NVIDIA and OpenAI say their strategic partnership targets at least 10 gigawatts of NVIDIA systems, with up to $100 billion invested progressively as each gigawatt is deployed and the first gigawatt planned for the second half of 2026 on Vera Rubin rather than a vague cloud ambition. [1]

The paper's June 14 feature said OpenAI's buying rails had gone public before its S-1, and Monday adds power and hardware to the same missing file because OpenAI says only that it submitted a confidential S-1 and has not decided timing. [2]

That means investors still cannot see whether commitments of this scale become capex, supplier financing, customer obligations, equity, debt or something else, which is why Oracle's AI-contract release matters as a map rather than a side story because power contracts are financial architecture. [3]

Oracle described huge AI remaining-performance obligations, negative free cash flow, debt and equity financing, and prepaid or customer-supplied hardware inside large AI contracts, the grammar of AI scale that allocation marketing on X turns into valuation, revenue and GPU claims before audited disclosure arrives. [3]

MSM can call the NVIDIA deal scale, but the paper's question is who pays for each gigawatt, who owns the hardware, who carries stranded power risk before revenue catches up, and when the S-1 makes the promise legible.

-- DAVID CHEN, Beijing