Ari Tanner

[Ari Tanner @Arzca83]

This guy is a diamond. Give him a follow!

[mon @moninvestor]

Anthropic's Fable 5 launch confirmed what I have been arguing for over a year now. There is simply not enough compute in the world to meet demand.

Keep in mind that Fable 5 and GPT-5.5 are not the end state. New models are shipping every few weeks, each one needing more computing power than the last, and the models coming in 2, 5, and 10 years will need vastly more than anything running today.

Way to invest? Own the supply chain, layer by layer.

Layer 1 is the chips. These are the processors that actually run AI.

> $NVDA is the leader and sells the most AI chips in the world.
> $AMD is its biggest competitor and is selling more AI chips every quarter.
> $AVGO designs custom AI chips for companies like Google that want their own alternative to Nvidia.
> $TSM is the factory that physically manufactures nearly all of these chips.
> $MU makes the high speed memory that gets attached to every one of them.

Layer 2 is networking. Thousands of chips only become a supercomputer if they can talk to each other at extreme speed, and these companies make that possible.

> $MRVL makes the chips that connect everything together.
> $ANET sells the switches that direct traffic inside AI data centers.
> $ALAB connects the components inside each server rack.
> $CRDO makes the high speed cables and signal boosters between racks.
> $CSCO is an old networking giant that is now winning large AI orders.

Layer 3 is optics. This equipment moves huge amounts of data using light through fiber instead of electricity through copper, because light is faster over distance.

> $LITE and $COHR are the two biggest players in lasers and the transmitters that send data as light.
> $AAOI is a smaller company supplying these parts to the big cloud players.
> $FN manufactures the components inside many of these systems.
> $CIEN carries this traffic across long distance networks between cities.
$SIVE is a speculative pick here, a small Swedish company working on light based chips for AI data centers with partners like GlobalFoundries. It is high risk.

Layer 4 is the data centers themselves, the actual buildings full of these chips.

> $NBIS, $CRWV and $IREN are building AI computing capacity at an incredible speed and rents out massive amounts of Nvidia chips to AI companies.
> $VRT supplies the power and cooling systems inside the buildings.
> $DELL and $SMCI assemble the server racks. > $EQIX owns the real estate where much of this compute lives.

Layer 5 is power, which I believe becomes the biggest bottleneck of this decade. A large share of planned US data centers is being delayed because there are not enough transformers, electrical equipment, and grid connections to power them.

> $CEG and $VST own the nuclear and gas plants that the tech giants are signing decades long contracts with.
> $GEV builds the turbines and grid equipment. > $ETN supplies the electrical gear inside every facility.
> $PWR is the contractor physically building out the grid.

Then there is the layer almost nobody connects to AI, and this is where the most interesting opportunities hide.

$CAT is a perfect example. Most people see a construction equipment company, yet its generator business is booming because every data center needs backup power on site, and management now brings up data centers on every earnings call.

$FCX mines copper, and this entire buildout consumes enormous amounts of it. $MP mines rare earths, which go into the magnets inside motors, turbines, and cooling systems. $UAMY is a more speculative one in antimony, a critical mineral used in batteries and electrical equipment.

The common thread is that the AI buildout reaches much further into the economy than most investors realize.

This is a buildout that will run for years. Own the companies supplying every layer and give the thesis time to work.

NFA

[Ari Tanner @Arzca83]

@VoltaWagen @klaus_koo Kävin koeajamassa myös Juniperin ja olin kyllä erittäin vaikuttunut. Sisustakin oli erittäin laadukkaan oloinen.

[Ari Tanner @Arzca83]

I don't understand this IREN vs. NBIS feud on X
$IREN 1y +734%
$NBIS 1y +978%

Just enjoy the ride! 😎They both do just fine!

[Ari Tanner @Arzca83]

@ErkkilaJorma Oma pelkoni onkin, että AI:n tuomat säästöt valuvat tuottajien taskuihin ja hinnat jatkavat kasvuaan inflaation mukana. Ihan oma mutu vain.

Annan karkeana ei-AI-esimerkkinä isojen kauppaketjujen keräilyautomatiikka. Nämä eivät ole heijastuneet kuluttajan hintoihin.

[Ari Tanner @Arzca83]

@KimmoMatikainen @AkiPyysing Uusi tuotanto on muna vai kana -ongelma. Sähkön hinnan on noustava, että uuden tuotannon rakentaminen on kannattavaa.
Jos halutaan edes haaveilla uudesta ydinvoimasta, niin tarvitaan uutta kulutusta, oli se sitten konesali tai alumiinitehdas.

[Ari Tanner @Arzca83]

@JarmoFriman Huhut X:ssä on kyllä välillä kullan arvoisia. Niin paljon pisteiden yhdistämistä ja muuta online-tutkimustyötä tekee moni accountti.

[Ari Tanner @Arzca83]

"Ain laulain työtäs tee..."

"AI:n laulain työtäs tee..."

Enpä tiennyt tuota lapsena. Miksei aikuiset tarkentanut tuota koskaan. Kaikkea sitä oppii...

[Ari Tanner @Arzca83]

@Ericc_6 @majulepp

[Ari Tanner @Arzca83]

@majulepp Mikä muuten sinulla hiertää BTC:n suhteen?

On ole vielä tavannut ihmistä, joka ymmärtää miten BTC toimii teknisesti, tietää mitä on raha JA olisi BTC:tä vastaan.
Fysiikan opettajana ja energia-alan asiantuntijana sinulla on hyvä lähtökohta BTC:n energia puolen ymmärtämiseen.

[Pete Karjalainen @PeteKarjalainen]

Tässä kun lueskelee datakeskuksiin liittyvää kritiikkiä. Tuntuu melkein absurdille, jotta louhinta nähdään vielä pahempana. Lämpötilat mitä voidaan kerätä talteen on jo lähtökohtaisesti todella korkeita, millä "priimaamisen" tarve jää varsin vähäiseksi kaukolämpöön. 1/x

[Ari Tanner @Arzca83]

@PeteKarjalainen Pitäisi vaan alkaa louhimaan oikein kunnolla. Olla osana reservimarkkinoita, väistellä kadulla mätiä tomaatteja, kuunnella kuinka keität valtameret. Kyllä aika korjaa tämän lopulta ja louhijat tulevat istumaan A-studiossa kuninkaina lopulta.

[Ari Tanner @Arzca83]

@VoltaWagen @MikkoJValtonen Tesla kuulemma korjaa nuo mukisematta ja omistajat tietävät tuon kyllä.
Kunnollisen huolto- ohjelman puute on se mistä Teslaa pitää kyllä arvostella.

[Ari Tanner @Arzca83]

@LyytinenReijo @vontuchman Tämä. Kaikki on ihan vauhkona GPU capexista. Aivan kuin HPC tarjoajat eivät osaisi hinnoitella tuotettaan tai solmisivat diilejä, mistä jää tappiolle.
Toki ala muuttuu kovasti ja kilpailu on kovaa, mutta fakta on, että IREN kaltaisilla on nyt se, mitä markkinat haluavat.

[Ari Tanner @Arzca83]

In every ER @danroberts0101 keeps reminding that ".. but the real world doesn't scale that way. Power, infrastructure, land, data centers, these takes years to permit"

Those who listen and realize this will win. IREN has such a head start!

[Ari Tanner @Arzca83]

@ConjurerThouse @JarmoFriman Tältä se juuri näyttää kun markkina ei ole ymmärtänyt vielä mistä on kyse.
IREN tarina on vasta alussa.

[Ari Tanner @Arzca83]

@elonmusk @SawyerMerritt @MobofJoggers @nottombrown Let's say OpenAI would ask for compute. What would you say?

[Ari Tanner @Arzca83]

Huh. Foundry mined two blocks 6 seconds apart.

[Ari Tanner @Arzca83]

@Bare_Birk Great article Birk! Very well written.

[Ari Tanner @Arzca83]

Finally got time to read this through. Must read for anyone interested in IREN!

[Birk @Bare_Birk]

https://t.co/0PKimVyr7i