US Data Center Briefing · January 07, 2026

January 07, 2026

Entergy Louisiana >$1bn 500-kV transmission proposal tied to 345MW data centre Hyperscaler capex projected >$600bn in 2026 amid >$162bn blocked/delayed projects AWS increases guaranteed GPU capacity pricing ~15% Behind-the-meter/fast-start power solutions (jet engines, LNG microplants) for AI loads High-density liquid cooling moving to turnkey reference designs (up to 142kW racks)

Top news (3)

Louisiana grid buildout tied to large-load growth: Entergy’s proposed billion-dollar transmission upgrades include a >$1bn, 145-mile 500-kV line to serve rising demand from data centres and heavy industry, linked to a planned $10bn, 345MW data centre in West Feliciana Parish and Hyundai’s $6bn steel mill. Entergy expects ~82% of the $1bn project cost to be charged to ratepayers and will seek approvals at the Louisiana Public Service Commission in 2026.
Hyperscaler capex acceleration + growing project friction: A sector roundup says hyperscaler capex is projected to exceed $600bn in 2026, while >$162bn of projects were blocked or delayed as of June 2025 (hyperscale expansion in 2026). Named commitments include Oracle “Stargate I” (1.2GW), AWS $5.3bn Saudi region and €7.8bn Germany sovereign cloud, plus multiple large AI campus initiatives.
GPU scarcity showing up in cloud pricing: AWS raising EC2 Capacity Blocks for ML pricing ~15% (example: p5e.48xlarge in Ohio from $34.608 to $39.799 per hour) signals tighter economics for guaranteed GPU capacity.

Key deals and projects (data centres, platforms, and supply chain)

North America

Fremont, California (AI/HPC colocation / hosting): AMAX expands its Fremont facility with a Phase A upgrade adding 2MW to support liquid-cooled AI racks at 150kW class (optionally up to 300kW) under its HostMax™ service. It references support for NVIDIA HGX and AMD MI355X platforms, with ASHRAE-compliant liquid and air cooling and a PUE-aware power/cooling architecture.
Bitcoin/HPC developer organisational buildout: Cipher Mining leadership hires add a Head of Policy and Government Affairs (Lee Bratcher; Texas energy regulatory and ERCOT experience) and Head of Strategic Initiatives (Drew Armstrong; infrastructure background). While not a capex announcement, it signals continued data centre / HPC expansion intent.

Global (platforms, racks, and enterprise AI stacks)

NVIDIA rack-scale platform launch: Nvidia unveils Vera Rubin NVL72 — a rack platform described as a six-silicon system (Vera CPU, Rubin GPU, NVLink 6, ConnectX-9, BlueField-4, Spectrum-6). NVIDIA states performance up to 3.6 exaflops NVFP4 inference and 2.5 exaflops NVFP4 training, with a configuration of 72 Rubin GPUs / 36 Vera CPUs and expanded memory (54TB LPDDR5x, 20.7TB HBM4). Named partner support includes AWS, Google, Microsoft, Meta, OpenAI.
AMD rack-scale roadmap: AMD unveils Instinct MI500 GPUs and Helios positions Helios at 3 AI exaflops per rack and flags MI500 (2027) as next-gen. A separate AMD report highlights major partnerships and supercomputing commitments, including a U.S. DOE agreement for two more supercomputers at Oak Ridge (Lux AI and Discovery) described as a combined $1bn investment (AMD’s 2025 push).
Enterprise AI software/hardware integration: Nutanix expands collaboration with Nvidia on an integrated AI operating environment (Nutanix Acropolis OS and AHV; supports Nvidia AI Enterprise, NIM microservices, BlueField-4 and Spectrum-X integrations).
Inference storage architecture aligned to Nvidia networking/DPUs: VAST Data’s AI inference storage architecture ties the VAST AI Operating System to NVIDIA BlueField-4 DPUs, Spectrum-X Ethernet, and RDMA NVMe paths; it will be showcased 24–26 Feb 2026 in Salt Lake City.

Power and grid / interconnection highlights

United States

Louisiana (bulk transmission for large loads): Entergy’s >$1bn 500-kV line proposal (145 miles) is explicitly linked to supporting a planned 345MW data centre investment and other industrial demand. Cost allocation mentioned: ~82% expected to be ratepayer-charged.
South Carolina (gas + pipeline to support demand): Dominion Energy and Santee Cooper’s 2,000MW gas plant is cited alongside a contracted Kinder Morgan $431m, 71-mile pipeline “Bridge Project” via Elba Express Company, targeting FERC filing in early 2027 and construction aimed for early 2029.
Behind-the-meter/fast-start power trend: US data centres turning to unconventional power sources describes developers deploying repurposed jet engines, aeroderivative turbines, LNG microplants, and imported power plants to secure fast-start capacity for AI loads. The story cites IEA estimates that US data centres used 183 TWh in 2024 and could reach 426 TWh by 2030.

Middle East

Saudi Arabia (grid readiness for renewables): Linxon’s turnkey high-voltage substation contracts cover multiple 380/110/13.8 kV substations, delivered as full lifecycle turnkey projects, positioned as supporting large-scale solar and grid readiness for variable renewables.

Asia

Japan (thermal generation coming online): Osaka Gas’ Himeji 1.25GW plant started commercial operation of No.1 unit, with a second 622.6MW unit due in May, lifting domestic thermal capacity from ~2GW to ~3.2GW.
India (green power commitment for an AI hub): Uttar Pradesh’s proposed Lucknow AI City is announced to be powered entirely by green energy, using solar systems and green hydrogen, with green building practices and data centres designed to meet global environmental and energy-efficiency standards.

Policy and regulation watch

Louisiana utility approvals: Entergy to seek LPSC approval in 2026 for major transmission additions; the story’s explicit note that ~82% of costs may be charged to ratepayers is likely to be a focal point in proceedings.
EU power demand and carbon cost implications: EU data centre electricity use forecast to double by 2029, a change expected to increase demand for carbon permits.
State-led “green data centre” positioning: Lucknow AI City’s green-energy requirement points to policy-led sustainability expectations (solar + green hydrogen; green building mandates).

Operational resilience and efficiency signals (select)

Cooling as deployment bottleneck and outage driver: Schneider Electric liquid cooling reference designs RD110/RD111, co-developed with NVIDIA, target AI racks up to 142kW; the piece highlights cooling failures as a significant source of costly outages and cites a market forecast from ~$2.8bn to >$21bn by 2032.
Global network reliability: ThousandEyes’ weekly outage summary reported 199 global outage events (Dec 29–Jan 4), including a 1-hour Hurricane Electric outage (Jan 2) impacting multiple APAC countries and a 9-minute Cogent outage (Dec 31) affecting multiple regions.

Two-line close

Transmission and fast-start generation are being pulled directly into the data-centre growth story as utilities and developers respond to large-load concentration.

At the same time, hyperscaler capex momentum is strong, but supply constraints (GPUs, cooling, and grid capacity) are increasingly visible in pricing and project delays.