For the first time, you're about to be able to own a piece of a frontier AI lab. Anthropic filed a confidential draft IPO registration (an "S-1") with the SEC on June 1, 2026, near a ~$965 billion valuation. OpenAI followed on June 9. Bankers are whispering "trillion-dollar debut." The hype machine is on.
Here's what almost nobody will tell you in the excitement: a confidential S-1 hides the only numbers that matter. A company can trade SEC feedback in private — without showing the public its losses, margins, or risks — until weeks before it actually sells you stock. Right now you're being sold a valuation and a story, not a business.
Don't buy the story. Wait for the document. When the public S-1 drops, here's what to read — in order.
You'll hear "$47 billion annualized run-rate, up from ~$10B in January." A staggering demand signal — but "annualized run-rate" means take one good month and multiply by twelve. It's revenue, not profit. The first number to find is the net loss and the cash burn. The question isn't "are they growing?" (obviously). It's how much are they lighting on fire to do it, and for how long can they?
These labs don't own their compute — they rent it, on enormous multi-year commitments to Google, Amazon, and Microsoft. Those obligations can be tens of billions and are the single biggest claim on future cash. Find the purchase commitments and related-party sections. A lab that owes a cloud provider more than it earns is not the same investment as one that doesn't.
Strip everything away and one question decides it: does the cost of answering a query fall faster than the price of staying at the frontier rises? That's the gross-margin trend. Cheaper-per-use faster than they spend → durable business. If not → a treadmill that needs endless new capital. The bull case and the bear case live in this single line.
Dual-class shares (do you get a vote, or just the risk?), the lock-up (when insiders can sell — the stock can drop when they do), and whether early investors are cashing out in the IPO or doubling in. Follow what insiders do, not what the roadshow says.
Notice three of those four come back to the same chokepoints: compute, energy, memory. The AI labs' entire fate rests on the cost of the picks-and-shovels — which is why, for most investors, the enablers of the buildout have been a cleaner bet than guessing which lab wins, and why the labs' own S-1s will be the best map yet of how dependent they are on those bottlenecks. Bet the bottleneck, not the believer.
Bending Spoons (Nasdaq: BSP, ~$19–20B) is the anti-AI-lab: an Italian "buy-and-revive" software roll-up that buys tired-but-loved apps cheap (Evernote, Vimeo, WeTransfer, Eventbrite, AOL), runs them ultra-lean, monetizes harder, and keeps the margin.
The lesson: one model sells you a profitable present, the other an extraordinary-but-unproven future. Neither is "better" — but know which one you're buying.
The part the hype skips. You almost certainly can't buy at the IPO price — those shares go to big institutions. What retail gets is the first-day pop: the stock often opens 20–50% above the IPO price before you can click. Chasing that pop is how regular people overpay. Two calmer truths:
None of this is "don't." It's: let the hype clear, read the public numbers, and let the calendar work for you. A great company bought at a euphoric price can still be a poor investment for years — the company and the stock are not the same thing.
We won't tell you to buy or avoid anything. We'll read the actual S-1 the day it goes public and give you the four numbers — losses, compute liabilities, gross margin, insider behavior — in plain English, while the rest of the internet reacts to the valuation headline. An IPO is a story a company sells you. The S-1 is the truth they're required to tell. Read the truth.
Educational, not investment advice. Dragonfly Lens is not a registered investment advisor. Figures as of late June 2026 and subject to change; the public S-1 supersedes all of it.