The week's capital events arrange on a single structural line. Google closed $85 billion in AI investment. Anthropic filed for what Wired is describing as the largest IPO in American history. Microsoft dissolved its exclusive compute relationship with OpenAI, decoupling the lab from Azure at the infrastructure layer. Each move operates the same flywheel: enterprise deployment revenue from API contracts funds the next training run on B200 NVL72 hardware that BIS October 2023 export controls keep out of Hsinchu fabrication supply lines and PRC data-centre procurement. The Anthropic IPO converts that flywheel into public-market capital, a financing mechanism unavailable to Baidu, SenseTime, or Zhipu AI, none of which operate a deployment surface generating the quarterly revenue that justifies a public offering at this scale.

Microsoft severed the compute-deployment bridge. What the company dissolved is not a research partnership but the mechanism that let OpenAI route Azure deployment revenue back into training runs on B200 NVL72 clusters inside Microsoft's infrastructure. OpenAI now finances that compute access independently, presumably toward its own public offering. The bilateral shape has three desks: Anthropic, OpenAI, and Google DeepMind each running independent capital-formation pipelines, each compounding BIS-controlled hardware access at a rate Baidu's Ascend 910B fleet at its Suzhou data centres cannot match, because the feedstock is US public-market capital formation, and that mechanism is closed to firms operating under PRC capital-markets constraints without a comparable enterprise API deployment surface. Trump's executive order reversal on AI oversight, reported by TechCrunch on Thursday, clears the last domestic friction before Anthropic's S-1 registration reaches the SEC Division of Corporation Finance.