company
Nebius Group is a Nasdaq-listed cloud infrastructure company offering GPU computing and AI cloud services, spun off from Yandex's international assets.
No evidence-backed relationships are recorded.
Reflection AI is preparing its first open-weight model after signing compute deals worth $150 million a month with SpaceX and over $1 billion with Nebius. Open weights leave the infrastructure and upkeep with the customer, so how the model deploys is the test buyers can run themselves.
Reality
Nebius bought Inferize, a 10-month-old Israeli startup with 17 staff, for an estimated $100 million to $150 million, according to Calcalist. The money goes on software that keeps inference GPUs from sitting idle, and Nebius is still adding data center capacity alongside it.
Perspective Coverage
3 publishersReality
GMI Cloud raised $663 million: a $223 million Series B with Nvidia in the round and a $440 million credit facility from ChinaTrust Commercial Bank. Teams that need GPUs in Taiwan, Thailand or Malaysia get a funded local supplier whose expansion rests mostly on borrowed money.
Perspective Coverage
3 publishersReality
g factor's Qwen 3.8 27B benchmark has Together AI fastest at one stream, at 189.61 tok/s, while four of five engines finish within about 10% at 64 streams. Choosing a provider from these numbers starts with knowing how many streams the deployment will run at once.
Reality
New Jersey fined DataOne $1.07 million, its largest data-center penalty, for running 62 unpermitted gas generators, about 123 MW, at the Vineland AI site. Until the permits clear, that power runs on a regulator's timetable at a campus built to deliver GPU capacity to Microsoft.
Reality
MOUs with six large asset managers aim to mobilize more than $500 billion for AI factories. Nothing is funded yet, but the underwriting model already changes who sets compute prices.
Reality
Contract assemblers have warned of increases above 15% on Vera Rubin and Grace Blackwell systems shipping from early next year. Every AI capacity budget written before that is now short.
Reality
The accelerator is in full production and the headline number is a single-request generation rate at 100,000 tokens of context. That is a different purchase order than throughput.
Perspective Coverage
3 publishersReality
Groq 3 LPX is in full production with Nebius as the named first customer. The benchmark is one model at one context length, and the 4x claim does not quite get from hours to minutes.
Reality
NVIDIA's open source Topograph discovers cluster topology from five named clouds or from ibnetdiscover on-premises, then republishes it as node labels, Slurm config or Slinky ConfigMaps whenever a watched part of the cluster changes.
Reality
The AI data centre company wants $35 billion from public markets on a first half where direct costs exceeded revenue, a $1 billion net loss and a contracted-demand total of $103.4 billion. Underwriters have yet to set a range.
Perspective Coverage
7 publishersReality
Halo adds expert and tensor parallelism to Hugging Face models and still saves SafeTensors that from_pretrained can load. Its best number, 9,009 tokens per second per GPU against TRL's 3,885, came from synthetic fixed-length sequences.
Reality
Bloom's own percentages imply a $13.3 billion non-compute capex baseline and a $61 billion five-year cost of ownership for that gigawatt, with two thirds of the claimed saving arriving before the plant burns any gas.
Reality
Nscale's registration statement pairs $140.6 million of first-half revenue with a $1.02 billion net loss, and says the company holds no binding commitments for the financing its four Anthropic agreements oblige it to seek.
Reality
The reported collateral is Google's proprietary processors plus contracts with customers Crux has not named. A lender's recovery case therefore rests on someone else wanting TPU capacity outside Google Cloud.
Reality
MLPerf Inference v6.1 preview submissions put Vera Rubin NVL72 at up to 3.7x GB300 on Qwen3-VL and up to 2.5x on DeepSeek-R1, on two different inference frameworks. The four-rack 99% scaling result is an offline number.
Reality
Microsoft's fiscal 2026 filing puts $329.1 billion of leases into a queue that starts in fiscal 2027. The 38 gigawatts Bloomberg reported for 2032 would need a slower build than Satya Nadella has guided to.
Reality
Nvidia reported $96.2 billion of revenue at 75% gross margins and guided the next quarter to $108.0 billion while assuming nothing from China Data Center compute, with gross margin slipping to 74%.
Reality
CoreWeave has disclosed $5.1bn of equipment financing and Nebius says debt plus one customer contract covers more than 100% of a deployment's capex, which turns an enterprise compute order into a credit judgement.
Reality
The pitch is a software-style test loop for machines: model a customer's own hardware and sensors, calibrate against real runs, then try thousands of scenarios before booking rig time. Ring says the match held on scenarios it withheld.
Reality
Invest · September 10, 2026 · 2 publishers
Product · September 9, 2026 · 1 publisher
Invest · September 7, 2026 · 1 publisher
Product · September 6, 2026 · 1 publisher
Leadership · August 27, 2026 · 5 publishers
Invest · August 30, 2026 · 1 publisher
Invest · August 29, 2026 · 1 publisher
Product · August 25, 2026 · 1 publisher
Build · August 25, 2026 · 1 publisher
Invest · August 24, 2026 · 1 publisher
Invest · August 24, 2026 · 2 publishers
Invest · August 19, 2026 · 2 publishers
Product · August 19, 2026 · 1 publisher
Product · August 18, 2026 · 1 publisher
Product · August 17, 2026 · 1 publisher
Product · August 15, 2026 · 1 publisher