Business Insider reports, citing internal emails, that Amazon is buying Nvidia GB300 server racks faster than it can power them, spending roughly $5 billion in August alone through its 'Early Rack Receipt' process and about $3 billion monthly since. Some hardware is being stored in intermediary warehouses because AWS sites lack the electricity to install it, and one internal team reportedly said there was no clear timeline for deploying the accumulating GPUs.
techspot.com
· 2026-10-24
Microsoft has pulled out as anchor customer for Nscale's proposed 50MW (expandable to 90MW) AI data center in Loughton, Essex, which was meant to house 23,040 Nvidia GB300 GPUs for Azure starting in 2027. The withdrawal reportedly follows warnings from UK Power Networks that sufficient grid capacity may not arrive until the early-to-mid 2030s. Nscale says the project continues and it is seeking alternative power sources and sites.
hardware.slashdot.org
· 2026-10-20
Reflection has unveiled Beam, its first open-weight model with 501 billion parameters, designed for tasks like coding, reasoning, and agentic workloads. The model was pretrained on 23.8 trillion tokens and refined through extensive reinforcement learning using over 100 million rollouts on NVIDIA GPUs, aiming to achieve high efficiency and performance. Beam is currently undergoing final testing before a public release of its weights and technical documentation later this month.
reflection.ai
· 2026-10-05
Elon Musk said xAI will bring 220,000 Nvidia GB300 GPUs online next week, another 220,000 in November, and potentially 220,000 more by late December. Combined with existing hardware at Colossus 1 and Colossus 2, this would give xAI about 1.1 million GB300 GPUs and 1.44 million GPUs in total. Musk also noted that Colossus 1, an inefficient mix of Hopper and Blackwell chips for training Grok, has been rented out to Anthropic for inference.
tomshardware.com
· 2026-09-25
OpenAI has fully unveiled Jalapeño, its first AI accelerator chip, boasting up to 13.4 petaflops of 4-bit compute and 232GB of high-speed memory. The company says benchmarks show up to 3.6x lower latency than Nvidia's GB300 at lower power, and that the chip went from concept to silicon in under 20 months using its own large language models to accelerate design work with a team of fewer than 100 engineers.
spectrum.ieee.org
· 2026-09-18
GoKawiil