February 07, 2026

Global Data Centres Briefing (UTC 2026-02-07)

Audience: Institutional asset managers and infrastructure fund managers (data centres, power, grid)

Top news (3)

1. New utility tariff structures proposed to accelerate clean power for data centres

• An RMI insight brief recommends Bring-Your-Own (BYO) and Clean Transition Tariffs to speed up clean energy procurement for large loads (including data centres) while insulating other ratepayers. Examples cited include PJM’s Bring Your Own New Generation Program (>250 MW), Evergy Kansas’s Clean Energy Choice Rider, NV Energy’s Clean Transition Tariff (used by Google), and Georgia Power’s Customer-Identified Resource option: Bring-Your-Own and Clean Transition Tariffs for Data Centers.

2. Google pushes for faster post‑quantum cryptography (PQC) transition across critical infrastructure

• Google says it has been preparing since 2016, is rolling out PQC across its infrastructure, and expects to complete migration within NIST’s guidelines. It also issued five recommendations for policymakers, including making AI systems PQC‑ready and prioritizing cloud‑first modernization: Google urges policymakers to prepare for quantum-era cryptography transition.

3. Nordics positioned as a renewable-powered, AI‑ready data centre hub (with heat reuse highlighted)

• A report frames the Nordic region as a model for sustainably scaling AI‑ready capacity, citing renewable energy, efficient power markets, and cool climate, plus Denmark examples such as atNorth using district heat reuse: Nordic region powers AI data centers with renewable energy.

Key deals and projects

Nordics (regional theme rather than a single transaction)

• The Nordic region is positioned as supportive for AI data centres due to renewable supply, power-market efficiency, and cool climate, with district heat reuse cited in Denmark (atNorth example): Nordic region powers AI data centers with renewable energy.

Note: No M&A, financing rounds, lease transactions, or capex figures were included in today’s story set.

Power and grid / interconnection highlights

US (tariff and utility program structures)

• RMI’s brief focuses on procurement mechanisms designed for large loads:
  • BYO structures (example: PJM Bring Your Own New Generation Program noted as >250 MW) to connect new supply with large customers.
  • Clean Transition Tariffs (example: NV Energy’s Clean Transition Tariff used by Google) aimed at enabling clean procurement while limiting impacts to other ratepayers.
  • Additional examples: Evergy Kansas Clean Energy Choice Rider and Georgia Power Customer‑Identified Resource option.
  • Implication for investors: these programs can influence time-to-power, contracting pathways, and who bears system costs for incremental generation tied to new data centre demand: Bring-Your-Own and Clean Transition Tariffs for Data Centers.

Grid-support concept from compute operations (research)

• A research proposal suggests data centre GPU loads could be modulated for distribution voltage regulation, by adjusting batch size in LLM inference to change GPU power draw (reducing GPU power to alleviate lower-voltage violations; increasing power to mitigate upper-voltage violations): GPU-to-Grid: Distribution Voltage Regulation via GPU Utilization Control.

Policy and regulation

Cybersecurity / national infrastructure readiness (global)

• Google’s PQC update is framed as both an implementation program and a policy agenda:
  • Migration underway across Google infrastructure; target completion aligned with NIST guidelines.
  • Policy recommendations include making AI systems PQC-ready and prioritizing cloud-first modernization.
  • Investor relevance: PQC timelines can drive security-related capex/opex, procurement cycles, and compliance expectations for critical digital infrastructure: Google urges policymakers to prepare for quantum-era cryptography transition.

Compute capacity, supply constraints, and operational efficiency (market signals)

GPU supply constraints and “neocloud” positioning

• A vendor article argues GPU shortages are hindering university research and proposes neocloud GPU rentals as a mitigation, offering hourly access to enterprise-grade NVIDIA GPUs via pre-configured VMs and a stated 15% coupon code; it also notes shortages can reshape research priorities and widen inequality across institutions: GPU shortages hinder university research, neoclouds as solution.

Efficiency and performance research (potential longer-term impacts on power density and cost-to-serve)

• LLM inference energy-efficiency “sweet spots”: an analytical model validated on NVIDIA H100 using TensorRT‑LLM, across 1B–9B parameter models and 64–4096 token sequences; reported mean MAPE 1.79%, recommending alignment of sequence lengths to reduce production energy usage: Energy Efficiency Sweet Spots in Production LLM Inference.
• Serving throughput improvement: TIDE proposes integrating online draft adaptation into inference, reporting up to 1.15x throughput improvement vs static speculative decoding and 1.67x reduction in draft training time: TIDE: Temporal Incremental Draft Engine for Self-Improving LLMs.
• Virtualization and latency dataset: a published dataset varying CPU affinity, packet injection frequency, virtual network driver, load types, and VM counts to study end-to-end latency; positioned to support ML-based admin decision-making: Data analysis of cloud virtualization experiments and network latency.
• NUMA migration technique (user-space): proposes page_leap() for fine-grained NUMA page migration (small and huge pages), as an alternative to Linux automatic NUMA balancing and move_pages(): Efficient and Reliable Fine-Grained NUMA Migration in User-space.

Data-intensive science workload signal (storage and workflow scale)

• A synchrotron facility (CHESS) deployed a data acquisition and analytics framework on a secure server with real-time web tools, implemented across beamlines ID3A/ID3B/ID4B, managing 50–100 TB and 10+ million files, tested with 43 research groups and 86 dashboards: Large Data Acquisition and Analytics at Synchrotron Radiation Facilities.

2-line close

Tariff designs that pair new generation with large loads are being positioned as a practical lever to speed clean power procurement for data centres without shifting costs to other customers.
In parallel, operators are being pushed by both security transition requirements (PQC) and GPU-driven scaling constraints, with efficiency research pointing to incremental gains in power and performance over time.