Ivan

[Ivan @Yarless_]

@jiahanjimliu @GavinSBaker If labs start to monopolize the market and design their own chips then they will become a bigger concern. Anyway helping neoclouds with good open-source models is in Nvidia interest.

[Ivan @Yarless_]

@jiahanjimliu Waymo is way ahead of Tesla with 10k autonomous cars without safety drivers with little incidents. Tesla is not even close.

[Ivan @Yarless_]

@FransBakker9812 It’s developed by Polarnode, not Nebius.

[Ivan @Yarless_]

@ilzmcfly Missouri, Alabama and Pennsylvania use BTM? What are your sources?

[Ivan @Yarless_]

@RealJimChanos @capocapuccino Do you include Starship RnD? Falcon 9 should be profitable.

[Ivan @Yarless_]

@Dr_Crossroads Coreweave don’t build their own datacenters and pay colocation fees, so they would have lower margins than both Iren and Nebius

[Ivan @Yarless_]

Nebius acquired both Clarify and Eigen because each team is better at one part of the process, it doesn’t seem they just outsourcing the process to one team or even two separate teams competing on the full-stack inference offering. So I don’t think it’s fair to say that strength of Nebius + Clarify + Eigen is equivalent to the strongest of the three. There should be some post-acquisition metrics.

[Ivan @Yarless_]

@jiahanjimliu I see your reasoning, it is legit. But they could acquire those startups not only because they were years behind Fireworks, but because they saw very high roi from those acquisition. Nvidia acquired Melanox not because they were behind AMD.

[Ivan @Yarless_]

@jiahanjimliu “Nebius launched Token Factory in Nov 2025, so IREN would be about 1-1.5 years behind” - why not “Nebius launched their inference offering in September 2024…”? I’m about that passage

[Ivan @Yarless_]

@jiahanjimliu Ok, but in your original post you said that Nebius just launched token factory in November 2025 and somehow September 2024 is not a right date for its inference offering. Why?

[Ivan @Yarless_]

@jiahanjimliu And can you elaborate why do you think Nebius partnership with hugging face changed betweeb ai studio and token factory? It is still an inference provider inside hugging face.

[Ivan @Yarless_]

I don't think he is dumb or he hates Nebius. I think he genuinely believes his thesis, you don't put 95% of your capital on a bet you don't believe in. But I think after one is all in one stock it affects one's judgement. Both stocks will be ok for the next 12-24 months if market expectations of ai revenue growth will stay at the current levels. Both stocks are risky bets because markets expectations can cool down anytime, but I think Nebius is more diversified, margin advantage on one datacenter site is not enough to say that Iren is a better investment than Nebius even with 50% discount. Enterprises are ready to pay 2.5x rate to AWS/Google Cloud/Azure compared to Coreweave/Nebius, so there is no way higher software margins wither in the next 2 years.

[Ivan @Yarless_]

@jiahanjimliu @jeremymjacquin Scarcity short-term, till eoy 2027 for sure, anywhere long-term - probably 2030 and later. So premium margins on software till 2030, base margins on best software after that; base margins for bare-metal till 2030, colocation-level margins after that for bare-metal.

[Ivan @Yarless_]

So then I'm not getting your point. Are you saying that Iren would show superior profitability in eoy 2027? How much so? Because Nebius would have a mix of Microsoft contract, other colocation and their greenfield sites with tax breaks then. Plus they will have software premiums on around 40% of their revenue, even if you believe that long-term all leading enterprises would vibe-code their own kubernetes and managed inference platforms, they won't be able to do that in the scarcity market. By the way, I think long-term kubernetes and managed-inference margins would compress to the current bare-metal margins levels and enteprises would choose clouds with superior software, because they can get just gpus anywhere. And bare-metal margins are manageble to them. So both short-term (2-3 years) and medium-term (3-5) years Nebius is most likely to have better margin profile. I agree that in the next 12 months Iren can have a better margin profile because it would commission its first large batches of gpus and Nebius would only start to put up its greenfield sites online.

[Ivan @Yarless_]

@jiahanjimliu @jeremymjacquin Cost calculations are pretty worthless if you don’t consider Microsoft contract terms during first 5 years and next generation of gpus after 5-10 years which will increase revenue per mw. It may be both costlier than Iren and be pretty profitable.

[Ivan @Yarless_]

@EndicottInvests With dilution there are 325m shares, so around 300 for 100B market cap.

[Ivan @Yarless_]

@AlexfromBabylon Iren still has more secured US greenfield sites, so that alone couldn't convince him.