Moonshot's $35B Valuation and CXMT's $464B Market Cap Put the Chinese AI Stack in Hong Kong
Moonshot AI closed a $3.5 billion round at a $35 billion valuation, having initially targeted somewhere between $1 billion and $2 billion, and plans a Hong Kong listing as early as this year. CXMT’s IPO prospectus disclosed that Alibaba holds a stake of nearly 5%, now worth more than $20 billion, against a closing market capitalization of roughly $464 billion on July 28. One model developer and one DRAM manufacturer, heading for the same venue, at valuations set almost entirely by domestic capital.
The Alibaba number is the one that reframes the last decade. A position taken years ago in a domestic memory startup has returned roughly twenty times and is now worth north of $20 billion, which makes it one of the best semiconductor venture outcomes anywhere in the world over that period. It was not available to Western investors and it was not priced by them. Whatever else the export control regime accomplished, it produced a domestic champion whose equity value accrued entirely inside the system it was designed to constrain.
CXMT at roughly $464 billion is the number with consequences beyond the listing. A DRAM manufacturer with that market capitalization can raise capacity capital at a cost no Korean or American incumbent can match, and it does not require Western customers to justify a fab. The Western memory bull case rests on supply discipline: three participants who have learned across multiple cycles not to build into a peak, reinforced this cycle by high-bandwidth memory consuming die area that would otherwise have become commodity supply. A fourth participant with near-free equity capital and a domestic demand mandate is not bound by that discipline, and its capex plan is a matter of public record now that the prospectus exists. That document is more important to 2027 DRAM pricing than any guidance a listed memory maker will give.
Moonshot’s mark is a different kind of information. Raising $3.5 billion against a $1 billion to $2 billion target is not a fundraise, it is an auction, and a $35 billion valuation for a model developer competing directly against both domestic state-adjacent efforts and the frontier labs is a price that assumes the open-weights strategy converts into a durable commercial position. The competitive logic is at least coherent: releasing weights builds the developer base that a hosted inference business monetizes, and it does so in a market where the leading Western models are difficult to procure. Whether that is a moat or a subsidy depends on who is paying for the training runs, and the round size suggests the answer is investors rather than customers.
For Hong Kong, this is the material development and it is not primarily about either company. A venue that spent several years being written off as a listings market is about to price both a leading domestic model developer and a top-twenty global semiconductor manufacturer by market capitalization. Capital market decoupling was supposed to starve these companies of public equity. Instead it relocated the pricing function, and the valuations coming out of the relocated venue are higher than the ones Western comparables carry.
The practical consequence for anyone holding memory equities is that the supply side of the thesis now has a listed, well-capitalized participant whose disclosure obligations will make its expansion plans visible before the pricing impact arrives. Read the capex section of the CXMT prospectus. It is the single most consequential document for DRAM pricing published this quarter.