NVIDIA CEO Jensen Huang admitted it on camera. In an April 2026 interview with Dwarkesh Patel, Huang said plainly: 'If we didn't support CoreWeave to exist, they would not exist.' That sentence is the thesis of this piece. NVIDIA is not Enron. It is something more structurally interesting: a chip company that created an entire class of debt-funded customers, invested in them at every stage from seed to post-IPO, signed $6.3 billion purchase agreements with them, backstopped their data center leases, and then watched outside capital from Blackrock, Jane Street, and overeager asset managers flood in to foot the bill. The risk sits entirely off NVIDIA's balance sheet. The revenue does not.
Ed Zitron's second installment of the Hater's Guide traces the specific mechanism: NVIDIA replicates vendor financing, the same tactic that destroyed Lucent and Nortel, without technically providing any vendor financing. Lucent loaned money to customers like Winstar so they could buy Lucent gear. NVIDIA instead funded neoclouds like CoreWeave, Nebius, and IREN as an early investor, then became their anchor customer, then backstopped their leases, then let third-party debt markets capitalize the whole structure to the tune of tens of billions of dollars. Each new NVIDIA investment in a neocloud signals safety to outside capital, which piles in, which funds more GPU purchases. Australian startup Firmus just closed a $2 billion round including NVIDIA, Jane Street, and Blackrock, following a prior $300 million round NVIDIA also joined. The pattern is consistent and documented.
The piece is worth reading in full not for the conclusion but for the architecture of the argument. Zitron walks through how each NVIDIA relationship with a neocloud functions simultaneously as investment, customer contract, and financing guarantee, and why that structure transfers all utilization risk, construction risk, and debt load onto third parties while locking in NVIDIA revenue. The Big Short 2 dialogue is a gimmick, but the S&P revising CoreWeave's outlook to positive mid-scheme is the real punchline. If these neoclouds cannot generate returns on the capital they have already raised, the losses land on banks and asset managers, not on NVIDIA. That asymmetry is the story.
[READ ORIGINAL →]