@Krokodil_V Profile picture

Krokodil Value

@Krokodil_V

Fundamental investor. Generalist but like tech, semis and healthcare.

Similar User
Olle Östergren photo

@ormenhairullah

X photo

@_anon_888_

sokarsokk👁️ photo

@sokarsokk

Bobby Jackson photo

@RobertJ33640650

even cheung photo

@Even12956Even

Peter Gutzwiller photo

@thegutz

StY photo

@rammsteinY0128

Evolution photo

@Evolut777

Krokodil Value Reposted

$VKTX coming in hot with the #Vikingtide data!

Tweet Image 1
Tweet Image 2

I mean, nice marketing and good improvement from $INTC but note they're using a 14 inch machine to compare to a 16 inch $AMD machine...

Tweet Image 1

The FTC can be overzealous but this is a rule I can get behind (although questions on reporting / enforcing).

1. People rely on online reviews to make informed purchases and compare products. When companies use fake reviews, consumers get misled and honest businesses lose out. Today @FTC finalized a rule to prohibit fake reviews. ftc.gov/news-events/ne…



Nice guidance bump by $HUBS. Don’t love the decel in net adds. Offering 25% discounts + flexibility on seats, yet still expecting subs to grow at a 1-2k lower pace for 2H24. Likely doesn’t matter LT, but still.


$AMZN seeing "significant demand" for Trainium and Inferentia, which are "very compelling on price performance" $MRVL $NVDA

Tweet Image 1

Krokodil Value Reposted

The trick to doing AI inference is to buy as few GPUs as possible to hold the model weights. Here’s some fun comparing Llama 3.1 405B inference on AMD and Nvidia GPUs. $AMD $NVDA nextplatform.com/2024/07/29/sta…


Further commentary from Amy Hood implies $MSFT CPU capex continues to be throttled. Nonetheless, $AMD DC CPU reaping market share gains off of strong TCO / performance and "some" return to growth.

Tweet Image 1

$MSFT CFO Amy Hood commentary re capex confirms flexibility of infrastructure to meet external / internal demand.

Tweet Image 1

Krokodil Value Reposted

$AAPL Intelligence Foundational Language Model (AFM) paper. AFM server trained on 8192 TPUv4 chips. AFM-on-device ~ 3B parameter model trained on 2048 TPUv5p chips. All $GOOGL $AVGO. Training, dataset, memory budget, usual eval discussions etc in paper. machinelearning.apple.com/papers/apple_i…

Tweet Image 1
Tweet Image 2

Another validation of custom silicon: $AAPL used 8,192 $GOOGL / $AVGO TPUs to train its foundational models.

Tweet Image 1

$LULU looks cheap but tough to underwrite when competition appears to be gaining steam and domestic growth is slow + could slow further. International (particularly China) is a big growth opportunity but still too small to ignore US/Canada situation.


Krokodil Value Reposted

$VKTX In a nutshell 3 big important updates from the call - Straight ph3 program for subq for obesity. It saves money and quicker to market. Puts the company in a driver’s seat for acquisition - No safety issues with 80mg oral cohort - Monthly dosing: Advantage 💥💥💥💥💥


Odd that $AVGO reaction negative even after $GOOGL increased capex and continues to expect quarterly $12B capex floor. $GOOGL comments on potential overinvestment and also trying to run their DC footprint more efficiently fully leans into custom silicon.


Krokodil Value Reposted

Y2K24 - fuckload Windows Machines are absolutely fucked. Crowdstrike $crdw down 19% premarket They pushed out a buggy update (.sys files are kernel drivers, Crowdstrike’s agent lives in the kernel) People’s computers crashed It also fucks up booting into loading All affected…

Note this will not work if your machine is bitlocker encrypted without getting the recovery key for each machine...



Re $ASML - is the "geopolitical" selloff based on fears that the crackdown will extend beyond the already impacted EUV / DUV chain? The BBG article notes the goal is to crack down on servicing and repairing restricted gear currently in China. Not much of an escalation to me?

Tweet Image 1

Loading...

Something went wrong.


Something went wrong.