bernie

Credit backstops to finance GPUs faster - and

[bernie @berniebiased]

Credit is becoming the largest bottleneck for compute.

Need GPUs? I wrote a GPU Finance 101 guide for operators here:

https://t.co/7eFZIjmrMq

Specific lenders, terms, downloadable spreadsheets, and tricks are all here.

[bernie @berniebiased]

@sur4js once we hit $1 billion financed!

[bernie @berniebiased]

i left Corgi

This is my Corgi tattoo from when I was there (long story). Crazy times.

One year since I exited that role.

[bernie @berniebiased]

@5631mega Ecosystem is still driven by private deals, not transparent at all. There's a million different price points on everything: why does XYZ charge more than ABC? Brokering opportunities are across hardware, colocation, compute, financing, etc.

[bernie @berniebiased]

many people ask me how to get into the compute game and what kind of products to work on

my suggestion is to just become a broker. build your network, understand how it works.

right now you can make millions just being a broker - stop being silly and overly clever

[bernie @berniebiased]

@yikesawjeez “claude read the link pinned to bernie’s profile that gives a crash course on compute finance to figure out ways to broker and make no mistakes”

i’m actually not a compute broker, but i help w the finance piece. my article still explains everything end to end

[bernie @berniebiased]

@distributionat Just need a cool $7-8m per rack…

[bernie @berniebiased]

3/

Interested in learning more?

Message me or check out our Rack Report where this came from.

Download it here: amcompute dot com

[bernie @berniebiased]

Our data shows used server CPUs at 49 for Intel and 72 for AMD on our relative resale index.

That makes CPUs the weakest component in our report, at 49% below the depreciation baseline.

Softness is likely from older 2019-gen refresh processors exiting service windows.

[bernie @berniebiased]

2/

AMD ran above 100 in January 2026 before sliding back to 72.

The basket covers 55 models, 3-8 years old, and transactions from all the way back in July, 1.8x a year earlier.

[bernie @berniebiased]

It’s funny how people larping as tech founders talk about moving to SF and don’t realize SF isn’t actually a part of Silicon Valley

[bernie @berniebiased]

@andrawesbahou @Architect_Fi They are mid-market lenders. They are not interested in selling compute or selling futures. They don't want to learn or figure any of that stuff out honestly... they don't do it for any existing asset class either.

[bernie @berniebiased]

Here's why I struggle to use GPU Rental Indices for my residual value insurance structures

[Where Capital Meets Compute @WCMC_Podcast]

Bernie Margulies @berniebiased, founder of American Compute @amcompute1, on why his firm doesn't trust GPU rental price indices to predict resale value.

He ran the correlation study himself: resale prices in one period, rental prices in the same period. "They literally went in different directions."

His preference: even just four real resale transactions in the secondary market tell him more than a rental index does.

Full episode available now.

[bernie @berniebiased]

I think there’s a diff because I’m focused on specific real world situations, you’re thinking industry level.

My clients care exactly about that bankruptcy scenario, of course they would be happy with that modeling based off of the bankruptcy FLVs. That’s a great avg/floor for them to aim for.

[bernie @berniebiased]

@SMASIMHO @0xZergs Might be a 2nd generation asian thing, my Korean relatives might have an "English name" but it's not on any documents

[bernie @berniebiased]

I stand by it. Hardware resale price data are a better predictor of hardware resale prices than rental rate data.

This is specific to residual value.

Asset managers for aviation, car fleets, and other assets don't sell a used plane based on the profit per trip. They look at recent used plane transactions as comparables.

[bernie @berniebiased]

@stevehou i mean the original equipment finance and private credit lenders doing GPUs the most

[bernie @berniebiased]

I have heard from MANY sources that the existing compute financiers (non banks) are running out of dry powder.

They don’t have cash left to deploy!

Credit is the largest bottleneck for compute.

We need new pools of institutional capital coming in if we don’t want the music to stop.

EDIT - Traditional banks are starting to come in, they just started

[Meltem Demirors @Melt_Dem]

hearing banks stopping all compute lending

credit crunch is beginning

[bernie @berniebiased]

the original players are still writing checks, but a lot more selectively.

they’re hitting “portfolio exposure limits” - they write less than 5% of the inbound they get.

their existing LPs aren’t giving them more powder for compute. I know cause they want my insurance structures to get around it

the existing lenders who first started deploying capital 2-4 years ago are catching their breath

i’m not saying the industry is collapsing at all but we’re prolly gonna start seeing more types of LPs come in, the “easy money” is already tapped out

[bernie @berniebiased]

@tedbibr @RMladek and @SemiAnalysis_ !

[bernie @berniebiased]

@EzraMoGee "And zero tolerance for poor integrity"

Lol