Miles Ashworth: Four hundred billion dollars. Two weeks. That's where I want to start.
Megan Skiendel: The post-IPO collapse.
Miles Ashworth: SpaceX IPO'd June 12th — Musk becomes the world's first trillionaire, the headlines write themselves — and then, in under a fortnight, a tech stock rout wipes enough to push him back below a trillion. SpaceX then floats a $25 billion bond offering and the stock wobbles further. That's the backdrop. And it is precisely into that wreckage that Reid Hoffman of Greylock Partners, invested in OpenAI and Anthropic, appears on Rana el Kaliouby's 'Pioneers of AI' podcast and calls xAI 'a complete train wreck' and SpaceX merely a premium-priced CoreWeave.
Megan Skiendel: The CoreWeave line is — wait, he said that specifically?
Miles Ashworth: Specifically. 'Not an AI company' — just leasing GPUs at a premium. And then he cited the Cursor acquisition, announced days after the IPO, as evidence: 'buying your way into relevance' rather than building it.
Megan Skiendel: Honestly — Grok is underperforming against the competition right now, that part's fair. But Hoffman has money on the other side of this trade. That framing matters.
Megan Skiendel: But here's what I want to separate out, because the headlines are blurring two completely different things. The xAI facts are verifiable and they're brutal — all eleven original co-founders gone by May 2026. Not some, not most. Eleven. And they're apparently on their third restart. That's not turbulence, that's the founding thesis collapsing twice. Grok lagging Claude and OpenAI's models — also real. That's the signal. But Hoffman then slides from those facts straight into dismissing SpaceX's compute business, and that's — actually, that's a totally different argument.
Miles Ashworth: The Reflection AI number.
Megan Skiendel: Six point three billion dollars. That's a real contract. SpaceX leasing GPU compute to Reflection AI — that's not branding, that's contractual revenue. CoreWeave does the same thing, sure. But the question is whether SpaceX's specific infrastructure, the Starlink integration, is somewhere CoreWeave literally cannot build.
Miles Ashworth: The warehouse analogy.
Megan Skiendel: Exactly — a company renting warehouse space isn't automatically a logistics company just because Amazon also rents warehouses. The question is whether your warehouses are in locations nobody else can build. That's the whole thing. If SpaceX's orbital infrastructure creates distribution nobody can replicate, Hoffman's CoreWeave comparison just falls over.
Miles Ashworth: Well, and he knows that. Which is precisely why he doesn't address the six-point-three billion directly. He addresses Cursor.
Miles Ashworth: The thing nobody's saying out loud is that Hoffman isn't a critic here. He's a competitor with a microphone. Greylock Partners holds stakes in both OpenAI and Anthropic — the two companies that lose most if xAI actually works. That's not incidental. That's the whole context.
Megan Skiendel: The conflict is real. I'm not disputing that.
Miles Ashworth: And then — wait, this is the bit that actually gets me — he called the U.S. government's ban on Anthropic models 'autocratic willy-nilly.' Which is a position that, purely coincidentally, benefits his Anthropic investment. Nobody in that interview pressed him on either of those.
Megan Skiendel: No, and — honestly, that's fair, that's the tell. But here's where I want to push back: the eleven founders leaving, Grok genuinely underperforming Claude — those facts don't need Hoffman's motives to be clean in order to be true. His conflict doesn't launder the data.
Miles Ashworth: The facts stand. Agreed. What doesn't stand is Hoffman positioning himself as the dispassionate diagnostician.
Megan Skiendel: Right — and actually, the inconsistency is more damaging than the conflict. He praised OpenAI and Anthropic coexisting as healthy competition. Multiple players, good for the ecosystem, all of that. Then xAI enters and suddenly it's pure zero-sum, complete wreckage. Same person. Nobody caught it.
Miles Ashworth: Because competition, it turns out, is healthy when he's winning it.
Megan Skiendel: And that's — honestly, that's where I land too. But I want to be specific about what's actually unresolved here, because it's not 'was Hoffman wrong.' It's more like — if SpaceX's Starlink integration turns the Reflection AI contract into something CoreWeave literally cannot replicate, then the 'premium-priced CoreWeave' framing doesn't just look biased, it looks prescriptively wrong. That's one future. The other future is the $25 billion bond offering and the post-IPO crash were telling us the AI growth story was always the inflation mechanism, and Hoffman was just — actually reading the room correctly. Those are genuinely different diagnoses and we're not close to knowing which one.
Miles Ashworth: Quite. And neither Grok's benchmark numbers nor Cursor's acquisition price settles it.
Megan Skiendel: No. So what actually moves the needle — is it a model benchmark, a compute contract renewal, a founder hire that signals the third restart actually stuck? What's the data point that tells you Hoffman was protecting OpenAI and Anthropic, or that he called it right from the beginning?
