Tropical Value

🇧🇷 15y Public Equities (ex-Head of Research @ top Brazil fund). Experience w/ Alternatives & Illiquids. Former Public Co. Board Member. Tech enthusiast.

[Tropical Value @tropicalvalue]

https://t.co/3ac5cFgWaB

[Tropical Value @tropicalvalue]

@oguzerkan Such an awful timing for the CFO departure though. Investors need some "credit cycle" reassurance. Now we have a "Global" CFO that doesn't know that well the Brazilian credit market that is where all profits come from today.

Full disclosure: I have a small position.

[Tropical Value @tropicalvalue]

@luiskruger Pois é, ele focou só na ponta do ativo. rs

[Tropical Value @tropicalvalue]

[1/7] Depois do meu post sobre #CSAN3, um gestor que respeito me mandou mensagem:

"Isso é idêntico ao que Buffett fez com as tradings japonesas via $BRK. Comprou empresas lastreadas em commodities, com dividendos acima do custo do capital, usando dívida."

Ele estava certo na tese. Errado na estrutura dos deals.

[Tropical Value @tropicalvalue]

@tuvidux @fahvallim #SUZB3

[Tropical Value @tropicalvalue]

@tuvidux @fahvallim cheguei a dar uma aprofundada no em P&P, mas quando esbarrei na verticalização dos players chineses e sourcing de madeira da indonesia eu desisti

[Tropical Value @tropicalvalue]

@rodrigohelbinho É por aí. O ponto central foi a estrutura de alavancagem., que quando o CDI subiu tirou a Cosan do jogo.

Se fossem passo a passo, usando os dividendos próprios ou de vendas parciais dos ativos do portfolio da holding como você colocou, o negócio teria parado de pé.

[Tropical Value @tropicalvalue]

Everyone is debating whether Zuckerberg's performance monitoring is surveillance. In 1991, Danaher was also tracking 50 employee metrics. That was 14 years before they narrowed it to 8 that made the whole system unique.

The issue for $META is whether the data will become legible to run a self-improving harness on.

AI companies are spending billions rebuilding what Danaher built in 1984. The vocabulary has changed, but the architecture hasn't.

1. Kaizen event → Fine-tuning loop
2. Defects per million → Hallucination rate
3. Escalation to supervisor → Human-in-the-loop

[ $DHR & $META - full piece ↓]

[Tropical Value @tropicalvalue]

https://t.co/AS04aTq5bj

[Tropical Value @tropicalvalue]

@jaminball If companies are already controlling their token budgets at sub-$50B per GW, there will either be construction (incl. electricity) or AI ROI bottlenecks for end-users. Unlikely scenario IMHO.

[Tropical Value @tropicalvalue]

$BRK likely would not scale its $GOOGL investment to $30 Billion without thoroughly analyzing the return on investment of the AI datacenter buildout. While markets worry that rising depreciation will stifle near-term EPS growth amid massive capital expenditures, Berkshire possesses the capital duration and patience required to let this scale of investment bear fruit.

Traditional infrastructure assets rarely achieve a payback period under 5 years. Yet, structured debt vehicles for Neoclouds ($CRWV / $NBIS / $IREN) currently amortize in 4 years or less. Because Alphabet utilizes its proprietary TPU chips, it should achieve a 30% faster payback timeline, reducing the capital recovery period to roughly 3 years.

Berkshire managers focus on owner's earnings and actual cash flows rather than accounting EPS. If the industry deploys the projected $7.5 Trillion in cumulative AI infrastructure (~$1.5 Trillion annually) and
$GOOGL captures a 20% incremental share, the company will deploy over $300 Billion in Capex compounding at a 3-year payback cycle at the TPU level for years to come.

[Tropical Value @tropicalvalue]

$BRK likely would not scale its $GOOGL investment to $30 Billion without thoroughly analyzing the return on investment of the AI datacenter buildout.

While markets worry that rising depreciation will stifle near-term EPS growth amid massive capital expenditures, Berkshire possesses the capital duration and patience required to let this scale of investment bear fruit.

Traditional infrastructure assets rarely achieve a payback period under 5 years. Yet, structured debt vehicles for Neoclouds ($CRWV / $NBIS / $IREN) currently amortize in 4 years or less.

Because Alphabet utilizes its proprietary TPU chips, it should achieve a 30% faster payback timeline, reducing the capital recovery period to roughly 3 years.

Berkshire managers focus on owner's earnings and actual cash flows rather than accounting EPS. If the industry deploys the projected $7.5 Trillion in cumulative AI infrastructure (~$1.5 Trillion annually) and $GOOGL captures a 20% incremental share, the company will deploy over $300 Billion in Capex compounding at a 3-year payback cycle at the TPU level for years to come.

[Compound248 💰 @compound248]

Greg Abel gettin' jiggy with it:

$BRK is investing $10 billion directly into $GOOG to help fund its AI buildout plans as part of an $80 billion overall equity raise:

$10B - BRK
$30B - underwritten offering
$40B - ATM to begin Q3 2026

Abel has now announced $17B across two deals in the past 24 hours.

[Tropical Value @tropicalvalue]

O desfecho separa os dois: $BRK detém ~9% das tradings japonesas e desde o investimento já se valorizaram entre 200% e 400%.

Já a Cosan foi obrigada a desfazer a estrutura alavancada e vendeu suas ações da #VALE3.

[Tropical Value @tropicalvalue]

[6/7] A tese era a mesma. A execução era estruturalmente incompatível com o Brasil.

Buffett capturou um regime único: juros zerados no Japão, sem risco cambial, sem chamada de margem, bonds de prazo longo.

A Cosan entrou no mesmo trade numa estrutura com custo flutuante e vencimentos curtos. O problema foi o descasamento entre ativo e passivo.

Por sinal, a #VALE3 hoje é cotada > R$80, fora os dividendos do período.

[Tropical Value @tropicalvalue]

@thomasrice_au 1/ Probably true; maybe will use this specific capital raise to fund their JV with $BX to fund AI datacenters

2/ ~30x P/E; so implied cost of equity likely lower than cost of debt given treasury @ 5%

3/ Their Capex plan is so massive this is all rounding numbers

[Tropical Value @tropicalvalue]

@Jesse_Livermore Rounding numbers... $10B is a VC series B round nowadays

[Tropical Value @tropicalvalue]

We got the growth. We got the value. We ain't got the momentum.

$META

[Tropical Value @tropicalvalue]

Generally agree. I find $GOOGL approach to LLM token cost-efficiency a differentiator in the West. Only the Chinese are approaching the ROI equation in the same way.

Shorter-term, (enterprises) both $MSFT and $UBER have recently demonstrated their preoccupation regarding tokens cost-benefit. From the investors side, pods only care about earnings revisions 2-4 quarters out - so it's kind of obvious D&A for the next couple years do not help. Forget about buybacks as well.

Longer-term, it seems inexorable AI is an irreversible trend. DC infra is the pre-requisite. Of course there might be air pockets along the way, but 10 years out ($BRK investment horizon) seems "money in the pocket" for them.

Time will tell.

Appreciate your response. Best of luck.