AI Capex Narrative Widens Beyond Chips as AMD Discloses $566 Million SpaceX Stake

AMD's $566 million stake in SpaceX is the most consequential disclosure in the batch, not for its size but for what it signals about the architecture of AI-era commercial relationships. Equity participation in a strategic customer blurs the line between revenue recognition and investment return, a structure that has become increasingly common across the accelerator complex and one that investors will need to scrutinize for circularity risk. The immediate read-through is that AMD is buying optionality on compute demand from sources outside the hyperscaler oligopoly. On the memory side, a five-star-rated analyst has published an aggressive 2026 target on Micron, reflecting the view that high-bandwidth memory pricing power is structural rather than cyclical — a call that implicitly assumes accelerator unit growth persists through the next capex digestion phase. The Broadcom-versus-Cisco framing addresses the networking layer: Broadcom offers custom silicon and switching ASIC leverage to hyperscaler build-outs, while Cisco's exposure is more diluted across enterprise campus and legacy refresh cycles. Underpinning all three is Anthropic's projected trajectory to $200 billion of revenue by 2028 as it approaches a public listing. That figure, if remotely achievable, validates the compute intensity assumptions embedded in current semiconductor multiples. If it proves aspirational, the entire AI supply chain re-rates downward from the demand side rather than the supply side — the more dangerous of the two failure modes.