What happened
Apollo and Blackstone are financing a reported $35 billion expansion of AI compute capacity linked to Anthropic, using Broadcom custom chips and networking and Fluidstack-managed data-center sites. The first phase is expected to add more than one gigawatt of capacity, with longer-term plans for much larger capacity by 2028. Reporting also notes Anthropic access to Google-supported processing capacity beginning in 2027.
Potential direction of power
Concentration · Concentration +2
The event strengthens the pattern that frontier AI compute is being built through highly capitalized alliances among model firms, chip/network suppliers, data-center operators, cloud-adjacent capacity providers, and large private-finance actors. The scale of capital and infrastructure required creates a high barrier to entry.
Mixed but concentration-leaning. Broadcom custom silicon may diversify the chip layer away from Nvidia dependency, but the overall structure bundles model demand, compute, private finance, chip supply, and data-center capacity into an infrastructure channel available only to very large actors.
Window of contestability
Narrows
