January 06, 2026

Top news (3)

• Big-ticket power/compute convergence M&A:Alphabet to buy Intersect Power for US$4.75 billion (cash; expected close 1H 2026). Intersect stays standalone under CEO Sheldon Kimber and will collaborate with Google’s technical infrastructure team on co-located data centre and power projects; certain Texas and California assets are excluded.
• Dispatchable power build to serve data centres in the UAE:TAQA, EWEC close AED 3.6bn Al Dhafra 1GW plant. Financial close reached on a 1 GW gas turbine plant (capex AED 3.6bn) to supply dispatchable power to data centres; structured at ~85% debt financing; backed by a 24-year PPA (signed April).
• US permitting/political pressure rising around hyperscale growth:Virginia urged to pause hyperscale data center approvals. The Piedmont Environmental Council asked the General Assembly for a pause pending full accounting and a transparent plan for grid, water, air, and community impacts; notes a warning to FERC about reliability/affordability risks and cites the state’s data centre sales tax exemption costing $1.6bn in FY2025.

Key deals and corporate actions

North America

• Cooling services consolidation:Vertiv completes $1 billion acquisition of PurgeRite to boost cooling. Vertiv completed a $1bn acquisition to expand thermal management and liquid cooling services for AI/high-density deployments; the story cites liquid cooling market forecasts of $17.7bn by 2030 (Grand View Research) and a $15bn opportunity by 2028 (Dell’Oro Group).

Asia

• India (portfolio recycling / fund alignment):CapitaLand India Trust divests minority stakes in three data centres. CLINT agreed to sell 20.2% stakes in three under-development data centre assets to the CapitaLand India Data Centre Fund for ~Rs 7bn (S$99.7m); the fund also receives a right of first offer on CapitaLand DC Bangalore (Bengaluru).
• Singapore (vendor capex into AI-DC enablement):Johnson Controls to invest S$60m expanding Singapore data centre hub. Johnson Controls plans up to S$60m over five years to expand its Singapore Innovation Centre, lift engineering headcount to 90–100, and focus on advanced cooling and automation for AI-driven data centres; includes prototype development/testing with universities/industry partners and deployment of OpenBlue and Cooling-as-a-Service across APAC.

Americas (people / platform execution)

• Ops/services leadership:Salute appoints Robert Heikaus as MD Americas for data centers. Appointment of Robert Heikaus as Managing Director, Americas (mission-critical background including CBRE GWS and IBM).

Power and grid / interconnection highlights

Middle East (UAE)

• Firming capacity explicitly tied to data centres:TAQA, EWEC close AED 3.6bn Al Dhafra 1GW plant
  • Size/capex:1 GW gas turbine; AED 3.6bn.
  • Contracting/tenor:24-year PPA (signed April).
  • Financing: ~85% debt-financed via a consortium of local/international banks.
  • Strategic frame: positioned to supply dispatchable power for data centres and support the UAE National Strategy for Artificial Intelligence 2031; described as part of a broader AED 36bn investment plan with Masdar (renewables + storage expansion).

India

• Grid stability toolkit for renewables + storage (relevance for data-centre-adjacent BESS and grid upgrades):India examines grid-forming inverters to bolster renewable integration
  • Grid-India discussion paper recommends phased adoption of grid-forming (GFM) inverter capability as India targets 500 GW non-fossil capacity.
  • Simulations (Rajasthan focus) indicate GFM outperforms grid-following inverters during disturbances.
  • Recommendation: BESS above 50 MW to include GFM capability, plus grid code updates and pilot projects.

United States

• Reliability and affordability flagged amid hyperscale concentration:Virginia urged to pause hyperscale data center approvals references a warning to FERC about reliability/affordability risks.

Policy / regulation / permitting

United States (state and city-level scrutiny)

• Virginia (approvals + incentives under review):Virginia urged to pause hyperscale data center approvals
  • Call for a pause on approvals and a transparent plan covering grid, water, air, and community impacts.
  • Incentive magnitude highlighted: data centre sales tax exemption cost $1.6bn in FY2025.
• Detroit, Michigan (health-linked permit conditions):Detroit must require health‑protective data center approvals
  • Advocates for health impact assessments, 24/7 clean energy procurement tied to approvals, closed-loop cooling, strict diesel testing, and fence-line monitoring as permit requirements.
• Maine (local transparency / siting pushback):Maine communities urged to scrutinize AI data center proposals
  • GrowSmart Maine released a community guide and urged towns to demand transparency; notes protests in Lewiston regarding a proposed AI data centre at the Bates Mill complex ahead of a Dec. 16, 2025 city council vote.

Sustainability framing (operational and compliance risk)

• Energy + e-waste as constraints on growth:Data center growth raises energy and e‑waste sustainability concerns
  • Cites US expansion with 1,240 data centres built/approved by end-2024 and notes Virginia data centres used ~26% of electricity in 2023.
  • Highlights global e-waste volumes (62m tons in 2022) and claims generative AI could add 1.2–5m tons annually.
  • Notes Illinois bills H.B. 3758 / S.B. 2497 targeting 15 GW of energy storage and virtual power plant programs.

AI infrastructure, networking, and data-centre hardware implications

NVIDIA/Dell platform updates (density, networking, and inference optimization)

• Next-gen AI racks and high-speed fabrics announced by a major OEM:Dell and NVIDIA unveil next-gen AI infrastructure and networking
  • PowerEdge servers integrated with NVIDIA Vera Rubin NVL72 (claimed 3.6 exaflops and 75 TB memory).
  • Support for HGX Rubin NVL8 (claimed ~400 petaflops), with BlueField-4 DPUs and 800 Gb/s ConnectX-9.
  • Networking: PowerSwitch with NVIDIA Spectrum-6 (claimed 102.4 Tb/s switching; up to 512×800G CPO ports).
• Inference performance lever (KV cache offload):Dell and NVIDIA advance KV Cache with BlueField-4 CMS
  • KV Cache offloading for LLM inference combining NVIDIA BlueField-4 with Dell storage (PowerScale, ObjectScale, Project Lightning).
  • Dell claims up to 19× improvement in Time to First Token and up to 5.3× higher queries per second using its KV cache offload solutions.
• Scaled reference design direction:NVIDIA launches Rubin platform and DGX SuperPOD scale-out
  • Introduced Rubin platform (codesigned stack of six chips) and Rubin-based DGX SuperPOD configs (DGX Vera Rubin NVL72 and DGX Rubin NVL8).
  • Claimed performance up to 28.8 exaflops FP4; availability indicated for 2H this year.
• Security/acceleration integrated into validated stacks:NVIDIA integrates BlueField cybersecurity into Enterprise AI Factory
  • Enterprise AI Factory validated design expanded to include NVIDIA BlueField cybersecurity and infrastructure acceleration; validated partners include Armis, Check Point, F5, Fortinet, Palo Alto Networks, Rafay, Red Hat, Spectro Cloud, Trend Micro.

Memory and component roadmap

• New memory form factor positioned for AI data centres:Samsung unveils SOCAMM2 LPDDR5 memory for AI data centers
  • Samsung introduced SOCAMM2 (LPDDR5-based CAMM2 module), claiming up to 2× DDR5 bandwidth with materially lower power.
  • Notes JEDEC standardized CAMM2 features including ECC; SK Hynix pledged support; expected launch around Q2 2026 alongside Nvidia’s Vera Rubin platform.

2-line close

Compute, cooling, and power stacks are being designed and financed together—vendor roadmaps and utility-grade generation are moving in lockstep.
Local permitting, incentive scrutiny, and grid-stability requirements are increasingly the gating items for hyperscale delivery timelines.