Nvidia Is Weighing a $10 Billion Stake in Anthropic's IPO. It Would Be Buying Its Own Demand. - The Globe and Mail

Nvidia's $10B anchor stake in Anthropic's IPO is a supply-chain fortification, not a financial allocation. The real threat isn't the investment—it's Anthropic's three simultaneous decoupling vectors: 5GW of Amazon Trainium, 5GW of Google TPU, and an in-house ASIC team. Nvidia's 0.5% equity slice buys 2-3 years of procurement inertia, securing Vera Rubin's inference architecture a penetration window before alternatives mature. On compliance, a vendor anchoring its customer's IPO sits in a gray zone under US antitrust law—vertical coordination without explicit price-fixing, echoing the structural risks seen in Qualcomm's 2010s patent-licensing disputes. The deliberate post-midterm IPO timing signals that bipartisan congressional pressure on AI compute concentration is now a real regulatory variable. Strategically, AMD's most probable counter isn't matching the investment but accelerating a de-Nvidia inference narrative; Google and Amazon directly capture the incremental demand Anthropic's diversification creates. More structurally, Nvidia's equity book ballooning from $3.8B to $47.9B in twelve months signals a balance-sheet drift from chipmaker to AI industrial fund, rewriting the sector's competitive coordinates. Within 18 months, chipmaker-invests-in-model-lab will become the standard AI infrastructure financing structure. But if Anthropic ships a first-gen inference ASIC by 2027, Nvidia's moat degrades from architectural lead to ecosystem inertia, and this $10B anchor becomes the last stake in a shrinking addressable market.