Serverless GPU Sandboxes (CoreWeave)
CoreWeave Serverless GPU Sandboxes are free during the preview
AI Year in Review · Oct 10, 2025 – Oct 10, 2026
Hardware, chips & infrastructure
61 stories in this category.
RL Rollouts (CoreWeave)
CoreWeave RL Rollouts hot-load weights about 15x faster
Serverless GPUs (CoreWeave)
CoreWeave launches serverless GPUs: GPU sandboxes by the hour, no contract
Vera Rubin NVL72 in production (CoreWeave)
Cognition becomes the first production customer on Vera Rubin NVL72 at CoreWeave
Serverless distillation (CoreWeave)
CoreWeave adds serverless model distillation to Forge
Monthly roundup: ThursdAI Oct 2026 ↗
Monthly roundup
Jev-style decision models (SGLang)
SGLang adds native support for Jev-style decision models
Constellation and Amazon
The companies signed a 20-year agreement for 690 MW from the Calvert Cliffs nuclear plant in Maryland, including a 190 MW uprate expected between 2030 and 2032 .
AIB Data Centers and Nebius
AIB contracted 50 MW of critical IT capacity in the southeastern US to Nebius for an initial 12-year term.
ClusterMax 3.0 Platinum (CoreWeave)
CoreWeave earns Platinum on SemiAnalysis ClusterMax 3.0 again
Monthly roundup: ThursdAI Sep 2026 ↗
Monthly roundup
Microsoft, Hyderabad
Microsoft opened its India South Central cloud region with three availability zones.
Boston Dynamics
Boston Dynamics opened a center at Hyundai’s Georgia Metaplant to train Atlas robots for car manufacturing.
CoreWeave
CoreWeave said its contracted power reached about 4.2 GW as of August 11, up from about 3.7 GW at June 30.
Figure Helix 2.5
Figure says Helix 2.5 achieved “zero-shot whole-body autonomy across 30 real homes.”
Agility Digit 5
Agility Robotics unveiled Digit 5 with AI-based detection of nearby people.
Oracle
Oracle’s fiscal Q1 2027 release reports 850 MW of additional datacenter capacity and more than 300,000 GPUs delivered in the quarter.
NVIDIA in Australia
NVIDIA said Australian partners are preparing land, power and shell capacity for its DSX AI factories, “with up to a 2-gigawatt buildout by 2027.”
Kimi K3 on Dedicated Inference (CoreWeave)
Kimi K3 (2.8T) runs on CoreWeave Dedicated Inference on GB300 NVL72
Figure and Nscale
Figure signed with Nscale for up to 100,000 NVIDIA Vera Rubin GPUs in Barstow, Texas, with initial deployment from the second half of 2027.
Broadcom
Broadcom reported AI semiconductor revenue of $16.7 billion for its third fiscal quarter, up 221% year over year.
Mac Studio (M5 Max / M5 Ultra) (Apple)
Apple announces Mac Studio with M5 Max and M5 Ultra plus a new Mac Mini
Monthly roundup: ThursdAI Aug 2026 ↗
Monthly roundup
MicroDuck robot kit (Hugging Face)
Hugging Face + Pollen Robotics ship a $399 walking mini robot kit
Nvidia Q2 fiscal 2027
Nvidia reported revenue of $96.2 billion for the quarter ended July 26, 2026, of which data center revenue was $89.0 billion. Its outlook for third-quarter revenue is $108.0 billion.
Nvidia Groq 3 LPX
In a release issued at Hot Chips, Nvidia said the Groq 3 LPX, an inference accelerator that extends its Vera Rubin platform, is “now in full production,” with Nebius as the first AI cloud to adopt it. Each rack holds 256 LP
PORTS-Pike data center (OpenAI)
OpenAI joins PORTS-Pike: 8 GW Ohio data center on a 20-year lease
Cerebras CS-4
Cerebras unveiled the CS-4, built on its WSE-3 Turbo (WSE-3T) wafer-scale processor with three wafers per rack, and said “first CS-4 shipments begin this quarter.” The CS-4 datasheet lists 44 GB of on-wafer SRAM per wafer and r
OpenAI PORTS-Pike, Ohio
OpenAI said it has agreed to secure approximately 8 GW-IT at the PORTS-Pike Technology Campus in Pike County, Ohio, the site of the former Portsmouth Gaseous Diffusion Plant, working with SB Energy, NVIDIA and the US
GPT 5.6 Sol ultrafast preview (OpenAI)
OpenAI previews ultrafast GPT 5.6 Sol on Cerebras at ~14x speed
Weave BYOB (Weights & Biases)
Weave ships BYOB: media stays in your own S3/GCS bucket
Cerebras and GPT-5.6 Sol
Cerebras said it powers an Ultrafast mode for OpenAI’s GPT-5.6 Sol.
GPT-5.6 API pricing (OpenAI)
Breaking on the show: OpenAI cuts GPT-5.6 Luna prices 80% and Terra 20%, crediting Sol's self-optimization
Crusoe and Aalo Atomics, Idaho
Crusoe partnered with Aalo Atomics on a nuclear-powered AI data center: a proof of concept at Idaho National Laboratory in 2027 using the 10 MWe Aalo-X reactor, and 50 MWe Aalo XMR plants at Crusoe sites by the e
Gemini Robotics 2
Google DeepMind introduced three robotics models: Gemini Robotics 2, a vision-language-action model; Gemini Robotics ER 2, an embodied-reasoning model; and Gemini Robotics On-Device 2, which runs on the robot itself.
SK hynix HBM4
In its second-quarter 2026 results, SK hynix said it “began mass shipments of HBM4 in the second quarter and will ramp up production in the second half of the year.”
Brookfield and NextEra, Kentucky
NextEra announced a data-center campus at the Department of Energy’s Paducah Site with more than 1.2 GW of compute capacity and up to 4.6 GW of dedicated generation, with construction to be complete by 2032, su
AMD Instinct MI400 Series
At Advancing AI 2026, AMD launched the MI400 Series: the Instinct MI455X for “frontier AI and AI factory deployments” and the Instinct MI430X for sovereign AI and high-performance computing. AMD’s MI455X product page
AMD and Anthropic
AMD said Anthropic will deploy up to 2 GW of Instinct MI450 Series GPUs in Helios racks, with the first gigawatt beginning deployment in the first half of 2027.
OpenAI, Georgia
OpenAI announced a campus in Effingham County with 3.2 GW of power from Georgia Power, delivered in phases between 2028 and 2032.
Gemma 4 (stealth update) (Google DeepMind)
Google quietly patches Gemma 4 with Flash Attention 4 and tool-calling fixes — no version bump
Crusoe and Lancium, Texas
The two companies announced a 1 GW campus on a 270-acre Lancium site in Childress, with construction expected to begin in the third quarter of 2026.
Pure DC, Finland
Pure Data Centres Group launched its Seinäjoki AI campus: a 110 MW first phase that is fully leased, with the full campus planned at more than 550 MW subject to permissions.
Meta, Louisiana
Meta said it is expanding its Richland Parish data center to 5 GW of compute capacity, describing more than $50 billion of investment in the region and a new energy agreement with Entergy Louisiana.
PyTorch 2.13 (PyTorch)
PyTorch 2.13 lands FlexAttention on Apple Silicon and big memory wins
Meta, Alberta
Meta broke ground on a 1 GW, AI-optimized data center in Sturgeon County, its first in Canada, which it says represents an investment of more than CA$13 billion once complete.
local.ai (Exo Labs)
Exo Labs launches local.ai to track the local-AI frontier
Jalapeno (OpenAI)
OpenAI unveils Jalapeno custom inference chip with Broadcom
GPT-5.5 and Codex on Bedrock (Amazon Web Services)
AWS brings GPT-5.5 and Codex to Bedrock as Azure exclusivity ends
Gemma 4 on W&B Inference (Weights & Biases)
Gemma 4 goes live on W&B Inference with LoRA inference support
Anthropic, Meta & Jane Street deals (CoreWeave)
CoreWeave signs Anthropic, Meta ($21B), and Jane Street ($6B + $1B)
GR LPX (Rubin NVL72 + Groq 3) (NVIDIA)
NVIDIA GTC: GR LPX pairs Rubin NVL72 servers with the new Groq 3 chip
Monthly roundup: ThursdAI Mar 2026 ↗
Monthly roundup
ChatJimmy (baked-weights chip demo) (Taalas)
Taalas demos 15,000+ tokens/sec with model weights baked into silicon
W&B Inference: MiniMax 2.5 & Kimi K2.5 (Weights & Biases)
W&B Inference adds MiniMax 2.5 and Kimi K2.5
Kimi K2.5 on W&B Inference (Weights & Biases)
W&B adds Kimi K2.5 to its inference service
W&B Inference (GLM-5 & Kimi K2.5) (Weights & Biases)
W&B Inference adds day-zero GLM-5 and Kimi K2.5 support
OpenAI x Cerebras Partnership (OpenAI)
OpenAI inks $10B deal with Cerebras for 750MW of high-speed compute
Monthly roundup: ThursdAI Jan 2026 ↗
Monthly roundup
Vera Rubin (NVIDIA)
NVIDIA Vera Rubin platform: 5x Blackwell inference at CES 2026
Project Digits (NVIDIA)
NVIDIA Project Digits: $3,000 desktop that runs 200B-param models
Iron (XPeng)
XPeng unveils 'Iron' humanoid robot with soft skin and 2026 production plan
NEO (1X Technologies)
1X opens orders for NEO, a $20k consumer home humanoid shipping in 2026
M5 chip (Apple)
Apple announces M5 chip with double the AI performance
DGX Spark (NVIDIA)
NVIDIA DGX Spark: a desktop personal supercomputer for local AI
OpenAI x Broadcom custom accelerators (OpenAI)
OpenAI and Broadcom to deploy 10 gigawatts of custom AI accelerators