle margins. In a situation like that, how could fundamentals not look great ($, WSJ)?

And if what happens instead is what the cyclical companies worry about—investor enthusiasm will diminish, and instead of asking for new deliveries in 2030, turbine buyers start asking if the orders they made for 2029 might get pushed back a year or two, then there will be a period where the last bout of prepaid, hard-to-cancel orders are still being delivered, and revenue is still getting booked. But by that time, asset price deflation will have moved from buyers of bottlenecked inputs to sellers (if you ever screen for stocks at very low P/E ratios, with very high recent growth, and you exclude anything noisy like comping against a uniquely bad year or growing because of a one-time gain, what you’re generally left with are the cyclicals. Much money has been lost buying homebuilders, copper miners, and airlines at 5x trailing earnings).

These sentiment shifts take time, in both directions. Sometimes, it’s funny to toggle back and forth between people who spend all their time thinking about tech stocks and people who actually work in tech. When there’s a drawdown, it feels like the party might be over, and we should all rotate towards more defensive names or maybe sit on cash. This might be the big one! But if you talk to someone who’s making an expensive decision at a big company—do we hire this person? Sign this contract? Go through with this acquisition?—it would be pretty surprising for them to say “Tech stock prices have plunged, implying that our economy has literally regressed to where it was back in the last week of October. In this kind of environment, there’s no way I can justify going ahead with this deal.” And that works in the other direction, too; tech stocks were rallying by early 2023, mostly concentrated in the AI plays (and also SaaS like Salesforce, Shopify etc.), but it took a while for private markets to decide that We Are So Back.

The lead/lag dynamic is different for different drivers of the economic cycle. In 2007, earnings growth went negative by Q3, and markets peaked in the middle of the next quarter. Impairments in long-term earning power tend to hit financial institutions’ balance sheets faster than those of operating companies: if a restaurant chain thought that their next location would get a 15% return on equity, but then demand drops and they figure 10% is more likely, they aren’t going to write down that asset. Whereas if a lender is no longer confident that they’ll collect 100 cents on the dollar, they do need to respond. (If nothing else, their investors and lenders will assume the worst, and a writedown is a good way to formally acknowledge that you’ve lost money while giving investors less motivation to speculate about exactly how much.)

All of this leads to a kind of funny outcome: if earnings per share deteriorate before the market peaks, it’s because of a slowdown in the more levered non-AI parts of the economy. Whereas if investors lose their appetite for funding private labs and the CEOs of hyperscalers decide they aren’t so desperate to win, there will be a brief period where reported numbers tell a great story about AI—just a story investors have stopped believing.

---

This is part of the social function of publicly-traded pure-plays like CoreWeave. It’s actually a bit analogous to the crypto wrapper business, albeit with more of a real business justification: CRWV gives the market a dial that they can turn up or down based on their view of what aggregate GPU capex looks like relative to demand. ↩︎ 1.

Bank accounting is sometimes quite figurative, given that they don’t have to take immediate writedowns from long-duration assets that lost value because interest rates rose. But this captures an economic reality, which is that many of a bank’s assets automatically liquidate at precisely book value at some point in the future. Their portfolio is often a mix of short-maturity assets with credit risk—the reason credit cards are free liquidity plus cash-back to people who pay them off regularly and expensive debt to people who don’t is that those expensive borrowers have very cyclical creditworthiness. ↩︎

Diff Jobs

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

• Series-A defense tech company that’s redefining logistics superiority with AI is looking for a MLE to build and deploy models that eliminate weeks of Excel work for the Special Forces. If you want to turn complex logistics systems into parametric models, fit them using Bayesian inference, and optimize logistics decision-making with gradient descent, this is for you. Python, PyTorch/TensorFlow, MLOps (Kubernetes, MLflow), and cloud infrastructure experience preferred. (Salt Lake City or NYC)
• 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.

Elsewhere

Negotiations

One of the most important features of the media business is that the same piece of content can function as a monetizable product, but also as an ad—trailers, free trials, excerpts, tolerating piracy—this shows up everywhere. And one of the trickiest places it happens is in carriage disputes, when cable systems, and more recently streaming services, negotiate with networks. A business like Disney has some channels that are differentiated assets, like ESPN, and some that are just replacement-level content. Sometimes the main force at work is that they’re using the good channels to acquire new customers/get more distribution for the worse ones, and sometimes it’s more because an ESPN-free television offering isn’t competitive (subscriptions happen at the household level, so it’s not a question of the fraction of American audiences that watch, but the fraction of American households with at least one member who watches).

Somehow, there usually ends up being an equilibrium, where cable companies or YouTube grumble that they’re paying so much, Disney gets to say that whatever they’re charging is a discount given the priceless nature of the content, and viewers can watch what they expect to. When both sides can’t reach an agreement, they drop coverage, and in this case, YouTube is offering customers a $20 credit as compensation. If ESPN usually charges about $15, YouTube is on average overpaying its customers to mollify them. But it’s also signaling that there’s a point at which they’re willing to walk. A little over a decade ago, Disney kicked off a cable network selloff because they’d finally reached the point where they couldn’t just raise prices indefinitely. Year to year, it’s still worth it for media distributors to pay for access to the content customers have the strongest affinity for. But in the very long term, it’s also in their interest to credibly signal that they can always say no to a lopsided deal.

Disclosure: long GOOGL.

Top-Down AI Deployment

One of the running gags in late-90s Dilbert was that clueless bosses would read about random technology trends and demand that their underlings figure out what that technology was and find some way to make it relevant to their business. (They weren’t the only ones; decade-too-early ideas like a B2B exchange for lab chemicals were briefly worth billions of dollars in that era.) Now, the WSJ has a piece on executives telling their employees to figure out what they can do with AI, and then start doing it, or risk getting fired ($). In a way, these managers are displaying some admirable humility: they’re not AI experts, and have to defer to those experts when deciding how big a deal AI is; they also know less than their employees about the specifics of each job those employees do. On the other hand, you’d think that an important stage of AI adoption is learning to ask a model questions like “I run a company doing X. What are some reasonable ways I could use AI for my business?” Many executives are outsourcing this work entirely to FDE armies ($, FT) at the big labs and companies like Palantir, Brain Co, Distyl, and the like.

The Nostalgia Economy

One of the important consequences of non-generative AI is that algorithmic feeds have gotten good, and that’s fragmented culture. We just won’t have as many universally famous people as we did historically, even if there will be more people out there with ten million plus rabid fans. (You will never have heard of 99% of these people, and for the other 1% you’ll be incredulous that whoever you’re talking to hadn’t heard of them.) This applies to brands, too; advertisers are starting to remake classic weird ads from the 90s. Since these ads were made when there were fewer media options, they have broader name recognition. But they’ll also have a narrow demographic appeal—if you were too young to directly experience 90s pop culture, you might have caught up on it later, but you wouldn’t have seen the ads. So these ads, too, will benefit from hypertargeting: once your feed realizes which specific years you were waking up early on Saturday mornings and adding various jingles and taglines to your permanent stock of pop culture references, they’ll show you the ads that reinforce them.

Full-Stack AI

OpenAI is looking at health-related products. There are several angles they’re probably considering here:

1. Health throws off lots of unique data, and the earlier they can collect it the sooner they can train better predictive models.
2. It’s also one more way to get consumers using multiple OpenAI products. Cross-selling predicts lower churn, though the causal arrow can be hard to find. But one of the easiest ways this works is if it’s a bundle—either getting some health app for free or subscribing to a paid one for a discount will keep more people using the core app.
3. OpenAI will constantly have to deal with questions about user privacy, and one of the best ways to mitigate those is to be able to accurately say that if they didn’t collect so much data, some of their customers would have literally died. (It doesn’t hurt that employees will also be motivated by the possibility that the work they do will help treat chronic health conditions rather than sniping viewer-minutes from TikTok.)

Tariff Dividends

The Diff approach to policy is to focus less on who gets what and who pays for it, and more on how different policies shift longer-term incentives. So, what does an income-limited tariff dividend actually do? It neutralizes some of the tariff’s negative impact on people whose consumption skews to imports (these will be lower-income consumers, in part because higher-income people consume more services and housing, and in part because low-income consumers are younger; as they age, more of their consumption will shift to education and then healthcare). But it still means that importers are worse-off, because not all of that dividend gets spent on their products. So the net effect of this policy mix is that it subsidizes the US economy shifting away from manufacturing things with complex global supply chains, and towards services instead. (To be fair, this might be the US’s comparative advantage. But it’s not the original intention of the policy.)