Version 5

What

Why it matters

Open questions

• What standards must New York establish before the moratorium lifts — will they require clean energy sourcing, water use limits, or grid impact assessments? [2]

• Will the NVIDIA efficiency argument hold at the grid level? The Blackwell B200 draws up to 1200W per chip vs. 700W for H100 [10][11], leaving open whether per-token efficiency reduces total power draw or enables more compute within the same energy budget.

• How far can S.4214 and H.R.9442 advance given that the Trump administration and congressional Republicans oppose such restrictions? [5][6][4]

• Will the $12.6 billion energy IPO surge of H1 2026 translate into supply that eases the grid bottleneck before the EIA's projected record power demand materializes in 2026–2027? [13][12]

Narrative

Timeline

• 2025: New York State Legislature passes Senate Bill S10642, sponsored by Senator Kristen Gonzalez, establishing the framework for a data center moratorium. [19][20][3]
• 2026-03-25: Roll Call covers the federal data center moratorium push, framing it as a response to AI infrastructure's grid and environmental impacts. [21]
• 2026-06-24: Rep. Frank Pallone calls for a national data center moratorium, per Politico. [7]
• 2026-06-25: A data center energy bill in Congress hits a procedural obstacle, per Politico. [8]
• 2026-06-29: City Journal publishes 'New York's War on Compute,' criticizing anticipated data center restrictions as economically harmful. [17]
• 2026-07-07: Reuters reports the EIA projects US power use will beat record highs in both 2026 and 2027 as AI demand surges. [12]
• 2026-07-14: Governor Hochul signs Executive Order No. 62, imposing a one-year moratorium on data centers consuming 50MW or more — the first statewide construction ban in the US. [1][2][4][15]
• 2026-07-14: Sanders introduces S.4214 and Ocasio-Cortez introduces H.R.9442, together the Artificial Intelligence Data Center Moratorium Act, seeking a possible nationwide ban. [5][6][16]
• 2026-07-14: NVIDIA publishes the case for performance per watt as the primary AI infrastructure metric, citing 25x improvement for GB300 NVL72 over Hopper on DeepSeek V4 Pro. [9]
• 2026-07-16: Ars Technica reports energy company IPOs raised $12.6 billion in H1 2026 — the highest first-half figure on record — as investors price in energy access as the central AI infrastructure bottleneck. [13]
• 2026-07-17: Simon Willison publishes back-of-the-envelope analysis showing Google consumed 10.9 billion gallons of water in 2025, adding water use to the environmental footprint debate around AI data centers. [14]

Perspectives

Tensions

• New York and federal progressive legislators argue data center growth requires a regulatory pause for standards-setting; the AI industry, conservative policy groups, and Republican legislators argue such restrictions damage US AI competitiveness. [4][17][18][1]
• NVIDIA argues that 25x performance-per-watt gains for Blackwell vs. Hopper make moratoriums unnecessary; but the Blackwell B200 draws up to 1200W vs. 700W for H100, leaving open whether per-token efficiency reduces total grid draw or simply enables more compute. [9][10][11]
• Regulators are restricting new data center construction at the state and potentially federal level, while energy sector investors are committing record capital — $12.6 billion in H1 2026 IPOs — premised on continued AI infrastructure growth. [1][13]
• Sanders, Ocasio-Cortez, and Pallone have introduced or called for a federal moratorium; the Trump administration is unlikely to sign such legislation, leaving state-level action as the operative constraint while federal bills remain stalled. [5][6][7][4]
• The EIA projects record US power consumption in 2026–2027 driven by AI, directly contradicting the industry efficiency argument that hardware gains will reduce total grid draw. [12][9]