Moonshot reportedly has a computing agreement with Alibaba (BABA) for the use of around 20k Nvidia (NVDA) chips

Compute-access agreements of this kind have become the standard structure in Chinese AI since export controls tightened: rather than buying chips outright, model developers rent capacity from cloud providers whose inventories predate or sit within the licensing rules, and Alibaba's cloud arm has been the recurring landlord in these arrangements. The precedent pattern is that the chipmaker's name in the headline does the market lifting while the actual revenue flows to the cloud operator, since the silicon is already sold and sitting in the lessor's data centres; NVDA sympathy moves on such reports have historically faded once that distinction is absorbed. The open question in episodes of this kind is provenance: whether the chips are legacy compliant inventory, domestically permitted variants, or something that draws regulatory attention, and prior cases have occasionally produced follow-on scrutiny from Washington rather than from Beijing. Moonshot is among the better-capitalised Chinese model developers, and commitments at this scale signal continued training ambition despite the compute ceiling, which fits the broader pattern of Chinese labs clustering around whichever domestic cloud capacity is available. Worth watching is any confirmation of the chip variant involved, whether the agreement is capacity rental or a deeper strategic tie, and any US official response, since prior large-count China compute headlines have tended to resurface in export-control debate.