How to Spot a Paradigm Shift Before Wall Street Does

This is the big one. If you only track one signal, track this.

Every new technology starts out as an expensive toy for rich people, governments, or nerds. Then the cost falls. And falls. And keeps falling.

Then, one day, the new way costs one dollar less than the old way... and everything changes.

That line, where new gets cheaper than old, is the cost crossover. It’s the moment a technology stops being a choice and starts being a spreadsheet decision.

Some crossovers you’ve already lived through:

• Solar vs. coal. Utility-scale solar got roughly 90% cheaper across the 2010s. Somewhere in the middle of that decade, it crossed below new coal in sunny places. After that, utilities stopped building coal plants, not because of activists, but because the math flipped.
• Lithium-ion batteries. Around $1,200 per kilowatt-hour in 2010. Under $150 by 2020. The crossover with the internal combustion engine on a total-cost-of-ownership basis happened quietly in the early 2020s. Then EV sales took off.
• Sequencing a human genome. $3 billion. Then $100 million. Then $1,000. Then a few hundred bucks. At $1,000 it became something a hospital could order. At a few hundred it became something a startup could bring to a million people.

CFOs don’t buy “cool.” But they do buy “cheaper.”

So the number that really matters in frontier tech, much more so than the stock price, is the unit cost.

• Cost per kilowatt-hour
• Cost per million tokens
• Cost per kilogram to orbit
• Cost per genome
• Cost per mile driven without a human

When the gap between the new way and the old way gets under 20%, you’re close. When it hits zero, the shift is already underway.

Two warnings:

First, “cheaper” has to mean cheaper all-in. A robot arm that’s cheaper than a worker’s salary but needs a $200k integration project isn’t across the line yet.

Second, crossovers usually happen in one narrow niche first. Solar crossed in Chile and Arizona before it crossed in Germany. Language models crossed for “summarize this email” long before they crossed for “write my legal brief.” Look for the first niche, not the whole market.

Utility-scale solar cost per MWh, 2009–2024 (Lazard): ≈$359 → ≈$60. Coal barely moved. The lines crossed around 2016—the market spent years arguing with the chart.

Wall Street watches capital flows. What it’s bad at is watching people flows, and people move first.

Think about who moves early: the 28-year-old with a PhD who could work anywhere and takes a 30% pay cut to join a 40-person company you’ve never heard of. That person has better information than any analyst. They’ve seen the demo. They’ve talked to the founders.

When a lot of those people move in the same direction at the same time, that’s the signal.

Some famous migrations:

• In 1957, eight engineers walked out of Shockley Semiconductor to start Fairchild. That one walk-out basically created Silicon Valley.
• In the early ’90s, physicists started leaving academia for Wall Street. Quant finance was about to eat the industry.
• Between roughly 2015 and 2020, the best machine learning researchers at Google, Facebook, and top universities started leaving for tiny AI labs with weird names and no revenue. You know how that turned out.

How to track it without inside connections:

Watch the job boards, not the press releases. When a company posts 40 roles for “robotics controls engineer” in a single quarter, they’re cooking something up.

Watch the grad students. Ask anyone finishing a top PhD program what their smartest classmates are doing. Whatever it is, that field has 5 to 10 years of tailwind. Grad students are choosing where to spend entire careers with no capital at stake, so they pick purely on where the future is.

Watch the “why.” This is important. Talent moving for equity in a small company is a shift signal. Talent moving for giant guaranteed salaries at big companies is a hype signal. In 2021, MBAs flooded into crypto for the pay. In 2016, researchers flooded into AI for the problems. Those are not the same migration.

Share of new AI PhDs going straight to industry, 2004 vs 2020 (Stanford AI Index). The repricing arrived in 2023. The résumés were public the whole time.

Every hot technology has two customer groups:

1. Enthusiasts. Early adopters. The ones who post about it. They’ll buy anything shiny and they’ll tell you all about it.
2. Boring customers. Insurance companies. Regional banks. Municipal water utilities. The procurement department at a Midwestern hospital chain.

Enthusiasts create hype. But boring customers create paradigm shifts.

A boring customer has to fight through six layers of approvals, a legal review, a security audit, and a CFO who’d rather set the money on fire than take a risk. If they buy anyway, it’s because the thing works and the cost crossover already happened.

AWS is the perfect example. From 2008 to 2012, thousands of real companies with real IT departments quietly moved workloads to the cloud. Amazon didn’t even break out AWS revenue until 2015, at which point it was already a $5 billion business and Wall Street collectively spit out its coffee.

Or take John Deere buying a computer-vision startup called Blue River in 2017 for over $300 million. A tractor company. Buying an AI company. That’s a boring customer telling you exactly where agriculture was going.

So where do you look?

Earnings call transcripts of companies that are NOT in the sector. Skip the tech companies. Read the calls from Walmart, Caterpillar, UnitedHealth, Duke Energy. Ctrl-F the technology you care about. When a utility CFO says “we’re deploying grid-scale storage at three sites this year” in a flat, bored voice, that’s the signal.

Capex line items and 10-Ks. Boring companies have to disclose big purchases. “Investment in automation equipment” going from $40 million to $210 million is a tell.

Supplier customer-concentration disclosures. Small public suppliers have to tell you when one customer becomes more than 10% of revenue. When a tiny sensor company suddenly says “one automotive customer represented 34% of sales,” we’ve just learned something important.

AWS’s size when Amazon first broke it out in 2015—after seven years of boring customers quietly moving workloads. Wall Street collectively spit out its coffee.

Most investors treat regulation as a threat. But there’s a moment in every paradigm shift when regulation flips from blocking the new thing to accommodating it, and that flip is one of the cleanest signals you’ll ever get.

Regulators are the slowest, most incumbent-friendly institutions around. They don’t bend for hype. They bend when the technology has become so inevitable that not having rules for it looks worse than having them.

The regulatory turn basically says: the incumbents lost the lobbying fight.

Some examples:

• Drones. For years, commercial drone use in the U.S. was effectively illegal. In 2016 the FAA published Part 107, a simple set of rules for commercial operators. The industry went from hobbyists to a real business shortly after.
• EVs. California’s zero-emission vehicle mandates and the EU’s phase-out timelines didn’t cause the EV shift. They confirmed it. Governments don’t ban a product category until they’re sure the replacement exists.
• Spot Bitcoin ETFs. After a decade of “no,” the SEC said yes in January 2024. Whatever you think of crypto, that was the regulatory turn, and tens of billions of dollars flowed in within months.
• Nuclear. After forty years of tightening, U.S. nuclear policy started loosening around 2024 with bipartisan legislation aimed at speeding up approvals for new reactor designs. That kind of reversal doesn’t happen because of a good podcast episode.

How to track it:

Read the boring dockets. The FDA, FAA, FCC, SEC, NRC all publish proposed rules. Read the summary and ask one question: is this making the new thing easier or harder? Early documents call a technology a “risk.” Later ones call it a “framework.” Framework means they’ve accepted it exists.

Watch for the first friendly jurisdiction. Arizona for self-driving cars. Wyoming for crypto. Singapore for cultivated meat. When one place bends, others usually follow within a few years.

Watch what incumbents ask for. When the old industry stops lobbying to ban the new thing and starts lobbying to regulate it, they’ve admitted defeat. Taxi companies in 2012 wanted Uber gone. By 2016 they just wanted Uber drivers to get background checks. That’s a turn.

Between the FAA’s last two new categories of civil aircraft: helicopters in the 1940s, and powered-lift air taxis in 2024. Regulators don’t move for vaporware.

This one’s our favorite, because it’s almost pure physics. You have to buy the parts before you can build the product, so a shift shows up in suppliers’ order books before it shows up in anyone’s revenue.

Basically, the picks-and-shovels guys sell out before the gold rush even hits the newspapers.

Some tells from history:

• Through 2023 and 2024, TSMC’s advanced chip packaging (the stuff that glues AI chips together) was completely sold out, with lead times stretching out for many months. That was visible in earnings calls before most people understood why every AI company was constrained.
• Lithium carbonate prices went up roughly 5x between early 2021 and late 2022. If you were watching the price of a dull white powder, you knew EV production was about to explode before the automakers told you.
• In the ’90s, the dot-com boom was visible in Cisco’s backlog before it was visible in anyone’s browser.

The supply chain can’t lie. A press release can say “we’re seeing strong interest.” But a backlog either exists or it doesn’t.

What to watch:

Lead times. When a component that used to ship in 4 weeks now ships in 40, someone is buying it faster than it can be made. Distributors like Digi-Key and Mouser show stock levels and lead times publicly, for free. This is one of the most underrated data sources in tech investing.

Price increases that stick. Normal supply hiccups cause spikes that reverse. Paradigm shifts cause price increases that stay elevated for years because demand keeps growing. Check the commodity or component price two years later. Is it still up? Then it wasn’t a hiccup.

Supplier capex announcements. Suppliers are terrified of building capacity nobody wants, so a giant new factory means they’ve already seen the orders.

Second-order suppliers. Who supplies the supplier? Who makes the machines that make the chips? Those companies see the shift earlier and get less attention.

The supply-chain tell can show up 12 to 18 months before the end-product companies report big numbers. That’s the window.

Lead times for large grid transformers, roughly 2020 vs today. Electrification demand hit a supply base built for replacement demand. Nobody put out a press release.

This is the last signal, and the one that tells you the shift is basically complete.

Every new technology gets an adjective when it’s born. The adjective is how we mark it as different from the normal thing.

And at some point, the adjective simply disappears.

Nobody says “digital camera” anymore. It’s a camera. Nobody says “mobile phone.” It’s a phone. When was the last time you heard someone say they “streamed” a movie instead of “watched” it?

The adjective drops off when the new way becomes the default way. At that point, the shift is already here.

So why is this useful, if it’s the last signal?

Two reasons.

First, the adjective drops in stages. Listen to who drops it first. Engineers stop saying “AI-assisted coding” and just say “coding” long before your uncle does. Watch the language of the people closest to the technology. When they stop qualifying it, you’ve got an early signal.

Second, the language shift has an evil twin: the adjective attaching to everything. In 2017, every company was a blockchain company. In 2021, everything was a metaverse. In 2023, every product suddenly had “AI-powered” in front of it. When the adjective spreads like that, it’s a pure hype cycle. Real adoption looks like the adjective falling off. Fake adoption looks like the adjective getting stapled on.

Okay, so we’ve covered six signals.

Think of them as a scorecard. For any promising technology, we can go through the list:

Then count.

Of course, the obvious disclaimer: spotting a paradigm shift is not the same as picking a winning stock. The internet was a real paradigm shift and most dot-com stocks went to zero. The shift tells you where the future is going. It doesn’t tell you who gets paid.

But here’s what this framework does:

It gets you out of the business of reacting to headlines and into the business of watching evidence. Headlines are what Wall Street trades on. Evidence is what Wall Street eventually catches up to.

The “next big thing” is already in motion.

Exo/Signals tracks frontier technologies before the breakout. Get the signals that move markets, early.