Sorry I missed this.When it comes to hyperscalers, I have to start by saying that I do not trust the numbers. If I look closely at the data, I want to see diversity in the companies signing contracts, and I would prefer to see sustainable two- or three-year agreements. My overall feeling about the industry is that continuous advancements in chips, compute, and memory will inevitably drive the efficiency of compute, power, and costs downward.
Furthermore, I see the current demand as inflated. Working on the advanced aspects of AI myself, I can see that the narrative being told to the public and the reality on the ground are far apart. I firmly believe in the potential of AI—I just don't believe the story currently being portrayed by the major players.
Your point on demand being inflated is the variable the filing cannot fully answer.
We called Nebius Needs Proof specifically because contract duration and customer diversity are unverified. If demand is more cyclical than the narrative suggests, those two data points become everything. Long-term agreements with a diversifying customer base means Nebius locked in a durable window. Short-term or Meta-concentrated contracts means 684 percent revenue was timing, not positioning.
What are you seeing on the ground that tells you the gap between the public narrative and reality is wider than most people think? That context would directly inform which risk matters more here.
From inside the industry, the picture is more nuanced than the mainstream narrative suggests: today’s AI infrastructure boom is driven less by broad enterprise adoption and more by temporary GPU scarcity and concentrated hyperscaler demand, with Meta and Microsoft alone representing the majority of Nebius’s backlog. Even leading AI companies are scaling back usage of certain models due to high inference costs, underscoring the economic friction normal enterprises face.
At the same time, current‑generation models still struggle with reliability, security, and production stability, and outside of coding‑adjacent workflows, AI has not replaced workers at the rate headlines imply : in most domains, human performance remains stronger and more consistent. The result is an 80% project failure rate and a market growing quickly but unevenly, constrained by real technical and economic limitations that investors should understand when evaluating long term sustainability.
In short, today’s AI infrastructure boom is driven by scarcity rather than durable demand, and the economics begin to break down well before true mass adoption. We are still early in this AI game and we face growing pains.I am a firm believer in the AI long game but not in the 12 to 24 months narrative.
That is exactly where the Firewall landed. Concentration risk and scarcity versus durable demand. Meta and Microsoft making up the majority of that backlog means two buyers have to stay. Scarcity-driven demand means customers using Nebius because there is nowhere else right now. Same revenue number. Very different business.
When supply normalizes those two situations diverge fast.
Your 12 to 24 month framing is the right lens. The question is not whether AI matters long term. It is whether Nebius locked in durable revenue before the scarcity premium fades. Contract duration is what tells you.
From inside the industry, which companies do you think have actually built something customers would still pay for if GPU supply doubled tomorrow? That is the one we want to find.
That is the question: will demand last? I will study the company further. This provides great insight into asking the right questions
That is the whole game, right question first. Would love to hear what you find. Contract duration is the one to watch."
Sorry I missed this.When it comes to hyperscalers, I have to start by saying that I do not trust the numbers. If I look closely at the data, I want to see diversity in the companies signing contracts, and I would prefer to see sustainable two- or three-year agreements. My overall feeling about the industry is that continuous advancements in chips, compute, and memory will inevitably drive the efficiency of compute, power, and costs downward.
Furthermore, I see the current demand as inflated. Working on the advanced aspects of AI myself, I can see that the narrative being told to the public and the reality on the ground are far apart. I firmly believe in the potential of AI—I just don't believe the story currently being portrayed by the major players.
Your point on demand being inflated is the variable the filing cannot fully answer.
We called Nebius Needs Proof specifically because contract duration and customer diversity are unverified. If demand is more cyclical than the narrative suggests, those two data points become everything. Long-term agreements with a diversifying customer base means Nebius locked in a durable window. Short-term or Meta-concentrated contracts means 684 percent revenue was timing, not positioning.
What are you seeing on the ground that tells you the gap between the public narrative and reality is wider than most people think? That context would directly inform which risk matters more here.
From inside the industry, the picture is more nuanced than the mainstream narrative suggests: today’s AI infrastructure boom is driven less by broad enterprise adoption and more by temporary GPU scarcity and concentrated hyperscaler demand, with Meta and Microsoft alone representing the majority of Nebius’s backlog. Even leading AI companies are scaling back usage of certain models due to high inference costs, underscoring the economic friction normal enterprises face.
At the same time, current‑generation models still struggle with reliability, security, and production stability, and outside of coding‑adjacent workflows, AI has not replaced workers at the rate headlines imply : in most domains, human performance remains stronger and more consistent. The result is an 80% project failure rate and a market growing quickly but unevenly, constrained by real technical and economic limitations that investors should understand when evaluating long term sustainability.
In short, today’s AI infrastructure boom is driven by scarcity rather than durable demand, and the economics begin to break down well before true mass adoption. We are still early in this AI game and we face growing pains.I am a firm believer in the AI long game but not in the 12 to 24 months narrative.
That is exactly where the Firewall landed. Concentration risk and scarcity versus durable demand. Meta and Microsoft making up the majority of that backlog means two buyers have to stay. Scarcity-driven demand means customers using Nebius because there is nowhere else right now. Same revenue number. Very different business.
When supply normalizes those two situations diverge fast.
Your 12 to 24 month framing is the right lens. The question is not whether AI matters long term. It is whether Nebius locked in durable revenue before the scarcity premium fades. Contract duration is what tells you.
From inside the industry, which companies do you think have actually built something customers would still pay for if GPU supply doubled tomorrow? That is the one we want to find.