Raman Bindlish

Macro economics and market analysis

CIO
Proflex Finance
Elevate your Investment Quotient!!!

[Raman Bindlish @Raman_Macro]

While airlines comparison feels apt when looking at economics, airlines can not create demand like AI can do.

There are only so many people who would fly even if it was cheaper or better experience.

So with all due respect to Ken Griffin, I think narratives are being conflated.

[Chamath Palihapitiya @chamath]

Super interesting comparison of GPU finance and the AI buildout to airlines. I’ve seen two other people make this comparison now:

1. Ken Griffin
2. Sridhar Ramaswamy

Both are very, very smart so worth learning more.

[Raman Bindlish @Raman_Macro]

@lukasgruber__ What are the revenues they posting?

[Raman Bindlish @Raman_Macro]

@JohnTinsmanAOT Asian conglomerates like to pay themselves a fat check in dividend rather than buying stock back only.

The buyback is better narrative is mainly a US thing with shareholder returns kept as first priority and was made even more popular by likes of Berkshire pushing for same.

[Raman Bindlish @Raman_Macro]

So we went pretty fast from

“Chinese CXMT is coming for US/Korean suppliers”

to

“Chinese CSPs want Samsung and SK Hynix to do 10 year contracts for memory to ensure they get enough”

$SKHY $MU

[Raman Bindlish @Raman_Macro]

@HomoSapienLCY @chontang Yes, every tech company should focus on minting existing tech for profitability and never invest in developing future tech. That would be perfect recipe for making money... Hilarious!!!

[Raman Bindlish @Raman_Macro]

@chontang Because you started from Deepseek tokens which is again tiny. Much of paid usage is driven by Opus / Sol.

And again 75% if compute is being used today for training and not for inference.

[Raman Bindlish @Raman_Macro]

$SKHY says the prices keep going up while supply is increasing.

Customers are willing to sign 5 year contract for a commodity play which is supposed to be disrupted by Chinese memory next year.

Bear narrative is falling apart on semi trade!!!

$MU $NVDA $AVGO $AMD

[Raman Bindlish @Raman_Macro]

Did you understand what he meant? He is referencing value of AI tasks... It is like saying you can use you expensive macbook to play online lottery or build trillion dollar empire.

and also he is telling you how the cost to run a repeated query is negligible for them while user finds same value.

[Raman Bindlish @Raman_Macro]

@jimcramer Would you recommend Jensen to buy treasury bonds if yields cross 6%? What kind of argument is this?

Junk bonds trade higher when yields are high. This is normal. CRWV issued coupon of 9%+ and it was never the senior most loan on their assets anyway.

[Raman Bindlish @Raman_Macro]

Add to that the amount of money that is already raised or approved by financing and not spent yet.

Are shorts just counting on political opposition to AI now because we do not like data centers in neighborhoods? Is that hope for bears at these valuations in AI trade?

[Andreas Steno Larsen @AndreasSteno]

The token demand is going up

The price of AI infrastructure is going up

And Anthropic proved that they can make money on selling inference

What's not to like, really?

[Raman Bindlish @Raman_Macro]

Bears are running the narrative recently for AI trade. But what does actual data tell you about demand?

GPU rent trackers show that rents are going up as we speak.

And the AI spenders are flush with cash.

So why so much pessimism? #AI $NVDA $AVGO $MU

[Raman Bindlish @Raman_Macro]

@HomoSapienLCY @rdd147 It is 4-5x from Blackwell to Vera Rubin (inference)

But we are comparing historical rates which was mostly H100 available in early 2026. Blackwell got lit up with SpaceX deal and that is why Anthropic paid much higher price.

[Raman Bindlish @Raman_Macro]

@Lifebass80 @rdd147 Each generation of GPU is improving inference per MW at much faster rate. It will be of the order of 20-30x improvement without counting the SW efficiency and cached responses that are even cheaper irrespective of HW used.

[Raman Bindlish @Raman_Macro]

@rdd147 AI is good at teaching fundamentals of how it works - you can ask Claude itself on how to run agentic AI and what is limitation of H100 (if any). Or for that matter, what is the meaning of agentic AI and how tokens are produced.

[Raman Bindlish @Raman_Macro]

For those who have a take on Anthropic S1 - do me a favor and simply just put S1 in Muse, Grok, or Claude (free version if you hate paying) and it will clear the confusion. And then decide to influence people online... Too many misinformed influencers with mostly bad takes.

[Raman Bindlish @Raman_Macro]

@GavinSBaker @rdd147 Bears have made up their mind. They don't understand that software can become more efficient and old GPU productivity can improve for future models.

And then we have this whole architecture level shift where new GPU class provides generational shift in cost per task.

[Raman Bindlish @Raman_Macro]

@rdd147 So let me try to agree with you. So Anthropic hiding compute spend and it was not reported in S1 as per GAAP requirements.

$AMZN and $GOOG did accounting fraud for Anthropic
$GS and $MS are collaborating on this fraud
And auditor Ernst & Young is also on board

Hilarious...

[Raman Bindlish @Raman_Macro]

While market is digesting Anthropic S1 and making AI trade decisions, here is a quick summary I put together of what is happening in parallel universe of open weight models and the growth of AI usage.

And do note the quality of Muse also as you go through the report and tool that was put together in the report for cost analysis.

Muse is free and still Anthropic paid usage is exploding higher.

#anthropicAI #muse #ai

https://t.co/IIZ2juf1qT