Q3 results in short (non official version)
1. we sold all we had
2. we build much more
70
128
1,333
482,334
Today we signed the almost 20 billions 5y contract with MSFT
@nebiusai now in the highest league.
nebius.com/newsroom/nebius-a…
55
100
1,173
119,073
As I get it from X: the last (and only) thing left for @nebiusai to make investors 100% happy is merch.
84
23
520
51,350
Lots of noise and bluff in this market.
Hard to tell what’s real and what’s just a press release.
@nebiusai’s first DC with first B300s in the UK isn’t “coming sometime maybe” — it’s live.
First customers will start in a days.
@nebiustf is using our footprint to scale inference for the UK’s vibrant AI builder community.
We are proud to announce our first deployment of @nvidia Blackwell Ultra in the UK.
Accelerated by NVIDIA Blackwell Ultra GPUs and NVIDIA Quantum-X800 InfiniBand networking, the deployment delivers unprecedented performance for generative AI and future foundation model development.
Read the PR: nebius.com/newsroom/nebius-a…
28
52
515
82,413
This is a big step and it gives much more than just revenue. It gives the fuel to build product we all came to @nebiusai for. And deliver more value to our beloved customers.
27
34
446
33,061
Today we’re launching Nebius 3.0 — “Aether”, our new version of cloud platform named after the Greek god of light.
v1.0: first true multi-tenant AI cloud
v2.0: unmatched performance for training & inference
v3.0 : enterprise-grade trust — SOC 2 Type II, granular IAM, full infra visibility
Read here: nebius.com/blog/posts/behind… 🔥
21
42
445
63,272
Guess how long after signing the pivotal deal it took our CEO to ask:
“Why hasn’t this (relatively small) new customer received an answer yet?”
Less than 1 hour.
That’s the obsession we need to keep as we build a true multi-customer cloud. The MSFT deal is the fuel for growth.
22
31
449
27,194
We’re launching Nebius Token Factory (@nebiustf), the next evolution of Nebius AI Studio.
The platform built for production-scale inference. 🔥
Now the best way to run scaled inference workloads:
- Vertivaly integrated with Cloud (flexible access to capacity, all the best Nvidia chips) drives the best prices
- Dedicated endpoints: Sub-second latency, autoscaling throughput and 99.9% uptime
- Enterprise trust: Zero-retention inference, SOC 2 Type II, with HIPAA and ISO 27001
- Governed collaboration: Teams & Access Management with RBAC, SSO and unified billing
- Fine-tuning and deployment: Adapt open models to your data using LoRA or full fine-tuning, then deploy optimized versions instantly, no infra setup, no GPU juggling.
- White glove support from AI Solution Architects
20
54
412
156,703
Cloud is the post-sales business. The hard work starts the minute the contract is signed.
But with this team in @nebiusai I’m confident.
No words to express the gratitude 🧡
23
13
350
21,478
What podcasts to participate in September-October to speak about Nebius full stack approach and building the truely software driven growth?
64
9
322
47,025
We’re launching Nebius Token Factory (@nebiustf), the next evolution of Nebius AI Studio.
The platform built for production-scale inference. 🔥
Now the best way to run scaled inference workloads
“Running inference at scale with healthy economics requires efficient on-demand and autoscaling capabilities. Nebius was the only provider that met our requirements — reducing overhead, simplifying management, and enabling us to deliver faster, more cost-efficient AI in production, ” said @alexmashrabov, Founder and CEO at @higgsfield_ai.
8
37
322
18,753
@nebiusai product is three layers.
Every layer has its customers.
Every layer above consumes the previous one.
Each layer below is designed to be a reliable platform for the next — providing flexibility and better economics.
1️⃣ Physical bare metal infrastructure
2️⃣ Multi-tenant cloud
3️⃣ Inference / fine-tuning platform @nebiustf
Stacked.
17
36
311
38,698
Nobody celebrates or argues the Gold status in ClusterMAX v2 in @nebiusai.
We just heads down back to work.
As usual.
21
17
304
24,547
Maybe our partners just want to be kind to us (or want larger GPUs allocation), but everyone who integrates with @nebiusai notice great experience with the platform and brilliant team.
18
28
301
15,932
Great to be on the list
Next step: moving from the ‘Hardware’ bucket (a bit of a confusing label 😅) to the ‘Hyperscalers’, haha)
We are updating the IVES AI Revolution 30 list reflecting recent momentum in the AI Revolution. We are adding CrowdStrike, Roblox, GE Vernova, and Nebius to our list of 30 tech companies that we believe will define the future of the AI theme over the coming years. 🔥🏆🐂🎯🍿
20
18
301
25,306
Q2 results for @nebiusai are out.
In my words:
1. We're growing fast & raised year-end guidance
2. Ramping up datacenters & megawatts to be able serving larger customers
3. Product improvements to serve largest customers of the market ⬆️: better MTBF, storage perf, 3rd-party benchmarks confirm it
4. Workloads diversifying (inference, @nebiusaistudio), customer base too (enterprises: @Shopify & @Cloudflare )
16
23
284
23,817
Something great
The world of neoclouds is full of fluff and endless comparisons of apples to oranges, and that's all in an environment of extremely expensive infrastructure.
nebius.com/economics-of-ai-c…
We spent the past few months analyzing the key factors that truly define the cost of AI model training, and showing how high-quality infrastructure can streamline development and maximize return on investment.
9
27
278
17,341
From yesterday’s talk with some very smart people:
If frontier LLMs stopped improving today, we’d still have a decade to unlock all the value across scenarios, verticals, and industries.
8
24
261
48,346
I’ve never spoken in public this much…
From new model architectures to the rise of agents, what will it take to power the next wave of AI? 🌍
That’s the question at AI Agenda Live NYC hosted by @theinformation: theinformation.com/events/ai…
Our own @romanchernin joins a panel with industry leaders from @WekaIO, @databricks, @AMD, @Cloudflare and more to explore: (Thread 1/4)
12
17
225
23,306
To make it clear
I have deep respect for what @CoreWeave has built and how they execute. It’s impressive work.
We do a lot of things differently. But learn from them.
As well as from other folks at the market
The real play is to build long term sustainable AI specialized platform.
11
10
231
23,707
I think the industry still hasn’t figured out what the AI native developer platform should be.
12
8
223
85,023
Why we’re building a hyperscaler-grade software platform at @nebiusai while much of the market focuses on just “deploying clusters” 🧵
7
20
221
45,193
I want @nebiusai to stay obsessed with every customer, every deal —
as if each one is life or death for the company.
The first year was wild.
I want us to keep that spirit alive.
15
13
221
17,337
Three phases of AI development
1. Foundational models
2. Vertical AI products
3. Enterprise adoption
11
11
220
35,078
We’re urgently looking for a Technical Content Writer / Technical Content Marketer for @nebiustf.
Can consider part or full time.
You’ll write:
• Technical deep dives
• Customer stories
• Real-world use cases
If “inference optimization” isn’t just a buzzword to you - DM me.
Location doesn’t matter.
28
12
216
28,874
Feels like we still can’t fully process how fast AI is distributing.
A few random examples:
- When ChatGPT-3 launched, every glamour magazine wrote about it in a few days.
- Phenomenal growth of new AI-first SaaS like Cursor or Lovable.
- “Neoclouds” moment when companies like @nebiusai going from 0 to multi-$B revenue in a year.
Even for those of us in the industry, it’s mind-blowing.
13
13
207
10,921
A few weeks ago we celebrated @nebiusai's 1-year anniversary. Everyone shared their anecdotes from the past 12 months.
One of mine:
In the beginning, we couldn’t afford 30-min meetings with potential customers and partners, because at least 20 mins went to explaining who the f*** we are and all the legacy.
Now we don’t have to. And can focus on business.
So proud (and relieved), haha :)
16
11
210
17,279
It’s funny how often people think Nebius mainly compete with CoreWeave.
The real game? Hyperscalers.
All “so called neo-cloud” players are expanding what AI-specialized cloud means.
The market will be huge — the real question is how we, convince customers that their critical AI workloads should be landed at AI specialized infrastructure.
3
14
200
18,563
Not merch, but also ok:)
nebius.com/newsroom/avride-s…
20
12
188
13,112
Something I’ve been meaning to post for a while.
We’re building Nebius on 3 core beliefs:
1️⃣ AI is real — not bubble, hype or a toy. It’s reshaping every industry. Infra demand will explode.
2️⃣ AI will stay competitive. Open ecosystem will remain strong, nations and regions will need their tech stack, many winners in products and models.
3️⃣ Nebius has the right to win: full-stack tech, resilient and passionate team, access to capital, real execution.
I think no one’s questioning (1) anymore.
I believe (2) turns out true — otherwise, the business challenges will be the least of our worries.
And (3)… we’re just doing it. Life will show:)
13
23
187
34,727
We first time mentioned some of our results in AI Research in shareholders letter
Here are some links
– SWE-rebench dataset on HuggingFace: huggingface.co/datasets/nebi…
– SWE-rebench leaderboard: swe-rebench.com
– SWE-rebench paper: arxiv.org/abs/2505.20411
– Reinforcement learning application towards long-context multi-turn SWE-agents: arxiv.org/abs/2508.03501
– Infrastructure blogpost about collecting datasets at scale and evaluating SWE-agents: nebius.com/blog/posts/infras…
10
6
177
18,121
@nebiusai is way more than just compute, we’re building a full developer platform to power AI products at scale.
A robust data platform is at the core of this vision, enabling data intensive workloads for AI.
tracto.ai/blog/better-data-i…
New article on how we make this possible.
5
23
164
5,360
Every time people got tools to iterate cheaper, faster, simpler (and without needing others), it triggered a nonlinear explosion of creativity and productivity.
That was with texts, photos, videos, you name it.
It will be the same with code.
7
8
159
18,080
new day, new release in @nebiusaistudio
our updated endpoint for DeepSeek is the fastest at the market
artificialanalysis.ai/models…
5
18
156
17,826
We had a great time last week with @Accel’s @caseyaylward and @weka talking about scaling inference
All the data centers and megawatts are cool, but the real play is we building inference platform and beyond to enable vertical ai builders and ISVs that applying AI at scale.
3
15
150
17,944
Funny fact: our AI platform’s internal codename is Newbius — a fresh build from scratch, the opposite of the old “Oldbius.”
We launched just 9 months ago, and it’s already one of the best. But we’re only getting started — the team ships big improvements every single month.
6
5
151
9,298
That's what Nebius Token Factory (@nebiustf) for
From yesterday’s talk with some very smart people:
If frontier LLMs stopped improving today, we’d still have a decade to unlock all the value across scenarios, verticals, and industries.
3
9
135
15,002
just @dylan522p in the push up contest at at @nebiusai booth
whom should we invite to the next contest🤔...
18
3
134
13,621
You need to onboard the new supplier @nebiusai to run at the pace and quality you need:)
11
4
132
21,393
We are going to @NeurIPSConf 🚀🥁🚀
🚀 Big news from Nebius AI R&D:
Our SWE-rebench dataset was the most downloaded on @huggingface earlier this year, and the paper has just been accepted at @NeurIPSConf 2025! 💡
Read it here 👉 arxiv.org/abs/2505.20411
Authors: @ibragim_bad, @agolubev13, Maksim Nekrashevich, @Shevan05, @simon_karasik, @Andrei Andriushchenko, @__vimary, Daria Litvintseva, @hr0nix
1
15
127
21,053
Arkady : The economics of the deal are attractive in their own right, but, significantly, the deal will also help us to accelerate the growth of our AI cloud business even further in 2026 and beyond.
4
4
126
9,482
🎙️ Recorded a 1.5h session with @Sam_Awrabi a couple of weeks ago.
How we build, quite comprehensive.
piped.video/AkBJgSAUWm0?si=X88w…
3
14
124
44,625
When our new CRO Marc joined, the team kicked off a project called Razgon. In Russian it means both acceleration (like overclocking a CPU) and the messy group brainstorming (like when creating comedy). Cultural fit💯
6
5
117
7,897
We joke that @nebiusai is more of a real estate business than a tech company—our most important meetings are always about capacity planning.
The toughest job? Capacity manager. Every seller is after you, begging for approvals.
And me? No good days. It’s always one of two crises:
- Not enough capacity
- Not enough demand
Guess what's the crisis today...
12
5
108
9,212
Folks, every tweet is retweeted multiple times with $NBIS
Why you need it from me?:))
20
4
124
26,681
cannot be more proud!
Here are some of the questions answered in the article:
- What causes reliability issues?
- How do people measure cluster reliability?
- How does cluster reliability affect goodput (compute utilization)?
- What measures do we have in place? (acceptance checks, health checks, node replacement process, etc.)
- What is our MTBF for a 3K cluster?
- What does a Nebius customer say about our reliability?
nebius.com/blog/posts/how-we…
4
6
102
7,892
I'm about to go live on the @ScalingEurope show with:
>> @aidancramer - Cofounder of @aiapplyco
>>@romanchernin - Cofoudner of @nebiusai
>>Tobias Bengtsdahl, partner at @AntlerGlobal
>> @charles_maddock - Cofounder at @strawberry
LETS GO
2
7
102
12,198
An inference platform is much more than ‘just inference’
The goal: lower the barriers for vertical AI product developers to build on open-source models.
1️⃣ Performance, reliability, scalability, security,
2️⃣ Tooling: evals, observability, integrations, etc
3️⃣ Turning data into quality of the tokens
4
7
100
4,794
September travel plans:
• London: Sept 10–11
• San Francisco: Sept 23–27
• New York: Sept 28–30
Let’s meet to talk inference platforms, scaling AI products, and enterprise adoption.
11
9
99
8,469
@nebiusai’s pitch is simple:
1️⃣ AI is a long-term trend – it will keep growing, and it needs massive infrastructure to grow with it.
2️⃣ The market stays competitive – we’ll have many models, strong open source, tons of products.
3️⃣ That’s where we win – scale + vertical integration. Not just a utility, but platforms for developers & businesses to apply AI at scale.
6
11
97
11,297
Had a lot of fun chatting with @antonosika, @Thom_Wolf and Matt Weigand in the @Accel’s beautiful space. There was even fight on open source vs close source models!
2
6
99
8,608
I’ll start to ask for feedback on different products we do.
First one
The coolest research effort at @nebiusai: SWE-rebench is the best benchmark out there for coding agents.
Drop a comment to @agolubev13 and the team!
Pls share 🙌
What are the key evaluation directions you'd like to see for LLMs as SWE-agents?
In SWE-rebench, we measure how well models solve GitHub issues. We're considering extensions, primarily measuring how well models write tests for these fixes, and extending the benchmark to other languages (like Java, Go, Rust, etc).
Some questions for you:
1. Which of these directions seems more important to you?
2. For multi-language evaluation, which languages are most interesting?
3. Are there other evaluation slices you'd like to see within SWE-agents?
2
9
95
9,962
Introducing Nebius MCP Server: The LLM-native way to manage your AI Cloud
In the last months, we made significant progress with product in @nebiusai
I’ll post a few highlights!
But this is my favorite funniest one!
The advantage of building a fully API-first multi-tenant cloud.
You let partners work with the platform OR AI-robots to use it!
nebius.com/blog/posts/introd…
3
12
93
7,735
One of my favorite initiatives of @nebiusai (with @NVIDIARobotics @nvidia support).
We still have few slots form judges.
Let me know if you are in robotics and want to join!
nebius.com/robotics-physical…
3
7
91
12,039
@nebiusai Q2 2025 earnings call is Thursday, August 7 — before market open.
As always, it’s more than just numbers: we’ll share key product updates and the progress that’s driving growth.
Product × happy customers = momentum.
group.nebius.com/newsroom/ne…
10
10
93
14,929
keynotes from @LiamFedus Natalia Gimelshein @apaszke @ezyang and the best crowd in the town
@caseyaylward @marksaroufim good luck!
@nebiusai is supporting the GPU Mode hackathon today
events.accel.com/gpumodehack…
1
8
91
7,299
Ok. The joke of last weeks
Thought a “champagne problem” was more champagne than problem… turns out it’s way more problem than champagne 🫣
7
2
86
10,356
SWE-Rebench is the most downloaded on HuggingFace.
We’re embarrassingly under-sharing the amazing work of our Nebius Research team.
Time to fix that! I'll post more about their main areas of work and why it's secret sauce of what we do.
🚀 Big milestone: our SWE-rebench is now the most downloaded dataset on HuggingFace!
Honestly, when we released it just two months ago, I didn’t expect such numbers.
So far, it has been downloaded 3.5M times in total, and 2.8M downloads came only in the last month.
2
9
84
6,922
Who is doing the best startup program in the industry?
What Nebius should/can do beyond providing compute credits?
15
6
82
8,227
Inference + infra, vertically integrated = better pricing & flexibility for customers at scale. That’s why🧵
3
6
72
5,172
Another great thing team at @nebiusai shipped recently is a new Object Storage, designed for streaming data to NVIDIA GPUs and checkpointing.
With up to 2 GiB/s write throughput per GPU.
nebius.com/blog/posts/q2-202…
3
5
62
3,070
@nebiusai QBR in Amsterdam & Finland – priceless for the sales team to see what’s behind those faceless GPU*hours.
5
2
59
3,793
We just deployed at @nebiusai new h200 clusters and have a unique offer to fill the short-term idle before the long-term contracts kick off.
Let me know if anyone needs burst capacities of H100 or H200 up to 2-4k for 2-12 weeks starting in December.
Let's play a bit
What price would immediately sell the capacities out?
Our pricing starts from $1.5 gpu/h.
6
56
34,675
@sama : We are ~doubling our compute fleet over the next 5 months (!)
—-
The largest, but still just one player at the market.
That’s what happens with AI products when they find the market fit and explosively grow. AI is real change.
Here is how we are prioritizing compute over the next couple of months in light of the increased demand from GPT-5:
1. We will first make sure that current paying ChatGPT users get more total usage than they did before GPT-5.
2. We will then prioritize API demand up to the currently allocated capacity and commitments we've made to customers. (For a rough sense, we can support about an additional ~30% new API growth from where we are today with this capacity.)
3. We will then increase the quality of the free tier of ChatGPT.
4. We will then prioritize new API demand.
We are ~doubling our compute fleet over the next 5 months (!) so this situation should get better.
5
2
55
3,230
run fast
fix fast
artificialanalysis.ai/models…
4
5
54
4,864
We are very proud to support the growth and speed of @recraftai 🧡!
🚀🚀 We’ve added a batch of new B200 GPUs, so image generation in Recraft is now 2x faster.
🚀🚀🚀On top of that, generative upscale is up to 3x faster.
Big thanks to our long-term partners at @nebiusai for supporting us on this move 🙌
1
49
3,552
The usual AI product journey:
1️⃣ Start with the best closed-source SOTA model → prove the use case, find PMF, scale fast.
2️⃣ Switch to open-source → fine-tune with proprietary data to match quality at a fraction of the cost → scale with far better unit economics.
agree?
4
3
45
25,674
@nebiusai is celebrating the 1st official anniversary of the company today. Insane. This kid of supercomputer and hyperscaler is just 1 yo:))
- Scale is critical. There are a lot of possible futures for Nebius as an AI infrastructure, but no successful scenarios if we are not large enough
- The scale is not enough. We need to provide value across the different segments of customers through the software. The real growth will come from “enterprises” (whatever we mean:)) where hyperscalers are the competitors, and pure infra play will not help.
That’s why we bet on vertical integration, and we go up the stack
Next year, I will be dedicated to making @nebiusaistudio (inference platform) super successful.
We build the foundation to help people build new SaaS products and apply AI to real use cases.
4
7
43
23,692
sifted.eu/articles/europes-a…
good stuff from @Siftedeu with @antonosika and my partner in crime building @nebiusaistudio, our CTO.
---
“You can’t rely on a closed-source ecosystem unless your plan is to be part of that ecosystem forever,” he says. “This isn’t just about cost optimisation but also ensuring business continuity.”
Loveable’s Osika agrees, stating that while the best coding models will likely remain closed source in the near term, open source is crucial for transparency, customisation and avoiding vendor lock-in.
“We aim to work with both,” he says. “The real value isn’t just in the model, but in the entire system we’re building around it that makes AI coding reliable, fast and secure.”
2
4
40
4,258
Cloud is not a sales-driven business, it’s post-sales driven. When we sign the customer, we only sell the promise. The behaviour of not first-time buyers is completely different!
4
35
3,215
Our beloved customer @trymirage under the new lovely name.
We're changing our name. Captions is now Mirage.
Today's most popular medium is short-form video. It's the fastest-growing format, it’s what people want, and it’s what brands need to grow.
But unlike other forms of AI content like text or image, which can be generated and ready-to-use with a single prompt, generated video still requires a lot of manual work and editing to be ready to share. This is where the real opportunity lies.
The way we see it, the real race for AI video hasn’t even begun.
The world is waiting for a foundation model for video that can take an idea and produce a fully usable, cohesive output from scratch. So that’s what we’re building.
We’re rebranding to Mirage to reflect our expanded vision for video, starting with the short-form format, through frontier AI research and models.
Check it out at mirage dot app
6
39
4,587
The most important event for open weight models this year?
🚨 Now live on Nebius AI Studio:
GPT-OSS-120B and GPT-OSS-20B: @OpenAI's new open-weight models.
✅ Instant access, no GPU setup
✅Playground + API
✅ Open source, zero-retention
✅Enterprise-grade, scalable inference
Try them now 👇
1
37
3,803
@nebiusaistudio is the launch partner of @NousResearch
@nebiusai is the go-to destination for serving your models efficiently once they’re released
Nebius AI Studio 🤝@NousResearch
Introducing:
🔹 Hermes-4-70B
🔹 Hermes-4-405B
We’re proud to power Nous Chat with these new models, and make them available today in the Nebius Playground + API.
👉 Try them today: studio.nebius.com/
8
37
5,235
My fellow Londoners — DM @nebiusaistudio if you’d like to join
(seats are very limited, not a public offer)
🚨 London Founders, we’ve got 5 spots left.
On September 11, Nebius AI Studio is hosting an intimate evening with @Accel, founders from @lovable & @snyksec, and other special guests.
Topic: How open-source models are redefining the AI game.
Expect a fireside chat, good food & drinks, and conversations with some of the sharpest minds building AI products today.
👉 If you want in, DM us. Seats are super limited.
1
4
31
6,450
> In the process of securing more than 1 GW of power by the end of 2026
a lot of work ahead
23
23
1,149
5/5 – Self-Service Experience
Everyone’s chasing the next Perplexity or Cursor — but what about the customer who needs 1 node for 1 week? We care about them, too.
1
28
2,803
build without limits with @nebiusaistudio
1
13
1,889