Infrastructure & Hardware

DCD Intelligence published its 2026 Data Center Workforce Survey, reporting that data center growth is outpacing the supply of talent. The findings point to skilled staffing for operations and maintenance as a constraint even as capital spending accelerates.

Serverfarm has expanded its North American data center development fund to $3.89 billion after closing an additional $895 million. The raise adds to the steady accumulation of capital earmarked for North American data center construction.

Microsoft has filed to build a data center campus in Prince William County, Virginia, proposing two buildings outside Gainesville. The filing extends the company's buildout in Northern Virginia, already the densest data center cluster in the US.

AWS said it is unable to restore access to data centers damaged by Iranian strikes, with the damage exceeding what its availability zone design was built to withstand. The account underscores the limits of redundancy assumptions under armed conflict.

Hokkaido published its stance on data center siting on September 15, asking operators to adopt decarbonized, energy-efficient technology and to coexist with host communities. The prefecture is promoting data centers for AI and chips, but flags power procurement strain and resident concerns over waste heat, noise and landscape.

OpenAI policy chief Chris Lehane said the company has spent weeks in AI safety talks with rivals Anthropic and Google DeepMind, arguing no antitrust exemption is needed. FTC chair Andrew Ferguson said he would treat such requests with deep suspicion, as rival bills on superintelligence and narrow exemptions surface in Congress.

Fujitsu said on Sept 14 it will sell FUJITSU-MONAKA, a domestically designed AI CPU, and the Fujitsu MONAKA Server from November, targeting data center operators and financial customers first. Derived from the A64FX Fugaku supercomputer chip, it anchors a 3 trillion yen AI business goal for fiscal 2035.

The Register reported on HALEU, the higher-enriched uranium fuel behind small modular reactors pitched for datacenter power. Production remains slow, but Nusano says federal backing from the US Department of Energy could accelerate output, eventually.

NVIDIA published a guide to choosing between dense and Mixture-of-Experts models. Using Nemotron 3.5 Lightning, it shows how a 30B model can activate only 3B parameters per token while keeping the larger model's capacity, and maps active parameters to throughput.

NVIDIA outlined how NVLink 6 builds resiliency across multiple layers for AI factories. Because every GPU in a training cluster moves in lockstep, one failure can stall the job; NVLink 6 adds redundancy and recovery from the link layer up to keep training running.

NVIDIA described how deterministic execution in Groq 3 LPX delivers power-efficient, high-interactivity inference on Vera Rubin. With power the defining constraint for AI factories, fixed execution timing cuts idle waits while preserving responsiveness.