Juniper Vale: You texted me at like 6:45 this morning, which you never do, and it was just a number.
Hope Sterling: 149 billion! I woke up and Bloomberg had just published that Meta is building a cloud business called Meta Compute to sell GPU capacity to outside companies, and I literally could not — like I sent you the number before I even finished the article.
Juniper Vale: And that's in one day, from a single report.
Hope Sterling: July 1st, one day, stock goes up 9 to 10 percent, $149 billion in market cap added, and they're talking about going directly at Amazon Web Services, Microsoft Azure, Google Cloud — like, not nipping at the edges, directly competing.
Juniper Vale: Okay but I want to back up for a second — because the infrastructure behind this isn't new. Meta's 2026 capex guidance is $125 to $145 billion. That's nearly double what they actually spent last year, which was $72.2 billion. Wall Street was calling that number reckless in, what, April?
Hope Sterling: Three months ago, yes, and now it's a feature — that's the thing that's messing with my head.
Juniper Vale: And Zuckerberg floated this at the May 27th shareholder meeting — his words were 'definitely on the table' — but only if Meta has excess capacity. That 'if' means they don't actually know yet whether this is a real product.
Hope Sterling: So the whole question is whether Meta Compute is an actual business or whether it's like, a story they're telling investors while the capex is already out the door — and I genuinely don't know, which is why I texted you at 6:45.
Juniper Vale: The story they're telling isn't wrong, it's just not new. Like, imagine you built a massive warehouse for your own inventory. Inventory doesn't fill it. So you start renting the empty shelves to other businesses. The warehouse didn't change. The strategy did. That's exactly what this is.
Hope Sterling: Okay but — wait — so the shelves were already built? Like the Hyperion campus in Louisiana, the Prometheus cluster in Ohio, all of that was already announced before July 1st?
Juniper Vale: Every bit of it. Hyperion's designed to scale to five gigawatts — five — and that's been on the books. Prometheus in New Albany, same buildout. The $600 billion U.S. investment plan through 2028, announced. Bloomberg just gave it a name — Meta Compute — and suddenly it's a pivot.
Hope Sterling: Stop. So Bloomberg named it and the stock added $149 billion.
Juniper Vale: Basically, yeah. And I mean — what makes this even harder to take seriously as a brand new strategy is xAI. SpaceX's xAI already signed a deal in early May 2026 to sell all the compute capacity at Colossus 1 to Anthropic. That playbook existed. It was documented. So this isn't even an original idea, it's the second time the market's seen it.
Hope Sterling: Wait, and CoreWeave — they're a current Meta customer, right? So Meta building this is like, their own tenant becoming their landlord's competition?
Juniper Vale: CoreWeave's stock cratered the same day, yeah. Which is actually the most concrete signal that the market read this as real — not just a vibe. Someone with money decided Meta entering this space directly threatens the neoclouds.
Hope Sterling: @AndreasSteno is out here calling Meta Compute 'the new AWS' and like — I get the energy, I feel the energy — but the math hasn't actually shown up yet, right? No margins, no pricing, no revenue targets disclosed.
Juniper Vale: Right — but the part that kills the 'new AWS' take is the timing. Meta's own AI model keeps getting delayed. As of early July, the Wall Street Journal's saying no confirmed release date exists. And the Meta Compute announcement drops at almost the exact same moment that news is circulating. That's not a coincidence.
Hope Sterling: Wait — like, simultaneously? The model delay and the cloud pivot are the same news cycle?
Juniper Vale: Basically the same week. And that's what makes me read this as defensive, not visionary. If your own models are stuck, you announce you'll sell compute to other people's models. It reframes the problem.
Hope Sterling: What's actually messing me up — like, why would an enterprise AI startup hand their mission-critical workloads to Meta Compute when Meta literally cannot ship its own model on a timeline? That's the question. Like, that's not a vibe concern, that's a real credibility problem.
Juniper Vale: And it gets sharper — Meta is the only one of the four major U.S. hyperscalers that doesn't already operate a cloud infrastructure business. AWS, Azure, Google Cloud each have decade-long enterprise relationships, compliance certifications, developer toolchains. Meta's starting from zero on all of that.
Hope Sterling: No way. Zero cloud business, and they're going directly at AWS?
Juniper Vale: From a standing start. And CoreWeave and Nebius — both stocks crater the same day — those are companies Meta is currently paying as infrastructure suppliers. So Meta's announcement punished its own vendors.
Hope Sterling: That's — I mean, that's a wild position to be in. Your landlord announces they're opening a competing building while you still have a lease with them. And honestly the model delays piece, like — I don't buy that's just bad luck, I think that's the actual tell that this whole thing is a pivot away from an embarrassing story. The math question though, pricing, margins, revenue targets — we need to get into all of that because that's where this either holds up or completely falls apart.
Juniper Vale: Yeah, and none of that's been disclosed. Which is exactly where we're going next.
Hope Sterling: None of it disclosed — okay, like, that's the part that physically stops me. The $149 billion single-day move came with no product launch, no customer name, nothing. Not one.
Juniper Vale: No pricing structure, no SLA history, no margin assumption — none of it. And I want to make that concrete, because it gets easy to let that slide as like, 'details to come.' Picture a CTO, late Q4 2026, sitting down to sign a multi-year compute contract. She's comparing AWS — fifteen years of compliance certs, enterprise support, the whole stack — against Meta Compute, which has... no disclosed pricing and a parent company that just delayed its own flagship model. That CTO is the actual test of whether this business exists.
Hope Sterling: Wait — and Muse Spark, like, that's literally the Meta model being floated for hosted access inside Meta Compute — but if Muse Spark has no release date, what exactly is she buying access to?
Juniper Vale: That's — yeah, that's the credibility loop. The cloud pitch partly rests on 'host our models,' but the models aren't out.
Hope Sterling: It's like — okay, actually no — it's not even just the models, it's the overcapacity assumption underneath all of this. Zuckerberg's 'if' on May 27th — if there's excess capacity — means Meta doesn't actually know yet whether they've built more than their own AI projects will eventually consume. And if their ambitions grow into every gigawatt of Hyperion, Meta Compute just... evaporates.
Juniper Vale: That's the question no one can answer right now. The whole business premise assumes the overcapacity is permanent, not temporary.
Hope Sterling: So what do we actually watch for? Like, what are the signals that tell us this became real?
Juniper Vale: Three things. A pricing disclosure — actual numbers, not a range. A named enterprise customer — not a rumor, a signed contract someone's willing to put their name on. And a confirmed release date for Meta's delayed AI model, because that one tells you whether the overcapacity is even real or whether it's already spoken for internally.
Hope Sterling: Until then it's a narrative with a $149 billion price tag and exactly zero receipts.
Juniper Vale: That $125 to $145 billion capex guidance has no disclosed Meta Compute revenue target sitting next to it. Nothing. So we literally cannot check whether the business justifies the spend. The math might exist internally. But it hasn't been shown.
Hope Sterling: And like — okay, this is the part that actually keeps me up — what if Meta's own AI projects eventually scale into every gigawatt of Hyperion? Like, the overcapacity that makes Meta Compute possible just... gets eaten. And the cloud business quietly disappears. Does that mean the capex was never a standalone bet? Or was it always just — temporary scaffolding, buying time until the story caught up to the spending?
Juniper Vale: I don't know. And I mean that — I genuinely don't know.
Hope Sterling: Yeah. Neither do I. Which is — honestly, for a $149 billion day, that's a wild place to land.
