You don't need to pick the next AI winner. You probably already own them
The AI monsters are hiding behind the balance sheets of the tech-giants. What's going to happen when these cash-burners go public later this year?
The PC revolution was supposed to belong to the bedroom programmer. Anyone with a Tandy or an Apple II could ship software on a floppy disk. The kid in Idaho with a soldering iron did not become a billionaire. Microsoft did, by selling the OS every machine had to run. Intel did, by selling the chip every machine had to use. This is the structural pattern of every great democratization of building. Railways minted Carnegie. The internet minted AWS. Ecommerce minted Shopify. When the cost of building drops, the builders compete each other into the dirt. The landlords compound.
I am the case study for the AI version of this. In the last 6 weeks, I have built two websites end to end, using Claude code (how ‘26 of me, I know). With end-to-end guidance from Claude and a few nuances along the way, I managed to get it done, even to the extent of building robust crawlers to facilitate the DC Index. The same work would typically cost me comfortably $20,000+.
To be clear, I am not a developer, despite my similarly nocturnal nature. I am, on a good day, a guy who can describe what he wants and hope the preview window handles the execution with the design I like. But with a few software subscriptions at probably less than $100, who’s picking up the added benefit? They redistribute to Claude, Vercel, GitHub, Stripe, and 30% to Apple or Google if any of it ever lands on a phone. The dev gets paid once. The toll roads get paid forever. But this is all pretty obvious, as the saying goes: ‘be the one that’s selling picks and shovels.’ So what’s the opportunity today for folks like us (those that aren’t yet tech-billionaires)?
Just a couple shovel makers in the early days
OpenAI and Anthropic are still hiding behind a shield of private capital.
In the 80s, you couldn’t buy Microsoft until 1986. In the 2010s, the great toll roads of commerce (Stripe, Shopify pre-IPO, AWS buried inside Amazon) were locked in private rounds. By the time retail could touch them, the easy 100x had already happened. This time, almost every toll booth is already public. Apple, Alphabet, Microsoft, Amazon, Shopify, Cloudflare, MongoDB, Datadog, Snowflake. Stripe is the last meaningful private hold-out. For the first time in modern technology history, the picks-and-shovels trade is fully accessible to anyone with a Robinhood account.
The second thing that’s different, and the part most retail commentary has missed, is the quality of the balance sheets.
In 2022, the tech and DTC washout was brutal because most of those companies were structurally fragile. Burning cash, or not profitable (some still aren’t), funded by cheap capital. There was a wild stat that came up in one of our Agora board meetings in 2021 that claimed:
‘There was only 1 company that was both founded and IPO’d since 2008’s financial crisis, that was TTM profitable.’
I dug for a week and found the answer at the time was Zoom. There may have been a few others in the end, but you get the idea. In an era where we’d seen more IPO’s than any other in history, these businesses weren’t turning a profit. Here are some of the crowd favorites and how they performed publicly in that time period:
The companies that boomed during 2019-2021 were in fact mostly not profitable. As a result publicly they performed poorly. Comparably a lot of the stocks today are still private, and backed by public companies who are still generating a huge amount of profit.
But the excitement this time is the security putting up collateral for these monsters. Privately, OpenAI and Anthropic amongst close competitors are all also burning through capital like the above but at much higher rates. BUT they’re not public, YET. So investors can keep making up valuations out of thin air and piling in dollars. The companies guaranteeing their energy in exchange for equity, or simply investing, are in fact cash flush, and right now the risk is being decided by these firms, not the public.
The toll roads are the opposite shape of business. Apple holds roughly $50B in cash and marketable securities. Microsoft, around $75B. Alphabet, $95B. Amazon, $95B. Meta, $70B. Nvidia, $45B. That’s not even total assets which is in the hundreds of billions. Combined, these AI mega-caps sit on more than $430B in liquid assets, more cash than the entire DTC sector had at peak market cap. Shopify ended 2024 with roughly $5B in cash and is now consistently free-cash-flow positive. Cloudflare, MongoDB, Datadog, and Snowflake are all either profitable or FCF-positive with multi-year runways. None of them needs the capital markets to be open to survive a downturn.
So the trade is genuinely uncommon. A multi-decade structural tailwind, applied to companies that are already over-capitalized, in a category that is now, for the first time, fully accessible to anyone with a Robinhood account.
The Shopify model, applied to all of software
The cleanest example of this pattern, and the one most relevant to brand founders, is Shopify.
Shopify does not sell products. Shopify does not have a brand the consumer really cares about. Most shoppers have no idea whether the store they bought a $40 candle from runs on Shopify, WooCommerce, or a website hand-coded by the founder’s cousin. They just bought the candle.
Shopify takes 2.9% of every transaction processed through Shop Pay, plus a monthly subscription, plus capital revenue, plus the new ad layer. They are completely indifferent to which merchant wins. They are also indifferent to which merchant loses, because losing merchants pay a Shopify subscription right up until the moment they shut down the LLC. In FY24, Shopify processed roughly $300B in GMV. They keep a slice of all of it. So of course, they’re pro-growth but they don’t really care if you survive over the next guy is what I’m saying.
A definite risk here for Shopify is the increase in platforms such as TikTok Shop, or similarly where users shop in-app natively, and so the ‘host’ of the transaction changes, which they’re already looking to solve.
This model is now being copy-pasted into every category of software. Vercel is Shopify for hosting. Stripe is Shopify for payments. GitHub is Shopify for code repos. The App Store is Shopify for distribution. They are all running the exact same play: do not pick winners, charge rent on every attempt and focus on pioneering the infrastructure.
This matters more in an AI build-out than at any prior moment, for one specific reason. The number of attempts is going vertical. AI does not just make the winners cheaper to build. It makes the losers cheaper to build too. Every founder who used to give up at ‘we need to hire a dev for $80k’ now has a working prototype on Vercel by Sunday night. Most will fail. The infrastructure layer charges the same whether they fail or not.
Where to park your savings?
The Mega Cap companies selling the picks and shovels for the AI boom. These now account for almost 40% of the entire S&P, meaning the NASDAQ technology ETF’s are trending similarly to the S&P.
Every one of these companies grows at the rate of all software, multiplied by the AI productivity multiplier. If AI makes 10x more apps possible, they grow at roughly 10x the underlying market. That’s not to say they’ll grow publicly by that amount, but they’re facilitating that growth in volume. None of them have to be right about which app wins. They just have to keep the lights on, which is increasingly an energy problem in itself, but that’s another article. So, you don’t have to pick the next AirBnB. You pick the company that hosts the next 50,000 founders trying to be the next AirBnB, of whom 49,997 will fail, all of whom will pay rent every month while they do. Apple’s App Store generated an estimated $100B+ in developer billings in 2024. Stripe processed over $1.4T in payment volume. Cloudflare runs roughly 20% of the web. GitHub crossed 100M developers. Each is a denominator AI is about to multiply.
The uncomfortable truth at the bottom of all of this is that the people getting squeezed are the developers. The mid-tier dev shop charging $25,000 to build a marketing site is living in fear. The agency taking three months to ship a CRUD app is gone. The freelance front-end dev billing $120 an hour for a landing page is, well, let’s say Upwork has slowed down for them. Not in five years. Right now, this quarter, while you read this. The money that used to flow to them flows in exactly one direction. Up. To the hosts, the rails, the stores, the repos.
And here’s the part most young investors miss: you don’t even have to pick the individual toll roads or even invest in the smaller hosts, that are however turning a profit. Why? Because as The S&P 500 is now closing in on roughly 40% exposure to the Magnificent Seven stocks. The S&P, in 2026, is no longer a ‘broad market’ index in any meaningful sense. It is an AI infrastructure index with some banks and consumer staples bolted on the side. Buying VOO or SPY gives you, by accident, the exact thesis of this article, with no stock-picking required. In a gold rush, you do not want to be panning. You want to be the bank. And for the first time, the bank is publicly listed, and you already own a slice of it.