China-linked influence operations tapping US divisions on AI bit.ly/4a25i81

actual new world order @nebiusai

On open source models: every version of the "AI margins collapse" argument rests on one assumption almost nobody states: that frontier grade oss models stay abundant and effectively free. That's the load bearing beam. It's why a business can route around an expensive closed API, why the cheap spot market exists, why the labs can't raise prices into their negative margins. Pull the beam and the whole structure shifts. For two years that beam was mostly Chinese 🇨🇳DeepSeek, Alibaba's Qwen, Moonshot's Kimi, ZAI, GLM, the big Chinese labs handed over frontier weights, and the commodity tier was built on them. So the popular conclusion is a clean binary: the day China decides to stop, the closed labs get their pricing power back. But the real picture is already more interesting than an on/off switch, and it's moving in two directions at once. First, China is already closing selectively, not someday, now. Alibaba has launched hosted, closed Qwen variants, Zhipu put a GLM turbo model behind a closed endpoint, and ByteDance and Tencent keep their flagship multimodal and video models proprietary. DeepSeek is increasingly the open outlier, not the rule. -> the reason is unglamorous: training frontier models is brutally expensive, Chinese labs have far less capital behind them than their US rivals, and good vibes open sourcing doesn't pay for the cluster lol. -> There's also a deeper logic: the open strategy was never charity, it was a flywheel: free weights drive adoption into China's vast manufacturing and robotics base, which generates proprietary real-world data, which trains the next model. Closing the frontier protects revenue and that data, but it risks slowing the very adoption that feeds the loop. That tension, not ideology, governs the switch. Second, and this is the part the binary framing misses, the cheap tier is no longer China's to control. The US is building its own open supply. NVIDIA's Nemotron ships open weights with open datasets and reproducible recipes, and a coalition has formed around it: @PrimeIntellect, which recently released INTELLECT-3 and brings the reinforcement-learning and environment tooling that turns base weights into working agents, @NousResearch on post-training and agentic tooling, plus @Mistral and others still putting out open models. If Chinese frontier weights get scarcer, that ecosystem can keep the commodity tier stocked, especially for the agentic and specialized work where post training matters more than raw scale. So the switch to watch isn't one lab's decision. It's a balance: Chinese commercialisation pressure on one side, the momentum and quality of the US open ecosystem on the other, and underneath both, whether open models keep closing the capability gap fast enough to keep price pressure structural. The bear case for closed labs needs the open supply to dry up. The bull case needs it to stall. Right now neither has happened, and the beam is being loaded from two directions instead of one. Everything downstream, the pricing power, the margins, the spot market, resolves out of that balance. Not the next benchmark. The incentives and the capital flows on both sides of the Pacific. (image courtesy of whatllm.org, showcasing the open models ecosystem)

"Attracting new and creative sources of highly efficient capital (news soon)" - $NBIS CCO A sovereign AI investment from EU, UK, or even US could be on the table. I'm expecting good news because an exec wouldn't hype up a financing news just to drop a new ATM. There's also a Softbank investment possibility, @DrTomsLens explained it in the tweet below:

$NBIS Weekly recap. Index Money Is Coming, but the Real Story Is Demand. If you only read one line about Nebius (NBIS) this week, read this one: the company is about to be force-bought by every passive fund on the planet, and it earned that spot by selling out of compute faster than it can pour concrete. Nebius joins the Nasdaq-100 on June 22 The Nasdaq-100 underpins more than 200 investment products and a vast pool of tracked assets. Every index fund and ETF benchmarked to it now has to own $NBIS, and most of that buying has to be in place before the 22nd. $NBIS Inflection event: Nebius finally explained the strategy to everyone that was invited to the private event in SF. From AI Cloud -- Inference -- AI Agentic to Physical AI & Robotics in the future. Some highlights: * Nebius Echo: Agentic AI for enterprises. * Data Centers: Expanding in USA & EU with 75% owned. * Nebius customer base growing with many top AI players & Fortune 500 enterprises coming in. * Nebius Customer Advisory Board: Advance Machine Intelligence, Black Forest Lab, Cloudfare, Cognition, Cohere, Core Automation, Higgsfield, Recraft, Revolut, Rhoda. It was a very important event to show the industry Nebius is leading in multiple fields. TD SYNNEX partnership dedicated GPU Cluster capacity is gone. Earlier this year the global distributor did something no IT distributor had done before, it reserved a dedicated, AI-factory-grade NVIDIA cluster (1,000 HGX B300s) directly from an AI-native cloud provider, putting Nebius capacity into a channel that reaches 150,000 customers across 100 countries. The update worth noting: that reserved capacity has effectively been SOLD OUT. Laurelle Roseman VP Global Partnerships mentioned planning for Cluster 2.0 is in progress. Roman Chernin (co-founder, CBO) on Harry Stebbings' 20VC said Nebius could sell 10x more compute if they had the capacity. A few things stood out: Jevons Paradox is the demand engine. As the unit cost of intelligence falls, total consumption rises, because tasks that were uneconomic suddenly aren't. Cheaper inference doesn't shrink the market; it expands it. Pricing power is capped by customer economics, not by Nebius. Push inference prices too high and customers' margins break and demand stalls. The durable edge isn't the sticker price on a GPU — it's total cost of ownership: caching, runtime optimization, and distillation can move token economics by an order of magnitude. And the part that ties back to this publication's earlier work: Stebbings noted Leopold Aschenbrenner has made Nebius one of his largest positions. The Situational Awareness thesis and the operator's own demand commentary are now pointing at the same place. Roman Chernin clearly states, the binding constraint on Nebius revenue is not demand, it's how fast they can build. Which is exactly what the next two news are about. A £1.7B UK buildout: $NBIS plants its flag as a sovereign-AI supplier in UK. At London Tech Week, Nebius committed approximately £1.7 billion ($2.26B) to UK capacity: three new NVIDIA deployments on top of its existing Chertsey site (Ark Data Centres, live since November 2025 on Blackwell Ultra), taking the UK footprint to four sites and 65 MW when fully ramped in 2027. The company is also expanding its London commercial and R&D hub. The strategic angle is the part to underline. The investment is explicitly aligned with the UK Government's AI Opportunities Action Plan, and the AI minister welcomed it in those terms, this is Nebius positioning as domestic compute for British enterprises, researchers, and public services, an alternative to the US hyperscalers. The proof of concept is already named: Revolut rebuilt its AI stack on Nebius, running FinCrime agents and a support orchestrator handling 1M tickets a month. Also out of London (June 9): the Physical AI Living Lab, a six-month program with NVIDIA for UK and European robotics startups. Founders get the full physical-AI stack: NVIDIA OSMO, Cosmos world models, Isaac Sim and Isaac Lab, the Physical AI Data Factory Blueprint, plus Voxel51's FiftyOne for synthetic data — running on Nebius's UK infrastructure (NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs). Applications run through NVIDIA Inception; the first cohort starts in September 2026. Netherlands: Founder and CEO Arkady Volozh is joining the Dutch AI Infrastructure Think Tank, an initiative led by VOLT's Han de Groot, sitting alongside former ASML chief Peter Wennink. The group is examining how AI compute can strengthen the Dutch digital economy — already ~200,000 jobs and roughly €21B in annual added value — and the conversation is tied to a Dutch "AI Gigafactory" ambition that traces back to the Wennink report. For a company headquartered in Amsterdam, this is Volozh putting Nebius at the center of European compute-sovereignty policy, not just supplying it. Between the UK commitment and this, a clear pattern emerges: Nebius wants to be the European champion that governments point to when they talk about keeping AI "at home." Marketing: refreshed web presence & new intro video Nebius also rolled out a refreshed public-facing site this week, leaning into the agentic and full-stack-platform positioning that ran through Inflection. Minor on its own, but it's consistent with a company repackaging itself from "GPU landlord" to "the platform you build agents on." Resuming, a week of lots of news as we progress into Nasdaq-100 inclusion on june 22 but with a very clear idea... the strategy is a total success given the sold out capacity, the increasing customer base & the new adition of Fortune 500 enterprises that will be the next target for our customer success cases. Execution continues at its finest from $NBIS team. Nebius defined is an AI Native Hyperscaler Ecosystem of different verticals all growing exponentially. The next few years will be of absolute hyper growth as $NBIS is positioned to capture massive TAM in several fields.

One year ago, we put humanoids in a ring. People came for the robots and stayed for the fights, pilots, chaos, and crowd. That became Ultimate Bots. June 26: Automata opens the gates for our One Year Anniversary Showdown, powered by @nebiusai Our last event sold out. DM us for early access - tickets are limited. See you ringside 🤖🥊

$NBIS I'm now closer to my goal of 6000 shares. Started 5340 and ended 5660 this week. Addition of 320shares & 340shares to 6000. I'll probably spread my buys across 2 weeks period from next week onwards. On hindsight, DCA-ing 10 shares each buy is my strategy so i do not get captured hostage in such a market correction. Instead of buying 200 shares at 270, i can do 10x20 shares at different price points each time. @nebiusai tagged an ATH of $282 and ended at $228 this week. Hold strong! Unfazed. Unshaken.

$NBIS ARR this year will be 50% hyperscaler deals, 50% own AI Cloud. This is by choice. Hyperscaler demand is "infinite". Their own AI Cloud is however where they'll make more profits from sustainable, recurring, high margin revenue long term.