Nine Engines Behind Every 100-Bagger

In 1956, IBM shipped the first hard drive. It stored 5 megabytes. It weighed about a ton. They loaded it onto airplanes with a forklift.

Today a microSD card the size of your pinky nail holds a terabyte. That’s 200,000x more storage in something 1,000,000x lighter.

Nobody at IBM in 1956 planned that. It just... happened. Every year, for 70 years, the stuff got smaller and cheaper and nobody could stop it.

This is the oldest and most reliable engine on the list. You know it as Moore’s Law: transistor counts on a chip double roughly every two years. But the real lesson isn’t about transistors. It’s this:

When something gets smaller, it gets put in places it could never go before.

• A computer that fills a room → a computer in your pocket → a computer in your earbuds → a computer in your pacemaker.
• Each step down in size opens up a market that was literally impossible the step before.

That’s why the money jumps. Apple didn’t get 500x bigger by selling more desktops. It got 500x bigger because the desktop shrank enough to fit in a pocket, and suddenly 7 billion people were customers instead of 700 million.

More storage in a fingernail-sized microSD card (1 TB) than in IBM’s first hard drive (1956: 5 MB, about a ton, loaded onto airplanes with a forklift).

Any technology where the cost-per-unit-of-capability is dropping 30%+ a year. Sensors. Batteries. Gene sequencers (a human genome cost $100 million in 2001; today it’s a few hundred bucks). When you see that curve, ask: what new place does this fit into in two years that it can’t fit into today? That’s where the next market is.

Here’s a weird question: why would anyone buy the very first fax machine?

You can’t fax anyone. It’s a paperweight.

But the second person who buys one makes the first one useful. The thousandth makes it essential. By the millionth, you’re an idiot for not owning one.

That’s a network effect. The product gets more valuable the more people use it.

Compare that to a normal business. If Ford sells you a truck, your cousin’s truck doesn’t get any better. If Visa signs up one more merchant, every Visa card on Earth just got slightly more useful. Visa spent zero dollars for that improvement.

This is why network-effect companies have insane profit margins that make old-school analysts angry. The product improves itself and marketing becomes optional. Competitors can’t catch up by building a better product, because the product was never the point. The people were the point.

Warning label: Network effects run in reverse too. When people start leaving, every departure makes it less worth staying. MySpace didn’t decline gracefully. It collapsed. So while this engine is powerful, it’s not a moat you can fall asleep behind.

Connections possible among 1,000 users. Among 10 users: 45. As user count grew 100×, value grew 11,100×.

Does the ten-millionth customer make the product better for the first customer? If yes, you’ve found this engine. Marketplaces, payment rails, communication tools, and any platform where other people’s presence is the product.

This one gets confused with network effects, so let’s split them clearly:

• Network effect: More users = more valuable.
• Feedback loop: More usage = a better product. Even from the same users.

Google search is the textbook example. Every time you click the third result instead of the first, Google learns something. Billions of clicks a day. The product you use tomorrow is smarter than the one you used today, and you trained it for free.

Tesla is a physical version. A million cars drive around with cameras. They send back footage of weird intersections and confusing lane markings. That footage trains the software. The software gets pushed back to the cars. Now every car handles that intersection better.

Here’s why this matters for your money: A company with a feedback loop has a time advantage that can’t be bought. A competitor with $10 billion can build the same factory. They cannot buy five years of usage data. They have to earn it, one click at a time, while the leader keeps pulling away.

Amazon reviews. Netflix recommendations. Spotify playlists. Every LLM you’ve used this week. All of them are feedback loops.

Google searches on an average day—each one a tiny, unpaid contribution to a product no competitor can replicate by writing a check.

Does the company’s product improve because it’s being used, automatically, with no new engineering? And critically: does the company own the data coming back, or are they renting it from someone else’s platform? Own the loop, own the future.

In 1936, an aircraft engineer named Theodore Wright noticed something strange at his factory.

Every time they doubled the total number of planes they’d ever built, the cost per plane dropped by about 15%.

Not because of any groundbreaking invention. Because workers got faster. Suppliers got better. Someone figured out a smarter way to route the wiring. A thousand tiny improvements that only show up when you actually build the thing a lot.

This is called Wright’s Law, and it’s held up across almost every manufactured product for 90 years.

The formula is dead simple: Double cumulative production → cost drops by a fixed percentage.

Solar panels are the poster child. In 1976, a watt of solar cost about $106. Today, it’s about a dime. Again, not from any one breakthrough. That’s fifty years of factories doubling and re-doubling their output, with each doubling shaving another ~20% off the price.

Lithium-ion battery packs: roughly $7,500 per kWh in 1991, around $108 by 2025. Same story.

Why should you care? Because learning curves are predictable. If you know the learning rate and you know how fast production is growing, you can forecast the price years out with scary accuracy.

A product where cumulative units are doubling every few years, and management talks about cost-per-unit going down (not just revenue going up). When cost drops below the price of the incumbent alternative... that’s when the floodgates open.

Pull out your phone. Now count the objects it replaced:

Camera. Camcorder. Flashlight. Map. Compass. Calculator. Alarm clock. Walkman. Radio. Notepad. Calendar. Photo album. Level. Scanner. Wallet (getting there). Landline. Rolodex. Encyclopedia. Boarding pass. Car keys.

Twenty physical products, each of which used to be a company or an entire industry, now live inside one slab as software.

That’s dematerialization. Something that used to be made of stuff becomes made of information.

And as a result, the economics flip completely.

A physical product costs money to make every single copy. Blockbuster had to buy a plastic DVD, ship it, shelve it, and hope it came back. Netflix’s copy of the same movie costs almost exactly nothing to deliver to the 100-millionth customer.

• Physical business: Sell 2x more, spend almost 2x more.
• Dematerialized business: Sell 2x more, spend... about the same.

This is why software companies run 80% gross margins while hardware companies fight for 30%. A dematerialized company can go from $10 million to $10 billion in revenue without building a single new factory. Netflix went up over 100x doing exactly that: turning a physical rental store into bits.

Any industry still moving atoms around when the value is actually in the information. Paper contracts. Physical bank branches. Radiologists reading films. Wet-lab experiments that could be simulated. The moment someone turns those atoms into bits, the old cost structure dies and the margin explodes.

Encarta was Microsoft’s encyclopedia, with professional writers, fact-checkers, and editors. Real budget, real quality.

Wikipedia was a bunch of random strangers on the internet.

Guess which one still exists.

Wikipedia didn’t beat Encarta by being better funded. It beat Encarta because it didn’t have to pay the people who built it. Millions of volunteers wrote more articles in more languages than Microsoft could ever afford.

That’s crowdsourcing, and almost every big platform of the last 20 years runs on it:

• YouTube makes no videos.
• Airbnb owns no hotels.
• Uber owns no cars.
• The App Store wrote almost none of its apps.
• Every AI model you’ve used was trained on text that other people wrote for free (controversial, yes; but here we are).

The company builds the stage. The crowd builds the show. The company keeps a cut of every ticket.

The risk: The crowd can leave. Drivers can switch apps. Creators can move platforms. The best crowdsourced businesses combine this engine with #2 (network effects) so the crowd is locked in by each other, not by the company.

Who creates the inventory? If the company creates it, growth costs money. If customers create it, growth is nearly free.

Every engine above still has one bottleneck: someone has to make the thing. A chip fab costs $20 billion. A battery plant takes years. Even software needs engineers (yes, even with agentic coding).

Biology doesn’t have that bottleneck.

A single E. coli cell splits in two every 20 minutes. Left alone with enough food, one cell becomes a billion cells in about 10 hours. Nobody built a factory. The product is the factory. It makes copies of itself.

Humans have used this for 10,000 years without thinking about it (that’s what beer, bread, and cheese are). What’s new is that we can now program it.

• Precision fermentation: Engineer a yeast cell to pump out insulin, or milk protein, or spider silk. Then let it copy itself in a tank. The tank scales; the R&D doesn’t have to.
• mRNA: Moderna went from receiving a virus’s genetic sequence to a finished vaccine design in about 48 hours. Once you have the sequence, your body’s own cells do the manufacturing.
• Engineered organisms that eat plastic, fix nitrogen for crops, or brew jet fuel.

Here’s the investor lens: in a self-replicating system, the marginal cost of scaling trends toward the cost of sugar water. You do the hard work once (designing the cell), then biology handles the copying.

Big fat warning: Biology is messy. Timelines slip. Cells mutate. Regulators move verrry slowly. Most synthetic biology companies from the last decade lost money for shareholders because “it works in a flask” and “it works at 100,000-liter scale” are very different sentences. This engine is real, but it’s early.

From receiving the virus’s genetic sequence to a finished vaccine design at Moderna. Once you have the sequence, your body’s own cells do the manufacturing.

A company that’s past the science-project stage and has actually proven commercial scale deployments.

In 1900, factories ran on steam. One big engine, one giant shaft down the middle of the building, with belts and pulleys running to every machine. Then electricity showed up.

The first thing factory owners did? Swap the steam engine for a singular, huge electric motor. Yet productivity barely moved for thirty years.

Then someone realized: you don’t need one big motor. Put a small motor on every machine. Now you can lay out the factory however you want. Now you can create assembly lines. Productivity exploded.

That’s a substrate shift. The thing runs on a fundamentally different physical foundation.

Computing has done this before: vacuum tubes → transistors → integrated circuits. Each shift wiped out the old leaders and minted new ones.

And it’s about to do it again, through three candidates:

• Quantum computing: Instead of bits that are 0 or 1, qubits that can be both. Overkill for email. Potentially world-changing for chemistry, materials, and cryptography.
• Neuromorphic chips: Instead of a CPU that processes instructions one at a time, silicon that mimics neurons and fires only when it needs to. Potentially 100x more energy efficient for AI tasks.
• Photonic computing: Instead of electrons, light. Faster, cooler, and it doesn’t care about the walls Moore’s Law is hitting.

Here’s the catch: Substrate shifts are the most lucrative and the most dangerous engine on this list. They’re lumpy. They take decades. Ninety percent of the early bets go to zero. But the winners don’t get 10x; they get a whole new S-curve to themselves.

Vacuum tubes in ENIAC (1945): thirty tons, a room of its own, less compute than a musical greeting card. The transistor didn’t beat the tube business—it deleted it.

Don’t buy the “steam engine swap” (the company using the new substrate to do the old job slightly better). Buy the company doing something that was physically impossible on the old substrate. And watch one number: cost per useful operation. Until that crosses the old technology’s line, it’s science. After it crosses, it’s a market.

Look at the iPhone:

Five engines, not one.

Now look at AI in the last five years:

Six engines running at once. That’s why it feels like it came out of nowhere. It didn’t, but six different curves crossed at the same time.

This is the whole framework in one sentence: One engine makes a good company. Three or more make a 100-bagger candidate.

Next time someone pitches you a stock in “exponential tech,” look past the TAM slide. Don’t ask about the CEO’s vision. Grab this list and start checking boxes:

Score of 0–1: Normal company. Might still be great. Not a 100-bagger candidate.

Score of 2: Interesting. Watch it.

Score of 3+: Now we’re talking. Do the real work.

Of course, this framework tells you which companies can go 100x. It doesn’t tell you which ones will, or when, or whether the current price already assumes it. Plenty of businesses have had three engines running and still died because they ran out of cash a year too early.

Godspeed.

The “next big thing” is already in motion.

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