Data Center Dynamics
Datacenter
Sept 3, 2026
Michigan miner flips to neocloud tenant — 20 MW by Q4, lease worth up to $1.2B
Hyperscale Data shut down mining at its Dowagiac site on September 1 to prepare for a California neocloud: 20 MW live by Q4 2026, scaling to 52 MW, on a 10-year lease with two 5-year extensions. This is the deal template of 2026, with public numbers on it.
NVIDIA Newsroom
NVIDIA
Aug 26, 2026
AWS and NVIDIA commit 2 million more GPUs for 2027–2028
The largest single allocation headline of the year: two million Blackwell Ultra, Rubin and Rubin Ultra GPUs land at AWS across 2027–28, on top of the million-plus already committed. The queue for everyone else just got longer.
ServeTheHome
AI
Aug 24, 2026
AMD details MI455X and the 72-GPU Helios rack at Hot Chips: 432 GB HBM4, 2.9 EF per rack
AMD’s first rack-scale system is real: 72 MI455X GPUs, 31 TB of HBM4 and 2.9 exaflops per rack, squared up against Vera Rubin NVL72. 2027 capacity is now a two-stack quote — and today’s MI300X-class silicon is the cheapest inference compute on the market.
NVIDIA Newsroom
NVIDIA
Sept 3, 2026
Vera Rubin ramps to full production — first shipments begin this fall
Production shipments start this autumn, with first allocation to CoreWeave, Lambda, Oracle Cloud and Azure. For everyone else, the practical question isn't when Rubin ships — it's when it ships to you.
Industry analyses
AI
Sept 3, 2026
Hyperscaler AI capex barrels past $600B for 2026
Estimates now cluster between $600B and $690B of infrastructure spend this year — absorbing GPUs, transformers and grid queues, and pushing everyone else to the secondary market.
Utility Dive
Power
Sept 3, 2026
FERC puts grid operators on the clock for large-load interconnection
Show-cause orders to six RTOs demand answers on co-location and behind-the-meter generation — a regulatory turn that favors sites with power already in hand.
Data Center DynamicsDatacenter6h ago
Bulgarian 800 MW campus enters phase-1 commissioning
The Sofia-region AI campus, backed by a consortium of Nordic hyperscalers, will bring 60 MW online in Q3 with a further 740 MW roadmap across 2027–2029.
The RegisterAI1d ago
AMD MI355X shows 1.3× throughput vs H200 on select inference workloads
MLPerf inference results published this week show AMD's latest Instinct accelerator pulling ahead on LLM inference — though training remains an NVIDIA stronghold.
BloombergDatacenter1d ago
US-East power constraints push new AI capacity toward Midwest and Nordics
Grid interconnection queues in N. Virginia now exceed 7 years — accelerating a shift of hyperscale AI build-out toward Ohio, Iowa, Sweden, and Finland.
