Why do you think @nebiusai released this piece of info? It's to tell the various countries that they are efficient in building data centers. Right now, governments are very concerned with these issues.. telling them, they have the scale and engineering knowledge. $nbis
$NBIS ENGINEERING MOAT: 🆒ing Edition🥶
5 key take aways:
- Up to 25x times less water usage per unit of power than the average US data center !
- Only 0.018 liters of water used per kWh !
- Nebius designs their own direct-to-chip liquid cooling systems !
- Nebius heats homes of local cities with their cooling system, resulting in elecricity savings for local cities, which also reduces emissions !
- Their system even saves electricity needs for their own primary heat-rejection solution (up to 60% of their dry coolers [water intake free])
---- They are best in class for a reason ----
Ray.NebiusMaxxing 🇸🇬 retweeted
NVIDIA Cosmos 3 Super With vLLM-Omni on Nebius | Cosmos Labs teapawt.net/i/broadcasts/1oKMvNPQy…
Look.. sovereign AI-native hyperscaler $nbis
Global-scale
$NBIS
Counting only sites that are running, being built or leased, Nebius is already in 8 countries:
🇺🇸 US: New Jersey, Pennsylvania, Missouri and Alabama
🇫🇮 Finland
🇫🇷 France
🇬🇧 UK
🇮🇸 Iceland
🇪🇪 Estonia
🇪🇸 Spain
🇮🇱 Israel
Signed, but earlier stage:
🇸🇦 Saudi Arabia: anchor customer for up to 50 MW at Oxagon from Q4 2027, under a preliminary framework
🇳🇱 Netherlands: letter of intent as first customer at Volt's 600 MW Rotterdam campus
And the hiring points at Asia: data center roles in India, a warehouse lead in Malaysia, and site selection roles in Singapore and Australia.
If the signed ones turn into sites, Nebius is in 10 countries. If Asia and Australia follow, it could be 14 across four continents. CoreWeave, the biggest neocloud, has data centers in the US, UK, Norway, Sweden, Spain and Denmark.
The spread matters because more and more AI buyers want compute inside their own borders: governments, banks, healthcare. That's the pitch behind Palantir naming Nebius its preferred sovereign AI infrastructure partner in September.
Ray.NebiusMaxxing 🇸🇬 retweeted
$NBIS
Key take aways from the $MSFT and $GOOGL earnings:
- Google Cloud revenue officially surged 103% YoY to 30.7 billion dollars ! (cloud made up 62% of the overall revenue growth this quarter‼️)
- Alphabet cloud operating income tripled to 11.1 billion dollars creating a massive 36% margin !
- Azure revenue growth accelerated to 48% reaching 33.2 billion dollars
- Microsoft operating income grew 21% which outpaces revenue growth and proves that AI investments expand margins
Translation: 🐂 $NBIS
Ray.NebiusMaxxing 🇸🇬 retweeted
Ever wondered what $NBIS Token Factory could be valued at if we carved out just that piece of $NBIS and valued it at competitor parity with players like Fireworks and Baseten?
Same managed inference, dedicated endpoints, post training, fine tuning, custom models and enterprise/agentic orchestration features but backed by Nebius’s own full stack AI cloud, their own physical infrastructure and rapidly expanding hardware capacity at that.
Let's start with the math shall we.
COMPETITOR BENCHMARKS:
- Fireworks: 40T+ tokens/day, $1B+ annualized revenue, $17.5B last closed valuation and reportedly discussing $30B - $40B.
- Baseten: approximately 40T - 50T tokens/day, reportedly $1B+ annualized revenue, $13B last closed valuation and reportedly approached at approximately $26B.
- Together AI: approximately 13T tokens/day, more than $1.15B in annual bookings and an $8.3B valuation.
Now compare that with $NBIS Token Factory.
The publicly visible number is only around 50B tokens/day through OpenRouter.
But Prosus alone has disclosed Token Factory workloads of up to 200B tokens/day through Nebius dedicated endpoints.
And this is where the public numbers can badly understate the real business.
Baseten shows roughly 368B tokens/day through OpenRouter, but says it processes 40T–50T tokens/day overall. (Based on Data recorded on 26th - 27th Oct)
That means OpenRouter may represent less than 1% of Baseten’s actual production volume.
If Nebius $NBIS has a similar public-to-private traffic ratio, its 51B visible tokens could imply roughly 3T-7T total tokens/day across serverless, dedicated endpoints and private enterprise workloads.
Again that is an estimate, not a disclosed Nebius figure but it means Token Factory may already be at 6%-15% of Fireworks/Baseten scale despite only launching in November 2025.
THE REVENUE MATH:
Fireworks produces $1B+ in annualized revenue from 40T+ daily tokens.
That implies blended monetization of approximately:
$1B ÷ 365 ÷ 40T × 1M
= approximately $0.068 per million processed tokens.
The formula is:
Annual revenue in billions
= 0.365 × daily trillions of tokens × net revenue per million tokens.
If $NBIS Token Factory reaches 45T tokens/day and monetizes at $0.08/M:
0.365 × 45 × $0.08
= approximately $1.31B ARR.
At 50T/day and $0.09/M:
0.365 × 50 × $0.09
= approximately $1.64B ARR.
Apply competitor-parity multiples of 20×-24×:
- $1.31B × 20-22× = $26B-$29B
- $1.64B × 22-24× = $36B-$39B
That gives Nebius $NBIS Token Factory a reasonable three year competitor parity valuation of approximately $30B-$40B.
So what should Token Factory be worth TODAY?
Take a $35B midpoint three-year parity value and discount it at 30% annually:
$35B ÷ 1.30 ^ 3
= approximately $15.9B today if competitor parity were guaranteed.
But it is not guaranteed, so I'll probability weight it:
- 60% probability of competitor parity: $35B
- 30% probability of partial success: $10B
- 10% probability of major underperformance: $2B
The calculation:
(60% × $35B + 30% × $10B + 10% × $2B) ÷ 1.30 ^ 3
= ~$11B today.
That gives me a, risk adjusted CURRENT standalone valuation with a slight bullish bias of:
$8B-$12B for Nebius $NBIS Token Factory.
My midpoint is approximately $10B–$11B.
Velocity matters. After a few of us devs explicitly pushed for public endpoint prompt caching on X, $NBIS Token Factory finally shipped it within months and the Open Router volume has gone instantly parabolic (scaling past 80B daily tokens just today). Get ready for $NBIS Token Factory to now take a massive chunk of public endpoint Open Weights Tokens processed from competitors like @FireworksAI_HQ & @togethercompute
Raw serving performance like Latency/TTFT and TBT or Tokens per Minute/Throughput was never the blocker for $NBIS Token Factory. Without public endpoints having input prompts with cached pricing user cost bottlenecks alongside Token Factory simply not being picked as a default routing option on routing platforms like Open Router, $NBIS Token Factory for sure had a massive impact to its true potential token volume as I have highlighted plenty of times before.
That single engineering friction point was holding back production grade volume from providers like Open router, Vercel AI Gateway and more. With $NBIS closing the feedback loop and delivering this as fast as they just did, the instant usage spike on Open Router to $NBIS Token Factory speaks for itself. Keep and eye on this chart as this will continue to balloon.
Software + Hardware Infrastructure execution doesn't get better than this. $NBIS
RT @AndersInvests: $NBIS Token Factory Valuation 📈
Conclusion:
"Risk adjusted CURRENT standalone valuation with a slight bullish bias:
-…
This quoted post is unavailable.
Ray.NebiusMaxxing 🇸🇬 retweeted
$NBIS director Arne Grimme keeps buying🔥
Last purchase made Monday @ $153.96
Bullish👀
Ray.NebiusMaxxing 🇸🇬 retweeted
All Material $NBIS quotes from $NVDA GTC
Demand for Next Year
"We have 1 GW worth of Veras coming next year, and already have significant demand for them."
"We're hearing from customers about what they need late in '27, early in '28."
Spot Pricing & Auction
"We want at least half our capacity to be allocated to today's needs [vs. long-term contracts]."
"A couple of hyperscalers asked us about our experiences [with spot pricing]."
On Raising Capital
"Capital is not a problem it's a task"
"We have a pretty healthy balance sheet, we also have been more conservative here."
"We are building a real company. Real company is measured by earnings. Healthy business is business that generates cash."
"Last week, some of capital sources temporarily closed like asset-backed or debt market. But we are working now with customer advances and through other instruments will probably provide us with enough financing to build 1GW at least"
Power
"We have more than 5GW+ reserved already"
Deployment Speed
"DSX POD, for complete NVIDIA Vera Rubin AI factories built by partners inside modular containers, reduce deployment time from six months to as little as 14 days"
"What used to take, in the case of Grace Blackwell, two hours to assemble one node [...] Now, Vera Rubin is 100% liquid cooled, and it takes less than five minutes to assemble it."
Bottlenecks
"GPUs will become the next bottleneck for the market. chips allocation is a concern."
Ray.NebiusMaxxing 🇸🇬 retweeted
$NBIS ENGINEERING MOAT: 🆒ing Edition🥶
5 key take aways:
- Up to 25x times less water usage per unit of power than the average US data center !
- Only 0.018 liters of water used per kWh !
- Nebius designs their own direct-to-chip liquid cooling systems !
- Nebius heats homes of local cities with their cooling system, resulting in elecricity savings for local cities, which also reduces emissions !
- Their system even saves electricity needs for their own primary heat-rejection solution (up to 60% of their dry coolers [water intake free])
---- They are best in class for a reason ----
"The direct-to-chip liquid cooling systems we design"
👀
Proprietary cooling system is just a hint of the engineering DNA $NBIS possesses.
Ray.NebiusMaxxing 🇸🇬 retweeted
I cannot overstate just how bullish this is for data centers that haven't sold all of their capacity yet
Ray.NebiusMaxxing 🇸🇬 retweeted
$NBIS
I agree 100% with Jon. Demand isn’t the problem. Delivering it is: capital, power, data centers and chips.
For the neoclouds that make it through, he says compute pricing is up 50%, older chips are worth more than people thought, and their data centers are leased below market.
So the question is who has the money to get there.
Nebius borrows at about half of what CoreWeave pays, has most of its debt in low-coupon convertibles, and gets customers to prepay much of the capex.
Same tailwind, very different paths to the other side.
Blackstone’s Jon Gray says neoclouds that can get to the other side will find compute pricing up 50% and legacy chips worth more than people thought, and the market is confusing deliverability with demand
“I think what we’re seeing in the market is a little bit of a conflation between deliverability of compute and demand for compute. And at our companies, as an example, we’ve seen a 20-fold increase in spend on models over the last 12 months. It’s pretty extraordinary."
"We have a joint venture with Anthropic to deploy these models. The company’s called Ode, and it can’t keep up with the business that’s coming in the door. So we’re seeing demand really growing for compute."
"And at the same time, the constraint, of course, is delivering that supply. And that’s hard because of amassing capital, data centers, power, chips, all of that."
"And so for the neoclouds, if they can get to the other side, the pricing of compute’s up 50%. Their legacy chips are worth much more than people thought. Their data centers are leased below market."
"Their challenge is, do they have the capital, the debt and equity and the quantum that’s needed to deliver?"
"I think some of these companies have very talented CEOs. CoreWeave’s a good example of this. They’ll find a way through it, but that’s the challenge they face. The underlying business is actually quite strong.”
Ray.NebiusMaxxing 🇸🇬 retweeted
Something interesting is happening between $NBIS and $NVDA. They are increasingly working hand in hand to optimize NVIDIA’s AI software to run faster, more efficiently, and at lower cost on NVIDIA GPUs and Nebius AI Cloud.
NVIDIA is now promoting their joint work on Cosmos 3 Super, a world model used to generate synthetic video for training robots and autonomous systems.
In this experiment, Nebius engineers managed to get ~43% more video output from the same GPUs, reducing the compute cost per second of video by 30%.
And this is huge, because video generation is directionally 100× more compute-intensive than text generation in language models. So even relatively small efficiency gains can translate into enormous savings at scale. And these gains here are by no means small, which to me shows just how early we still are in making compute efficient at scale for physical AI.
I wrote earlier that I think NVIDIA has been moving more and more toward the application layer of AI lately, especially in robotics and physical AI.
And every time one of its software platforms takes off, it allows a lot more people to build things they otherwise couldn’t, because they no longer have to build everything from scratch. This brings in more builders, more use cases, and ultimately more demand for NVIDIA’s compute. Just look at Cosmos and Alpamayo, NVIDIA’s platforms for building intelligent robots and self-driving cars.
And it looks like Nebius is positioning itself to benefit from all this in two ways.
First, directly, by helping NVIDIA build and improve these AI applications, making its own cloud the number one place to run NVIDIA’s AI software on NVIDIA’s hardware. The entire stack would be perfectly in sync, from the hardware at the bottom, through cloud infrastructure, all the way to the AI application software at the top, giving builders maximum value for every dollar they spend.
And second, indirectly, by becoming a closer engineering partner to NVIDIA and helping them scale their applications, which in turn might open some important doors down the road.
That second part could be particularly valuable when it comes to getting access to the latest GPUs, technology, customers, and capital. And I think the GPU part could become especially interesting a year or two from now. More and more people in the industry are saying that the next big bottleneck might not be securing power, but securing enough compute to actually use it.
So it’s really fascinating to watch this Nebius–NVIDIA partnership keep expanding well beyond GPUs, with Nebius derisking its investment thesis in almost every way imaginable.
This is one more example of why I keep saying Nebius is becoming NVIDIA’s concierge for AI.
I think people are seriously underestimating how big Tavily could become for $NBIS especially as AI agents continue to take off.
Tavily just announced that it surpassed 3 million developers, up from 1 million in January and 2 million in June. That's 3X growth in just 10 months, which is pretty impressive considering how early we still are in the adoption of autonomous AI agents. For those unfamiliar, Tavily essentially gives AI agents access to the live internet. It allows them to search the web, extract information, gather real time data, and use that information to complete tasks without constant human involvement. And as AI agents become more advanced, the demand for this type of technology could increase significantly. Unlike humans who might perform a handful of searches, an autonomous agent could potentially make hundreds or thousands of requests while completing complex tasks.
But here's why I think this is particularly bullish for Nebius. The company acquired Tavily to expand beyond simply renting out GPUs. Nebius already provides the computing infrastructure, Token Factory handles AI inference, and Tavily provides the real time information agents need to operate. Essentially, Nebius is building an ecosystem where developers can access multiple layers of AI infrastructure through one provider. What's even more interesting is the potential for recurring software revenue. While Nebius's core infrastructure business is capital intensive, Tavily gives the company exposure to usage based software revenue as more developers integrate its search capabilities into their applications. Obviously, 3 million developers doesn't mean 3 million paying customers, and we still need to see how much revenue Tavily can generate. But I think the bigger picture here is that Nebius is positioning itself to benefit from AI adoption across multiple layers of the stack and that's one of the reasons I'm bullish on $NBIS over the long term.
ALT Tavily by Nebius graphic: Now trusted by 3,000,000+ developers worldwide, with a globe
Ray.NebiusMaxxing 🇸🇬 retweeted
$NBIS: FUD vs REALITY
FUD: No more debt financing is available — the credit window for neoclouds is shut.
Reality: Within days of this doom FUD Nebius announced a $3.5 billion senior secured facility from Citi (364-day bridge, starting at Term SOFR + 2.75%, capped at +3.50%). This follows a $775 million GPU- and contract-backed facility (SOFR + 2.50% to 2030). And yesterday a new €250M loan was secured for Bethune France. The company cites more than $40 billion in long-term contracts with investment-grade customers (including Microsoft and Meta) that can support additional asset-backed financing, plus substantial customer prepayments that cover a large share of related capex. So the reality is financing access remains open on attractive terms.
FUD: This is an AI bubble — demand will collapse and GPU prices will crash.
Reality: GPU rental prices keep rising. Nebius raised on-demand rates effective October 1 (H100 roughly +17% to $4.50/hr, H200 +20%, B200/B300 up similarly). Its CRO has said prices are rising faster than expected, 2027 capacity would sell out if brought online, and a similar dynamic could extend into 2028. Older chips (A100s, H100s) are pricing above levels from when they were new. Elon Musk / SpaceX has described a lasting shortage of superintelligence compute, with rivals wanting capacity, strong pricing power, and aggressive buildouts (targeting roughly 13 GW of compute by the end of 2027). Demand is not cooling.
FUD: AI infrastructure isn’t profitable — it’s just endless cash burn.
Reality: In Q2 2026, Nebius’s core AI cloud business (≈98% of revenue) delivered an adjusted EBITDA margin of ≈50% (49.7%, or ~$286 million on $575 million of AI cloud revenue). Group adjusted EBITDA was $236 million on $582 million of revenue (~41% margin), up sharply from a loss a year earlier. Revenue grew 454% year-over-year, ARR reached $3 billion, and management reaffirmed guidance for ~40% group adjusted EBITDA margins, $3–3.4 billion full-year revenue, and $7–9 billion year-end ARR. Pricing and utilization are improving, not deteriorating.
FUD: Customer concentration and hyperscaler competition will kill the model.
Reality: Large contracts with investment-grade customers (Microsoft, Meta, and others) provide multi-year visibility and prepayments that de-risk buildouts. Nebius is also signing billions worth of deals across other major AI labs, quant firms, and enterprises while expanding its software/stack offerings. Hyperscalers are simultaneously capacity-constrained and still renting externally, so the shortage narrative from SpaceX and others reinforces that external supply remains valuable.
FUD: Capex is unsustainable and will force massive dilution or distress.
Reality: Capex is large ($20–25 billion guided for 2026) because demand is pulling capacity forward. The market got spooked by Coreweave 10X1 debt to cash balance but Nebius balance sheet is roughly 1X debt to cash! Nebius funding is diversified across customer prepayments (covering 50–60% of related capex on many deals), operating cash flow, convertible notes, and the new secured facilities. The company has repeatedly demonstrated the ability to raise capital on non-dilutive or attractively structured terms while maintaining a substantial cash position. Nebius keeps building the right way by respecting the balance sheet.
Bottom line: The dominant FUD narratives treat Nebius like a pure high-risk story with closed funding markets, evaporating demand, and structurally unprofitable economics. But the operating and financing REALITY shows the opposite… a company with strong core-cloud margins (~50% adjusted EBITDA), rising prices, sold-out or sellable capacity, multi-billion-dollar contracted backlog, elite clients/partners including Meta, Microsoft Nvidia and Palantir and continued access to debt and prepayment funding.
NFA - Always Do Your Own Research