Anthropic’s IPO data has surfaced: it plans to raise $100B, targeting a valuation of $2T—both exceeding SpaceX’s ceilings. But this isn’t just a fundraising headline; it’s a pricing signal for the industry chain. While it tells investors a $30T market story, it’s also looking for Matx to develop its own chips. The underlying economics force it to break free from Nvidia’s pricing power: Meta internally once estimated that investing in Anthropic models could reach as high as $10B per year, and the scale of training costs has become unsustainable. The clear beneficiaries are: TSMC’s advanced process technology; SK hynix’s $4B packaging plant in Indiana; and Kioxia/SanDisk’s $31B storage expansion in Japan—all positioning themselves upstream in this arms race. The ones harmed are Nvidia’s forward pricing power and all intermediate-layer inference service providers. NVDA’s earnings report beating expectations is a fact, but the market’s $209 valuation already prices in the probability of “self-developed chip substitution.” The prospectus by the end of September will be the first validation point: watch how it discloses its self-developed chip mass-production timeline. If it clearly targets tape-out in 2027, Nvidia’s forward-valuation logic will need to be rebuilt. Keep an eye on the next funding move by Meta and OpenAI.