Nvidia's Blackwell-generation rack-scale AI server system pairing Grace CPUs with Blackwell GPUs for large-scale training and inference workloads.
Nscale's 90MW Essex datacentre may lack grid power until the early-to-mid 2030s, years after its planned 2027 start. The date depends on national transmission work outside the company's control, and Nscale is preparing a US listing.
Reality
Inferact measured 709 output tokens a second on 16 Ironwood chips against 452 on 16 GB200s, at low concurrency with speculative decoding. The engineering worth reading is the hand-written memory schedule underneath.
Reality
Where a robot's inference runs decides what the machine costs to build. SemiAnalysis puts the industry's upfront compute bill in the trillions at a billion robots, paid by manufacturers on every unit before it does any work.
Reality
A JAX and Transformer Engine version of the same GB200 run reports 1,068 TFLOPS per GPU, a 10.4x gain. The 84% share bounds what fixing communication alone could have bought, so the expert matmuls got faster too.
Reality
The 1.7x to 3.6x latency range is set by the baseline systems, not the chip, and the report's own publication date is unsettled. Read it as direction, not evidence.
Perspective Coverage
9 publishersReality
Dynamo can now run the vision encoder as its own worker, and the win is real where encoding dominates a request. The arithmetic behind the headline number tells you which traffic it transfers to, and NVIDIA names the one topology it never beats.
Reality
NVIDIA reports a five-fold memory cut, 100,000 instruments on one GPU, and a 13-second refresh. The structural-break signals it sells the pipeline on go unmeasured in the post.
Reality
Lightricks moved the detail recovery step inside the VAE decoder. That removes a stage from the pipeline, adds a second sampler to the stack, and makes decode latency something you now have to measure.
Reality
NVIDIA's tutorial post-trains a 4B model into a Franka manipulation policy that runs on Jetson Thor with about 0.6 seconds of latency slack per cycle. Closed-loop success is 22.9%.
Reality
GPU rental prices fell more than 60% in 2025. SemiAnalysis data says H100s have since climbed about 40% to $2.35 per GPU-hour, with 12 to 18 month waits for clusters.
Reality