John_Galt @NoesisPrime

Asymmetric risk/reward. Space tech. AI buildout. Washington, DC Joined June 2018

Tweets

456
Followers

1K
Following

2K
Likes

89

John_Galt @NoesisPrime

5 hours ago

@YodaStockInvest I would say $nlst, they just pressed Samsung for a deal while being 800 mcap company

0 0 1 358 0

View Details

John_Galt @NoesisPrime

9 hours ago

@capybaraReborn x.com/noesisprime/st…

John_Galt @NoesisPrime

9 hours ago

0 0 5 225 0

0 0 0 66 0

View Details

Autor pyta dlaczego IREN i Nebius emitują akcje zamiast brać dług na Fed + 3%. Odpowiedź leży w fizycznym łańcuchu dostaw. Kredyt bankowy wycenia ryzyko operacyjne. Tradycyjny dług finansuje twarde, pracujące aktywa. 15-letni lease. Wpięty prąd. Działające serwery. Teraz spójrzmy na liczby z placu budowy. Kolejka oczekujących na podłączenie w teksańskim ERCOT wynosi 410 GW. GE Vernova ma backlog na turbiny gazowe rzędu 116 GW. Czas dostawy dużych transformatorów (LPT) przekracza 3 lata. Żaden bank nie sfinansuje 4-letniego stania w administracyjnej kolejce po pozwolenia środowiskowe. Dlatego mniejsi operatorzy muszą używać ATM. Muszą przetrwać fazę high-risk development, w której sprzęt fizycznie nie istnieje. To samo dotyczy hyperscalerów. Google i Microsoft nie leasingują gotowych data centers dlatego, że brakuje im kapitału na własny budynek. Płacą premię za ominięcie 36-miesięcznego bottlenecka na switchgeary. Kupują time-to-market. Autor trafnie diagnozuje lewar opcyjny ulicy. Jednocześnie kompletnie ignoruje strukturę kosztów i opóźnień na fizycznej sieci.

Capybara Stocks @capybaraReborn

20 hours ago

The following is a warning to all investors in the AI compute space. Over the past year nearly all equities linked to the AI hardware space have experienced 10x runs or more. While there is real promise in the AI trade, it is my belief that compute equities will be responsible

93 6 120 28K 79

0 0 5 225 0

View Details

John_Galt @NoesisPrime

21 hours ago

Most people still think the AI bottleneck is chips. It’s not. It’s power + cooling + grid interconnection. Crypto mining already proved this years ago the winners weren’t the ones with the best ASICs, they were the ones who secured cheap, reliable electricity at scale. The same movie is now playing with AI data centers. Only this time the numbers are 10–20x bigger. $NVDA $IREN $CIFR $HUT $BTBT $NUAI $APLD #AI #DataCenters #Energy

0 0 7 100 0

View Details

John_Galt @NoesisPrime

23 hours ago

x.com/noesisprime/st…

John_Galt @NoesisPrime

a day ago

$CRWV started by mining $ETH. LiquidStack scaled immersion cooling through Bitfury’s mining infrastructure. Crusoe built modular data centers next to stranded energy and used them to mine $BTC. Today all three stories run directly into AI infrastructure. Crypto did not invent

4 0 2 238 0

0 0 0 16 0

View Details

John_Galt @NoesisPrime

23 hours ago

Anthropic: ~$19B with $WULF. $MSFT: $9.7B with $IREN. $AMZN: ~$5.5B with $CIFR. All three suppliers were built around crypto mining. AI is now renting the power infrastructure crypto spent years assembling.

2 0 1 358 0

View Details

John_Galt @NoesisPrime

23 hours ago

So the question is no longer: “Can Bitcoin miners pivot to AI?” Some already did. The better question is which mining-era power portfolios can survive the conversion economics and become long-duration AI infrastructure. $BTC created a market willing to finance power-intensive compute almost anywhere electricity was cheap. Now Anthropic, $MSFT, $AMZN, $AMD and $CRWV are putting very different economics on top of some of the same sites.

1 0 1 43 0

View Details

John_Galt @NoesisPrime

23 hours ago

There is a catch. A Bitcoin mine is not automatically an AI data center. $RIOT estimates roughly $11M to $12M of new construction cost per IT MW for its latest 191 MW AI project. The tenant wants Tier 3 infrastructure, redundant systems, dense networking and reliable cooling. Mining tolerated interruption. AI tenants pay for uptime. And shareholders paid for the journey. $IREN had ~188M shares outstanding in June 2024. By April 2026: ~357M. Roughly +90%. $WULF had 495.5M shares outstanding in May 2026, plus 80.9M warrants and 47.5M RSUs outstanding at quarter-end. Power may be scarce. Capital is still required to turn it into AI infrastructure.

1 0 0 126 0

View Details

John_Galt @NoesisPrime

23 hours ago

x.com/noesisprime/st…

John_Galt @NoesisPrime

2 days ago

Sometimes I think about how crypto mining quietly prepared the entire compute layer for AI. The GPUs. The high-density power infrastructure. The operators who learned to run racks 24/7 at the edge of thermal limits. Bitcoin didn’t just create a currency. It stress-tested the

2 0 1 192 0

0 0 0 16 0

View Details

John_Galt @NoesisPrime

a day ago

4 0 2 238 0

View Details

John_Galt @NoesisPrime

23 hours ago

@naiivememe It is exactly opposite for me

0 0 0 36 0

View Details

John_Galt @NoesisPrime

23 hours ago

@Nexora_iQ Can we call crypto a R&D department at this point?

0 0 1 5 0

View Details

John_Galt @NoesisPrime

a day ago

So I would rewrite the original thesis slightly. Crypto did not build AI. It pre-funded a parallel experiment in cheap power and dense compute. $ETH trained GPU operators. $BTC trained power operators. Immersion cooling got a brutal commercial test bed. Flexible-load software learned to treat compute as part of the grid. Modular data centers learned to follow stranded energy. AI is now buying those skills, those interconnections and, increasingly, those exact sites. $BTC $ETH $NVDA $AMD $CRWV $CORZ $WULF $HUT $RIOT $CIFR $IREN

0 0 0 42 0

View Details

John_Galt @NoesisPrime

a day ago

The limit matters. You cannot move an AI model into a mining shed and call it a data center. AI needs much more: dense east-west networking fiber redundancy power quality liquid cooling tenant-grade uptime large synchronized clusters That is why $WULF, $RIOT, $HUT and $CIFR are spending billions rebuilding infrastructure around power they already control. The inherited asset is not the building. It is the head start.