Dr. Zuck in the Metaverse of madness

THE NUMBER: $83.6 billion — what Meta has lost on Reality Labs since 2020. A loss every single year, $19.1 billion in 2025 alone, and another $4 billion-plus gone in the first quarter of 2026. That’s the ghost in the room now that Zuck wants to raise tens of billions more for AI. When somebody asks you to fund their next big vision, the only fair question is how the last one went. Eighty-three billion dollars says ask it loudly.

On Friday the Nasdaq fell 4%, its worst day since the tariff mess of April 2025. The S&P dropped 2.64%, the Dow shed nearly 700 points, and the chip names got taken to the woodshed — Micron off 17% over two sessions, AMD down 12.6%. Everybody on X called it the AI crash. They grabbed the wrong half of the story. The trigger was macro: a hot May jobs report (172,000 jobs against 85,000 expected) lit up rate-hike fears, and Broadcom whiffed on its AI-chip outlook. Rates are the vise here, not the technology. Higher rates make a $145 billion capex bill more expensive to carry and discount every dollar of far-off AI revenue harder, all at once.

But underneath the macro, one specific thing actually spooked people, and it’s the story that matters. Meta floated the idea of raising tens of billions in stock, and the shares dropped 5.5%. Sit with that. Meta, one of the great cash machines ever built, got punished for wanting money. A year ago it couldn’t hand cash back fast enough. Now it’s passing the hat. Alphabet is already there, raising $85 billion. SpaceX is in line for up to $75 billion. The richest companies in the history of capitalism are all reaching for other people’s money in the same week.

Here’s the frame for the whole issue. A young lab raising money is doing its job — you bought the growth, you signed up for the burn. A cash cow raising money breaks a promise. And the market is now asking the question underneath the question: do the business models of Meta and Google even survive the AI era intact, and if so, what do they look like on the other side? That’s not a valuation debate. That’s an existential one for Meta and a dominance one for Google. Same instrument, three different kinds of madness, because Musk — the third body in this story — never signed the cash-cow contract in the first place.

💲 Meta Got Punished for Wanting Money

Start with the company in the headline. Meta (NASDAQ: META) fell 5.5% Friday, touching $585 intraday, after the Financial Times reported it’s weighing a stock sale of tens of billions to fund AI. Meta called the report “pure speculation.” Maybe. But the market’s reaction was the honest part, and it wasn’t really about dilution math.

It was about pattern recognition. This is a company that already torched $83.6 billion on Reality Labs chasing one man’s vision of a world nobody asked to live in. So when Zuck says “trust me, tens of billions more, this time for AI,” Friday’s selloff is the market muttering the obvious: we’ve seen this movie, and it had a metaverse in it.

Now look at the inventory the way an analyst would, not a fanboy. Facebook is going grey, increasingly the place your aunt argues about politics. Instagram still prints money but the feed is sliding toward AI slop, and the day the feed becomes slop is the day the engagement that is the ad asset starts to bleed (and a widely cited report this week pegged automated traffic at 57.2% of the internet, so the slop problem isn’t hypothetical). WhatsApp is a beautiful product that, fifteen years on, nobody has truly figured out how to monetize. The Ray-Bans are fine. Fine doesn’t carry a $145 billion capex year, and Apple is walking into that category at WWDC tomorrow. And Reality Labs is a crater.

There’s a scene in Other People’s Money where Danny DeVito’s Larry the Liquidator explains that he loves money more than the things it can buy, then reminds a room full of true believers that the last buggy-whip company made the finest buggy whip you ever saw, right up until it died. That’s the risk hanging over Meta’s raise. The capex itself isn’t crazy. The company is funding it three different ways at once — a contemplated equity raise, a $27 billion joint venture with Blue Owl where Blue Owl covers 80% of the Hyperion buildout, and literal fabric tents to bring compute online in half the time. When a trillion-dollar company is pitching tents and tapping private credit and floating a secondary, that’s not confidence. That’s a scramble.

The desperate part: the bull case left for Meta is “four billion users is the best distribution rail in the world to push AI through.” That’s a real asset. It’s also a bet, not a moat, and the market just told you it’s pricing the bet, not the moat. If you own Meta, you’re underwriting Zuck’s second world-changing vision in five years. Decide whether you’d have funded the first one knowing what you know now.

⚡ Google Rents Its GPUs From a Rocket Company

Now the other raise, and why it isn’t the same trade. Alphabet (NASDAQ: GOOGL) is raising $85 billion, and Warren Buffett anchored the first $80 billion of it — the patron saint of patient capital bought the toll booth (we flagged that on June 4). Google has the TPUs, the data, DeepMind, and a genuinely good Gemini. It’s raising from strength, to own the compute it runs on so it never has to depend on anybody.

Except it just did depend on somebody. The same week, Google agreed to pay SpaceX $920 million a month — roughly $30 billion through June 2029 — for access to about 110,000 Nvidia GPUs, described as a short-term bridge for Gemini Enterprise demand. Read that twice. The one company on earth that designs its own AI silicon, builds its own data centers, and runs its own cloud still got caught so short it had to rent capacity from Elon Musk. If Google can’t build fast enough to feed its own product, compute scarcity isn’t a talking point. It’s the binding constraint on the entire industry.

Connect the dots: Google’s raise is offense and Meta’s is defense, and Friday’s tape sold them as the same thing. They’re not. Google’s question isn’t survival — it owns your data, your browser, your phone, and a chip nobody else has. Its question is narrower and meaner: can it win the AI interface war without cannibalizing the roughly $175 billion search-ad machine when the world moves from ten blue links to one answer? That’s Clayton Christensen’s nightmare with Google’s own name on it. The innovator has to eat its golden goose on purpose, before someone else does it for free. I’m long Google here, and the $85 billion is why — it’s buying the freedom to make that choice on its own clock.

🦞 The Cartel Nobody’s Drawing on the Whiteboard

Here’s the part almost nobody is connecting, and it reframes everything above. In February, SpaceX and xAI merged — an all-stock deal valuing the combined company around $1.25 trillion, with xAI folded in as a subsidiary. So Colossus, the Memphis-and-Mississippi supercomputer, now belongs to SpaceX. Which means when Anthropic pays $1.25 billion a month to rent Colossus (our June 1 issue) and Google pays $920 million a month for “compute capacity at xAI data centers,” they are renting from the same Musk building. Two of the three frontier players are anchor tenants in Elon’s data center.

And Google doesn’t just rent it. Google owns roughly 5% of it — a stake from 2015, diluted by the merger, worth somewhere between $87 billion and $122 billion at the IPO range. So the $920 million a month Google pays out the front door comes back through the window as appreciation on its own equity. Better still, that contract is the thing SpaceX’s prospectus desperately needs: the AI segment lost $2.5 billion last quarter on just $818 million of revenue. An $11 billion-a-year Google contract doesn’t nudge that line, it rewrites it — roughly 13x the segment’s revenue overnight. Google isn’t hedging. It’s the anchor tenant who also owns the building and signs the lease that makes the building financeable. That’s not a hedge. That’s a flywheel.

Now watch where the knife points. Musk is the only player in this game who is short nothing. If the labs win, he collects rent (Anthropic, Google). If his own model wins, that’s Grok. If the application layer wins — and coding is the richest seam in it — he’s got a $60 billion call option on Cursor, the coding tool now living inside more than half the Fortune 500. Cursor is built on a fork of Microsoft’s own VS Code and it’s eating Microsoft’s GitHub Copilot for lunch. And Musk named his next $20 billion data center “Macrohard” — a literal middle finger to Microsoft, with the stated goal of building all-AI software to take the company down. He’s not hiding the strategy. He spiked the ball before the snap.

The real story: the board is consolidating into two empires — Musk (SpaceX/xAI/Grok/Cursor/Macrohard) versus OpenAI and its backer Microsoft — and the casualty is OpenAI. It’s the one company that can never rent from the Musk colossus, because Sam and Elon are in active litigation and the door is welded shut by blood. So OpenAI goes third through the IPO window, after Anthropic and a $1.75 trillion SpaceX have already drained the public’s appetite, while both its rivals stroll in with a glamorous compute alliance and OpenAI pays retail for Stargate.

I’ll push back on my own read, because the cartel story is seductive and half-true. Going third isn’t all bad — it gives OpenAI optionality to price off how Anthropic and SpaceX trade, and owning Stargate capacity is more durable than renting from a man who hates you and wrote a 90-day termination clause into Google’s deal. The Musk alliances win the narrative round. OpenAI is betting that owning your compute beats renting it from the enemy. Both bets are live.

📉 The Era of Cheap AI Is Over

One more leg, because the capital scramble has a twin on the revenue side. The reason everybody needs money is that nobody can yet charge enough to cover what this costs. Jacobin ran the math: ChatGPT costs something like $17 billion a year to operate against 800-900 million weekly users and only 35 million who pay. One analysis this week argued the labs may be spending more than $1,000 for every $100 of revenue they book. Sam Altman himself conceded that corporate worry about AI costs is “fair criticism.”

Shelly Palmer thinks the next 18 months of enterprise AI come down to one phrase: price per intelligence unit. The rate card already tells the story. Anthropic and OpenAI are priced in lockstep at $5 per million input tokens; Google undercuts both at $2 with Gemini 3 Pro. Subsidized intelligence is ending, and the company that can’t drag its cost down or prove its return faster is the one passing the hat next.

What this tells you: the capex problem and the revenue problem are the same problem read from two ends. The frontier labs can’t fund the compute and can’t yet charge enough to cover it. That gap is the entire reason Google, Meta, Anthropic, OpenAI, and SpaceX are all in the capital markets in the same fortnight. The whole industry is financing the distance between what AI costs and what anyone will pay for it.

What This Means For You

Capital is the moat now. Not the model, not the benchmark, not the founder’s pedigree. In a world where compute leads the corporation by a mile, where even Google has to rent from a rival, the only durable sovereignty is a balance sheet big enough to feed your own machines. Everything Friday was the market discovering that even the richest companies alive may not have one.

Don’t buy “Big Tech capex” as one trade. Google raising to attack and Meta raising to survive are opposite bets wearing the same costume. Sort your exposure by who’s holding a war chest and who’s holding a ransom note.

Watch the contract, not the headline. A growth company raising capital is on-thesis; a cash cow raising capital just told you its old model can’t fund the new war. The tell isn’t the dilution. It’s the broken promise.

Own your compute or own a backup. Single-source compute makes you a hostage, and even Google found that out this week. Route the frontier model for the hard 10%, a cheap good-enough model for the rest, and keep a second supplier wired up.

The bubble and the buildout are the same event read from two chairs. The technology keeps compounding the whole time the financing gets harder. Don’t bet on whether AI is real. Bet on who can still write the check when money stops being free.

Three Questions We Think You Should Be Asking Yourself

If you own the megacaps, can you tell offense from defense in your own portfolio? Google buying compute sovereignty and Meta funding a second moonshot are not the same risk. If you’re holding both because they’re both “AI,” you’re not investing, you’re indexing the narrative.

Where is your business renting its most important capability? Google got caught short on the one input it thought it controlled. Whatever your equivalent is — compute, talent, distribution, a single platform — what’s your 90-day plan if the landlord raises the rent or locks the door?

Are you funding a vision or paying tuition? There’s a real difference between spend that’s building a base and spend that’s chasing a Reality Labs dream. Before your next budget cycle, be honest about which one your AI line item actually is, because the market is about to grade everyone on exactly that.