Nebius Group (NBIS): Q2 2025 Earnings Review
Today, Nebius Group (NBIS) released its Q2 2025 results.
As of writing, the stock is up over 20% after beating expectations and raising guidance.
It’s the largest position in my portfolio, so I couldn’t be happier.
In this article, I’ll break down everything you need to know about the Earnings Report.
Financial Highlights
Revenue of $105.1M vs. $101.2M est. (+625% YoY and +106% QoQ)
This might seem like a small beat, but it’s not. It’s much more than that.
As I explained yesterday on my X account, this was the first quarter that excluded Toloka’s numbers from the consolidated results. However, analysts didn’t adjust their expectations accordingly, which had me concerned about a possible miss.
Believe me, beating these expectations without Toloka’s help is a very meaningful beat.
June ARR of $430M (+72.7% QoQ)
Gross margins of 71%, up from 47% QoQ
Adj. EBITDA of $(21M) vs. $(59.6M) est.
More important than the beat itself is the fact that NBIS announced its core business is already Adj. EBITDA profitable, well ahead of schedule. That’s a major milestone.
The company also emphasized that Group EBITDA will be positive in H2 2025, still negative for the full year, but expected to be positive for the full year 2026.
GAAP Net Income was positive, well above negative estimates.
However, this was due to a non-cash gain from the revaluation of ClickHouse and the third-party investment in Toloka.
ClickHouse recently raised funding at a much higher valuation, which increased the value of NBIS’ stake.
Maintained CapEx guidance of ~$2B
Even more important: the guidance was raised across the board.
Year-end ARR guidance RAISED from $750M-$1B to $900M-$1.1B
Year-end capacity guidance raised from >100 MW to 220 MW
Year-end 2026 capacity guidance raised from “up to 1 GW” to “ over 1 GW”
This is HUGE.
Honestly, 220 MW is enough for NBIS to reach well above its current ARR guidance. I seriously think they’ll raise it again in the next quarterly report, assuming the capacity rollout continues as expected.
This is still a fairly young company, so it wasn’t yet clear how they'd approach guidance on revenue/ARR or margins. We had some clues from the 4-year GPU depreciation schedule (compared to 6 years at CoreWeave), but it’s becoming pretty clear they prefer a conservative approach: underpromise and overdeliver.
I love that, and I think there could be a lot more upside surprises ahead.
Everything’s heading in the right direction, and there’s nothing more we could’ve asked for.
Why the company raised ARR guidance but not revenue:
“Of the 220 MW of connected power we expect to have at the end of the year, we will have 100 MW of active power. And as we are building out our data center capacity, most of our GPU installations will take place in Q4.”
Earnings Call Highlights
The earnings call was extremely bullish, at least in my opinion.
Peak Utilization
It opened with this key quote from Arkady Volozh, Founder & CEO of NBIS:
“We could grow faster, but we were oversold on all of our supply of previous-generation Hoppers, and we decided to wait for the new generation of GPUs to come.”
This means they were completely SOLD OUT of all H100 and H200 GPUs.
One of the bear cases I see floating around is the idea that older GPUs quickly become obsolete as new generations are released. But that misses a key shift in the industry: training is moving to the newest chips, while inference, which will drive most real-world demand, is increasingly run on older, broadly available GPUs.
Nebius is proving that with the right engineering, you can extract world-class inference performance from older chips. That’s why the company was recently ranked, in an independent study, as the fastest inference platform on public NVIDIA GPUs (meaning GPUs already generally available, not the very latest ones).
Of course, not every workload can be easily optimized for older hardware. Very large-scale or latency-critical inference will still benefit from newer chips like the Blackwells. But if Nebius keeps pushing the boundaries of efficiency on legacy hardware, it could become the go-to platform for maximizing the value of older GPUs, keeping them relevant far longer than most expected.
Very importantly:
“Pricing trends remain relatively stable for the Hoppers.”
It’s also worth noting that the company was at peak utilization overall:
“By the end of Q2, we were at peak utilization. The demand environment is very strong: as we brought on more capacity, we sold through it. If we had more capacity, we probably would have sold more as well.”
There’s no demand problem, only supply constraints. Things are just getting started.
“Looking ahead to our increased annualized run rate revenue guidance, a significant portion of it is already under contract, which gives us strong visibility.”
In other words, the updated guidance should be easily achieved.
Capacity Expansion
To support the aggressive ramp-up in capacity, Arkady announced two new data centers in the U.S.:
“We have nearly closed on two substantial new greenfield sites in the U.S. Each one will be able to deliver hundreds of megawatts of power in 2026, and we’re confident we’ll be making an announcement about that soon.”
We should expect more news on this front soon.
Nebius favors greenfield data center projects, building from the ground up, over build-to-suit or colocation options. This choice comes with multiple strategic and financial advantages.
By building greenfields, Nebius gains full control over design, construction, hardware installation, and phased deployment. This allows the company to match infrastructure development closely with actual demand, avoiding rigid timelines or long-term lease obligations.
Additionally, because Nebius handles everything in-house, from fiber and power integration to rack and server installation, it can operate at a lower total cost of ownership, estimated to be around 20% below the market average.
In short, greenfield development allows Nebius to fully capitalize on its competitive advantages.
UK Opportunity
Nebius is making a major push into the UK AI market, which it sees as one of the most promising globally. The company highlighted the UK as the third-largest AI market after the U.S. and China, and praised the government’s strong support for AI, including £14B in confirmed private sector investment.
As part of this expansion, Nebius is launching its first large-scale GPU cluster just outside London, expected to go live in early Q4 2025. The company believes it will be the first to bring NVIDIA B200s to the UK market, placing it at the forefront of enterprise and research AI infrastructure in the region.
One of the most exciting UK opportunities lies in healthcare and life sciences. Nebius has built a dedicated team for this vertical, based in the UK, and is working closely with NVIDIA to support the sector. Together, they plan to announce new initiatives aimed at accelerating innovation among life sciences startups, embedding Nebius deeply into the UK’s AI ecosystem.
The company has been highly focused on this segment, even launching the “AI Discovery by Nebius” event, an annual award recognizing startups using AI to revolutionize drug discovery, biotech, genomics, and healthtech.
Honestly, I wouldn’t be surprised if at some point, they decide to invest directly in startups innovating in this space. Time will tell.
Europe
Nebius also expects to benefit significantly from rising AI investment across Europe. With deep regional roots and strong connections, the company plans to expand its data center footprint throughout the continent. The goal is to remain one of Europe’s major AI infrastructure builders, leveraging both public and private partnerships to capture the growing demand.
Strengthening Go-To-Market Strategy
Nebius is expanding its go-to-market leadership and global sales operations to support its next growth phase. The company announced the hiring of a world-class VP of Sales Strategy & Operations, who starts this week, along with regional general managers to lead business efforts across the Americas, Middle East, Asia Pacific, and Japan.
Beyond geographic growth, Nebius is also building leadership dedicated to strategic customers and major enterprises, a clear sign of its intent to deepen relationships with large AI buyers and accelerate adoption in key verticals.
The company is boosting its customer-facing capacity and distribution strategy, focusing first on regional AI builders, targeted software vendors, and select enterprise segments. This approach helps Nebius deeply understand high-impact use cases and the full customer journey.
Looking ahead, Nebius plans to scale its global distribution and sales infrastructure, enabling it to meet the surging demand for AI infrastructure across all regions and industries.
Enterprise Customers
Another criticism Nebius has faced in recent months is that its clients were mostly AI-native startups, raising concerns that its business model was too risky since many of those startups might not survive.
But I’ve been saying this for a while: this is just the first wave of customers. Once the company has enough capacity, we’ll start seeing contracts from frontier AI labs (think OpenAI, Anthropic, etc.), hyperscalers, and enterprise customers, the last of which will likely drive the largest share of long-term demand in the AI infrastructure space.
Well, that’s already starting to happen.
Nebius added Cloudflare, Process, and Shopify as enterprise customers during the quarter, while still remaining the leading cloud provider for AI-native startups.
“We're very excited about these customers. They’re leaders in their categories, pushing the frontier of using AI to build and deliver great solutions.
Shopify is using Nebius' AI infrastructure along with Toloka's training data to optimize every step of the merchant journey. It's a very exciting opportunity for us.
Likewise, Cloudflare is using Nebius’ powerful inference at the edge, a key part of their overall offering.
Both relationships are growing and scaling opportunities for us. We’re also seeing similar interest from other major tech companies and category leaders, reinforcing the broader market opportunity.”
I can’t wait to hear more about these “major tech companies and category leaders” the team is referring to.
As for a large contract with a frontier AI lab or a hyperscaler, I think it’s pretty much guaranteed to happen as capacity scales up. Here are a few hints:
“Our mid-term guidance of several billion dollars in revenue does NOT factor in a large deal with a Frontier AI lab or a hyperscaler. So if, or when, they come, it will all be incremental and a nice surprise.”
“As we bring on larger GPU clusters, we are able to bring on new large customers who want to purchase greater and greater capacity. This allows us to expand and diversify our customer base and has been a clear signal that there is growing opportunity in the market.”
Once again, I’ve been saying this for a long time: it’s not a matter of if, it’s a matter of when. The demand is already there. We just need to wait for the company to finish building the required capacity.
Everything is going according to plan. Nebius is the AI cloud provider best suited to the level of customization and performance enterprise customers demand, and I have little doubt it will become a massive player in this space.
Building and Improving the Full-Stack AI Cloud
Nebius continues to invest heavily in its vertically integrated AI cloud platform, with major enhancements to both its training and inference software stack in Q2 2025. The company emphasized its focus on delivering speed, reliability, and flexibility, the three core demands from customers running large-scale AI workloads.
Key software improvements:
To support the upcoming deployment of NVIDIA’s Blackwell architecture, Nebius rolled out several improvements this quarter:
Network Speed Doubled: Resulting in direct performance gains in MLPerf benchmarks.
Reliability Upgraded: Through improvements to the core platform, auto-healing systems, and proactive health-check software, Nebius increased mean time between failures.
Seamless Data Flexibility: Customers using S3 storage can now easily migrate data for AI workloads on Nebius clusters, simplifying onboarding and reducing friction.
In the MLPerf 5.0 training benchmarks, Nebius delivered impressive results, successfully training a 405 billion parameter model. The company achieved near-linear scaling in cloud environments, with performance comparable to bare metal setups. That’s a huge win for customers who want cloud scalability without sacrificing speed or efficiency.
Expanding Across the AI Ecosystem
Nebius also expanded its reach through several high-impact partnerships aimed at simplifying AI deployment and improving developer experience:
Launched integrations with Mistral, Base 10, and SkyPilot to make its AI cloud more accessible across various workflows.
Partnered with Lightning AI and Anyscale, strengthening support for both open-source tools and enterprise use cases.
On the infrastructure side, Nebius expanded its portfolio with NVIDIA AI Enterprise and became a launch NCP partner for NVIDIA DGX Cloud Lepton, further anchoring its role as a high-performance AI platform.
The company measures the success of these partnerships by platform adoption, revenue contribution, and strategic access to new user segments, all of which are trending positively.
Regarding DGX Cloud Lepton, here’s what Roman Chernin had to say:
“We are one of the largest partners of NVIDIA there, and we see that it generates a significant pipeline of customers who start using NVIDIA Lepton and then continue directly with us.
In general, we think this partnership is a very good extension of the broader work we do with NVIDIA. It’s also part of our ongoing efforts to develop ecosystem partnerships, channel partnerships, and value-added partners that we mentioned earlier in this call.”
That said, I was curious to hear more from management about how they view Lepton in terms of long-term commoditization risk. While it may accelerate AI adoption in the short term, over the long run it could also drive commoditization of inference, which would hurt Nebius’ ability to differentiate from competitors.
My honest thoughts:
The commoditization of AI infrastructure is one of the most important long-term risks to monitor, and yes, DGX Cloud Lepton could play a role in that. But let’s be real: none of the parties involved (including NVIDIA) would benefit from such an outcome.
That’s why I don’t see this as a black-and-white issue. Companies like Nebius have welcomed Lepton’s launch, and there may be underlying mechanisms in place (just speculation here) to limit the downside risk.
So while it’s something to watch closely, I view it as a long-term concern, not an immediate threat, at least for now.
AI Studio: Enterprise-Grade Inference as a Service
Demand is rapidly shifting toward inference, and Nebius is responding by building the next layer of its AI stack: a fully integrated, enterprise-grade inference-as-a-service platform (Nebius AI Studio).
This offering is designed for AI-centric ISVs and enterprises, enabling them to deploy and scale open-weight models such as LLama, Qwen, Falcon, and even OpenAI's latest releases. The platform is built on Nebius’ proven infrastructure and targets the biggest production AI pain points: unpredictable latency, GPU bottlenecks, and lack of scalable, flexible platforms.
The result is a high-performance, dedicated infrastructure that enables customers to build, deploy, and scale AI applications reliably and efficiently.
Capital Raises
“We have raised over $4B in capital so far. We have a strong balance sheet, as you can see, and we have access to potentially billions of dollars more, thanks to our non-core businesses and other equity stakes, such as Avride, ClickHouse, and Toloka. In short, this is an exciting time for Nebius. We are in the midst of a once-in-a-generation opportunity, that’s what we believe in. The demand for AI compute is strong and will only get stronger. We are rapidly increasing our capacity to pave the way for accelerated growth in 2026 and beyond.”
“We have significant cash on hand and will approach any additional capital raising opportunistically, depending of course on timing and market conditions. At the moment, our focus is on securing land and power and moving quickly to reach our one gigawatt target.”
As I’ve been saying for a while now, ClickHouse will be an important source of non-dilutive capital in the mid-term:
“The way we're thinking about this stake is that there’s a lot of value to be created in the business. But if there were to be a liquidity event in the coming years at a significantly higher valuation, then that’s something we’d potentially consider as a source of several billion dollars.”
The same logic applies to Toloka and Avride, although personally, I’m much more excited about ClickHouse.
Here’s what they said about Toloka:
“This industry generally is a hot one. Scale AI, which is a comp for them, recently sold about half the company at a $30B valuation. So we think there is very significant upside to Toloka’s business prospects and valuation. What was important for us in that deal was that we retained a significant majority economic interest, so we feel like we have a lot of exposure to the upside, as and when we feel it's the right time to try and tap into that.”
We can expect several billion dollars in non-dilutive funding to gradually benefit Nebius over the coming years. That’s a huge advantage over peers and a major lever for creating shareholder value.
However, it's also important to consider other funding sources, which may involve some dilution. Personally, I'm not concerned about that. This team has tremendous capital efficiency, and I'm confident that any raise will result in net positive returns for shareholders in the long term.
Tariffs Impact
Nebius is closely monitoring the evolving situation around tariffs, including the most recent developments. The company noted that it's still too early to draw definitive conclusions, but emphasized that any policy changes would impact the entire industry, not just Nebius.
While acknowledging the risk of short-term fluctuations, Nebius remains confident that the market will adjust and stabilize. They’ve committed to providing updates as more clarity emerges.
Subsidiaries
ClickHouse and Toloka are no longer consolidated in the Group’s financials, so there were no updates about them this quarter.
Avride
Avride is positioned as an exciting and rapidly growing company within its industry. The business structure for Avride is expected to follow a partnership model similar to what was done with Toloka, where a strong partner takes a controlling role to help scale the business further.
Avride operates two main business lines:
Delivery Robots: Avride is expanding coverage with existing partners, adding new cities and service areas. Key collaborations include continued work with Uber and new projects with Grubhub at several university campuses. The company is also entering new verticals, such as grocery delivery, highlighted by a recent contract with retailer HEB in Texas, and indoor delivery operations in Japan via a partnership with Mitsu and FuruSan.
Autonomous Vehicles: On the AV side, Avride is growing its fleet and expanding road testing in Dallas in partnership with Hyundai. The company plans to launch its autonomous vehicle road test service with Uber later this year, following an early partnership agreement.
Overall, Avride is seen as a significant source of value for Nebius, with strong momentum and promising growth opportunities across both delivery and autonomous vehicle markets.
TripleTen
TripleTen continues to post strong growth, driven by:
~6,000 new learners per quarter
Increased average revenue per enrollment
Solid performance in core markets like the U.S. and Latin America
Customer acquisition remains efficient, powered by referral-based enrollments and strong sales productivity.
On the ops side, the company has leaned into AI tools to improve learning outcomes. Notably:
Project completion rates are up nearly 35%
Manual instruction has been significantly reduced
Expanded job support services: including earlier career prep, more externship opportunities, and dedicated CV placement specialists to improve interview conversion rates
TripleTen also launched a new AI Automation course in Q2.
It’s worth noting that Nebius still owns 100% of both Avride and TripleTen, though TripleTen remains a much smaller player compared to the other non-core assets.
Final Thoughts
All in all, this was a massive quarter from Nebius, far better than I expected.
Everything is moving in the right direction, and the company even raised its short-term projections. Honestly, we couldn’t ask for more.
Given how new this company is, it’s been difficult to make firm calls on execution or the quality of the management team, until now. This report reinforced what I’ve been saying for a while: this team operates with exceptional capital efficiency and a clear bias toward underpromising and overdelivering, and I absolutely love that.
I’ve had a lot of people ask if I’m trimming or selling my position, considering it’s already a large part of my portfolio and I’m up over 150%. My answer is simple: No, I’m not.
In my honest opinion, this is just the beginning of a fairy tale story. If the company delivers over 1 GW of capacity by the end of 2026, then today's valuation still looks ridiculously low. As I’ve explained multiple times, had this business gone public via a traditional IPO, it would likely be trading at 2–3x its current valuation. There are still many catalysts ahead, and I truly believe I’d regret selling even a small portion of my stake right now.
That said, and as I always remind you, this remains an investment with several risks to monitor. While I’m confident in the calculated risk I’m taking, make sure to do your own research, manage your own risk, and adapt your allocations based on your personal goals, situation, and tolerance.
Thank you so much for your continued support on this journey.
I can't express how much it means to me, and I’ll keep doing my best to provide you with valuable research.
Disclaimer: As of this writing, M. V. Cunha holds a position in Nebius Group (NBIS) at $25.67/share.
Don’t forget to subscribe for more content like this.






Excellent write up. I love reading your reports. Thank you
The comment of uk being the 3rd largest AI market other than USA China, is very interesting. I always thought Europe was a weak number 3, And worried that NBIS would not achieve as fast as it could have achieved due to the slow participation in AI. BUT I Didn’t realize that uk is the number 3. I think that will be a win win for uk and NBIS.