Every B300 Earns $8/Hour: Why the Grid Queue Gives Communities Massive Leverage Over AI Data Centers

Exponential View (Azeem Azhar) · rss · 2026-11-06

Azeem Azhar argues the US data center backlash — voters passed 43 of 44 local measures to slow data centers — is a lever for communities to strike deals, not the end of the AI build-out.

Time is power, not energy prices: Per Silicon Data, each plugged-in B300 generates $8/hour and pays for itself in under a year. A 1GW data center holds 400,000 B300s at $40bn total cost; annual opex is under 2.5% of CapEx ($800m, two-thirds electricity). Shells take 18-24 months to build, but the average US grid connection wait is four years — until electrons flow, no cash flows, which is extraordinary leverage.

The Lanarkshire template: DataVita agreed to pay up to $100k per MW per year into a local community fund — $50m/year at 500MW, a 3% boost to the council's income, up to £543m over 15 years. By contrast, Oracle's New Mexico package worked out to $84k per MW as a one-off, nearly 18 times stingier.

The jobs illusion: Meta's $50bn Louisiana investment yields 1,000 permanent jobs (20 per $1bn); Anthropic's $50bn builds bring 800. Locally negotiated levies can create annuity streams instead. At the Lanarkshire benchmark, every GW of connection could yield $100m — just 0.2-1% of what a gigawatt rents for at today's surge prices.

Original post →

More from Infra

Infra channel →