Tokyo Alpha Mind

Global Investing 📈 AI 🤖 Tokyo Life 🇯🇵
Insights for your financial and personal growth from Japan.

[Tokyo Alpha Mind @TokyoAlphaMind]

@KrisPatel99 The real issue is that most “cheap” models are chinese models and are just not okay to use in enterprise.
If for example $MSFT releases an hyper low-cost model, Anthropic an OpenAI ARR will plummet overnight

[Tokyo Alpha Mind @TokyoAlphaMind]

@EquityBrian I think CCP helped a lot on this one by limiting the involution war on Quick Commerce.
With less promotions to be made under less competitive pressure, this makes sense

[Tokyo Alpha Mind @TokyoAlphaMind]

Je regarderais plutot tout ce qui se fait chez Neos comme $QQQI ou $SPYI meme concept de covered call ETF mais une strategie plus simple sur la base de QQQ ou SPY plutot qu’une gestion active des stocks underlying.

Ceci etant dit ces covered call ETF ont tendance a surperformer quand le marche est relativement flat mais va underperformer sur de long rallye ou bien meme en cas de crash. Le vrai avantage est pour ceux qui veulent une partie du total return distribué en dividende plutot que de le laisser compound

[Tokyo Alpha Mind @TokyoAlphaMind]

You should absolutely not DCA if you enter the memory trade and $MU goes 30% lower
Semis are cyclical, if $MU start going lower and lower, most likely it will be because the market would sense the end of this cycle.
The Buy The Dip mentality does not have to work for everything.

I think this is a misguided advice in a good take. If you enter the memory trade now you should treat that as a trade on the basis that the shortage is going to last beyond 2028 and that $MU will continue to have significant pricing power.
So the thesis should be that Memory goes even higher from here

If $MU decline by -30%, most likely the thesis would be broken and most likely Micron will not see its ATH anytime soon. DCAing in this case will make you DCA in a very long way down.

Not Financial Advice, but be careful out there folks

[Tokyo Alpha Mind @TokyoAlphaMind]

Excellente analyse sur la situation actuelle.

Effectivement tout le monde cherche le nouveau bottleneck. Point interessant d'ailleurs que j'ai pu voir autre part est que la partie memory est la seule de la chaine de valeur a avoir repondu a ce goulot d'etranglement par une hausse extreme de son pricing, un peu a la NVIDIA. Mais au vu de la demande et des backlogs affiches par a peu pres tous les acteurs on devrait voir le phenomene se repercuter sur toute la chaine de valeur.

Interessant egalement c'est l'aspect reflexif de la penurie qui fait augmenter les prix et donc les montants de CAPEX (Comme mentionne par $META durant leur earning call). Les montants en dollars augmentent mais la capacite de compute pas autant.

Enfin je note que la tres grande majorite de la demande sur les semis-infra et des CAPEX associes viennent des Hyperscalers et de META (les 700 BnUSD cette annee et les 1 Trillions en 2027). Ce sont eux qui vont sonner la fin de ce cycle quand ils vont commencer a reduire drastiquement leurs investissements.
Ceci etant dit, le consensus reste sur un effort d'investissement qui s'etend sur 5-10 ans (on l'entend souvent chez les Brookfield, KKR, Apollo, Blackstone qui financent ces projets data center)

[Tokyo Alpha Mind @TokyoAlphaMind]

It seems obvious now to anyone that no enterprise is going to vibe code any SAAS themselves just to potentially save money.

However the less obvious is that many will and are trying to build their own agent architecture and agentic systems (see the JP Morgan presentation)

Some will succeed, but many will not and will be faced with the reality of enterprise-level software.
Just the matter of Access & Identity management is complex

IT department will never let anyone build agentic system outside of platforms like $NOW, $CRM, $MSFT, $PLTR or $SNOW just to mention a few

I think it will take time for IT department to catch-up because AI is improving too fast for those departments to keep the pace. The governance is going to come for the only fact that agents are using the company most valuable asset, its knowledge.

Having said that, $MSFT has one of the best position since they could (and they will) capitalize on the fact Copilot will be the safe chat and agent for the enterprise where employee can safely use company documents.
OpenAI and Anthropic has been so fear-mongering that many enterprise do not really trust them with their data and mostly tolerate them because their offering is way more advanced than Microsoft.
With Microsoft Copilot and Github Copilot, $MSFT could just play the safe Switzerland of LLM in the enterprise.

Coming back to $NOW I think that the real slide that people should look at is this one because it shows you why enterprise is going to pay ServiceNow and why $NOW is going to keep great margins going forward

[Tokyo Alpha Mind @TokyoAlphaMind]

First there is a prisoner's dilemna for both OAI and Anthropic now that they have heavily subsidized their token. The first one to reflect the true economic value of the tokens into their price is going to lose a lot of demand and you don't want to show that pre-IPO

Then, we should not see these AI Labs in isolation of everything else. Google has frontier models, there are open-source frontier models as well. These are much cheaper sources of tokens, and AI Labs cannot have pricing completely disconnected from that. Otherwise same issue as above.

The crux of all of this is that OAI and Anthropic do not have the best economic structure here and they are forced to grow their way out (a la Uber as many mentioned).
But the reality is that demand is not concentrated only at Opus-level demand. I would guess that most workloads can be handled by much cheaper models and this is where the infinite demand is and is going to be in the future. The issue for OAI and Anthropic is that they are not tooled for a cost and efficiency battle vs Google/Amazon or Chinese Open Source. While Opus-level tokens are cheap people will use them without second thoughts by default. As soon a price rises, people will switch, optimize, eval. and profits will shift to the most efficient inference providers

[Tokyo Alpha Mind @TokyoAlphaMind]

Indeed it is clear that expanding credit is a net positive for $MELI.
If I understand correctly we should consider the credit - e-commerce flywheel as a whole and not transaction by transaction since The interest paid on loans cover enough of the defaulted loans (i.e NIMAL positive).

Having said that I think it is fair to say as well that e-commerce revenue is partly supported by the credit provided by Mercado Pago

[Tokyo Alpha Mind @TokyoAlphaMind]

@DrewCohenMoney

[Tokyo Alpha Mind @TokyoAlphaMind]

@DrewCohenMoney $MELI provide you an adjusted free cash flow specifically for this reason in their presentation

[Tokyo Alpha Mind @TokyoAlphaMind]

Stay with the big 3 $APO $KKR $BX
You should really deep dive though as they have very different exposure and even business models.
I like $KKR as it is well diversified and have started a Berkshire-like approach with their "Strategic Holdings". Though $KKR is much more geared towards Private Equity, which means they are much higher beta and relies on M&A activity

$APO on the other side is almost all private credit. So maybe more stable than a $KKR but with less upside. $APO though has been the best at Private Credit origination compared to a $OWL for example. Their dominant position is actually important because this means they did not need to take as much risk in the underwriting as other smaller player for growth

This is simplified but these are the 2 I would look at

[Tokyo Alpha Mind @TokyoAlphaMind]

@KrisPatel99 The trap actually is that those cyclical businesses usually have low P/E at the top.
The sign of a reversal is when these component companies start talking about expanding capacity, especially into slowing demand growth

[Tokyo Alpha Mind @TokyoAlphaMind]

All Private Credit/Equity are selling off look at $APO, $KKR or $BX
As an asset manager $BN is thrown away with the bath water. No distinction is made on the Private credit players and Brookfield business with Oaktree and any Insurance-led models is sold off today indistinctively same as SAAS.
The market is going to realize though that Oaktree is actually opportunistic distressed debt not just private credit, and that Brookfield actually has a very low exposure to Private Credit.
Though fundraising might be more difficult this year going forward due to the perception on private assets today

[Tokyo Alpha Mind @TokyoAlphaMind]

This practice has stopped actually under Alex Chriss where they renegotiated many contracts to align in terms of profitability with Stripe and Adyen.

The reason why it was a loss leader under Dan Schulman was merely an attempt to undercut Adyen and Strip and take market share, in a land and expand strategy.

There are not really synergies between Branded Checkout and Braintree

[Tokyo Alpha Mind @TokyoAlphaMind]

Could have productized Google Finance years ago but they never cared that much. Google Finance does not even have an iOS app.
Maybe this time they will take it seriously and start going after perplexity finance. They just need to bundle it into Google AI Pro with some Youtube Premium and they have millions of the FinTwit/ Finance Youtube community with their own version of Prime for Google Products. They could even add a NotebookLM dedicated to Equity Research.

They just need to realize that a lot of people are ready to pay to increase the probability of superior returns

[Tokyo Alpha Mind @TokyoAlphaMind]

I think you said it best yourself :

I buy quality at a reasonable price and hold until it's egregiously expensive, or I find a drastically better opportunity.

Now ASML trades at almost 40x Forward P/OCF and you are fairly concentrated in your top ideas. Trimming would not be a bad move on a risk management perspective.

[Tokyo Alpha Mind @TokyoAlphaMind]

@Adamtradez @fiscal_ai Great Income portfolio

[Tokyo Alpha Mind @TokyoAlphaMind]

I had a model too when the stock was at 60$ which I though was conservative.
I too was modeling 15% EPS growth so was clearly blindsided when they guided for EPS decrease in 2026. To me it makes no sense given the share repurchase program and this would imply something like a double digit decrease of Net income. If they actually sandbagged it and they are going to continue to grow EPS double digit, this guidance was borderline securities fraud.
Anyway we will see next quarters
https://t.co/MkW3K7qIjh

[Tokyo Alpha Mind @TokyoAlphaMind]

Actually as a developer you have to change all the time because there are always new models and always models optimized for different tasks. Also another trend is that because models become more performant on price/intelligence basis you tend to use them for tasks that previously required non-LLM algorithms.