BRIGHT EVIDENCE PACK / Emerging
AI data centers and the grid bottleneck: what real progress would look like
A new Goldman Sachs forecast highlights the scale of the buildout. The more useful test is whether new computing demand brings reliable electricity, workable connections and fair terms for the people sharing the grid.
Canonical Bright record · JSON evidence pack · Key-facts embed
Dates and assessment
- Source published
- 2026-10-20
- Bright published
- 2026-10-20
- Substantive update
- None recorded
- Evidence state
- Emerging
- Independent verification
- Not established by this source review
- Last source review
- 2026-10-20
The claim in context
The human problem
Electricity connections, equipment, project delivery and cost allocation shape whether AI infrastructure can serve people reliably.
The prior constraint
Proposed generation, requested consumer connections and actual deliverable electricity are different quantities with different timelines.
AI’s actual role
AI computing creates electricity demand while efficiency per useful task and flexible operation can influence the infrastructure needed.
The documented result
Goldman Sachs’s October 19 analysis forecasts global data-center power demand increasing 170% from 2025 to 2030. The IEA separately projects 485 TWh in 2025 rising to about 950 TWh in 2030. Berkeley Lab counts 1,312 GW of generation and approximately 749 GW of storage seeking transmission connections at end-2025.
Why it may matter
Credible demand plans, enforceable customer obligations, verified flexibility and completed projects could help turn computing growth into dependable infrastructure with clearer local costs.
Limitations
- Forecasts are source-attributed projections, not measured AI consumption or guaranteed construction.
- Gigawatts of capacity and terawatt-hours of electricity are different measures.
- Berkeley Lab’s generation and storage queue excludes consumer loads, distribution-connected and behind-the-meter projects; applications are not future supply guarantees.
- Historical queue attrition and elapsed connection-to-operation times do not establish the outcome or schedule of an individual project.
- A protective tariff or payment for a substation does not establish that all wider costs are covered or that residential bills have fallen.
- Efficiency per task can improve while total consumption grows; future usage and workloads remain uncertain.
Original evidence
- analysis published on 19 October · institution
- electricity-measurement guide · institution
- April 2026 outlook · institution
- Berkeley Lab's 2026 queue research · institution
- Powering Artificial Intelligence, published on 14 October · institution
- executive summary · institution
- AEP Ohio's data-center process · institution
- guide to who pays data-center grid costs · institution
- Ohio's tariff · institution
- PJM co-location cost rules · institution
Attribution
Credit Bright AI Future and link the canonical Bright record.
- Link to the canonical Bright record.
- Keep material limitations with the claim they qualify.
- Link to the original evidence when repeating a substantive claim.
- Do not describe a source check or organization-reported result as independent verification.
Linked source material, quotations, trademarks and media remain subject to their owners’ terms. No reuse right is granted for third-party media.