Hyperscalers Seal the Infrastructure Monopoly

Meta's GPU cloud entry signals hyperscaler domination is complete—neocluster providers built on commodity pricing and debt face defection and refinancing risk as buyers lock captive infrastructure.

Meta Compute's announcement crashed CoreWeave 14%, Nebius 17%, IREN 6.5%, while Meta jumped 9%. The market read this correctly: the neocloud age is over. Hyperscalers control both AI model demand (the buyer) and infrastructure supply (the asset). CoreWeave and Nebius compete on price; Meta competes on integration. Price loses.

The data confirms it. Berkshire backed Google's capex with $80B; Meta committed $59B to Louisiana and Canada; Anthropic locked TeraWulf for $19B over 20 years. These aren't spot purchases—they're buyers securing supply or third parties guaranteeing it. CoreWeave's $24.9B debt (co-financed by Nvidia customers) only makes sense if generalist GPU cloud demand accelerates forever. SK Hynix just destroyed that assumption: the CEO warned HBM shortages last until 2030, forcing AI builders into vertical integration, not rental optionality.

TeraWulf survives because it's not competing on commodity pricing. Its Anthropic deal is a long-term utility contract—fixed revenue, off-balance-sheet capex for the buyer, neocloud as a service provider to a single captive buyer. That model scales if replicated with other large models. CoreWeave and Nebius don't have that lock-in. They face acquisition, specialist pivots (inference-only, safety testing), or debt refinancing stress.

The neocloud sector's bet was that hyperscalers would buy GPU cloud rather than build it. That bet is lost. Memory supply constraints through 2030 only accelerate the consolidation—AI builders optimize for in-house infrastructure, not flexibility. Watch CoreWeave/Nebius for debt stress and distressed M&A; watch whether SK Hynix's supply squeeze becomes the shock that forces neocloud refinancing cycles into crisis.