Domenico Mosca

Investitore e Consulente finanziario (PFA) FinecoBank. Porto chiarezza su investimenti e allocazione di portafoglio
#Finance #globaliquidity #Bitcoin

[Domenico Mosca @DomenicoMosca3]

Sono felice di condividere il primo articolo di Capitale Intelligente, la mia nuova newsletter per investitori evoluti.
AGI prima del 2030: cosa significa per un investitore https://t.co/IcBdSvbY2n

[NVIDIA @nvidia]

10x more tokens per megawatt.

CoreWeave has the first measured performance of NVIDIA Vera Rubin NVL72, showing 10x improvement in tokens per second per megawatt on DeepSeek-R1 compared to Blackwell.

[Domenico Mosca @DomenicoMosca3]

@hamids @chamath @shiri_shh @chamath is the dumbest billionaire in the world

[AI Edge @aiedge_]

https://t.co/yCzQfzrh1B

[Domenico Mosca @DomenicoMosca3]

https://t.co/AWqEwf3EA6

In questo articolo della mia newsletter analizzo cosa rende i ricchi (che hanno investimenti) sempre più ricchi e i poveri sempre più poveri

[Domenico Mosca @DomenicoMosca3]

@RihardJarc Solid post but there is an error on Google Cloud profit margin. It is way higher (33%) than the 17% stated

[vas @vasuman]

Must read. Very thorough analysis on AI scaling.

[Domenico Mosca @DomenicoMosca3]

@ariaradnia Mercadolibre

[Claude @claudeai]

Claude for Excel, PowerPoint, and Word are now generally available, and Claude for Outlook is in public beta.

As Claude moves between your Microsoft apps, it carries the full context of your conversation.

[Domenico Mosca @DomenicoMosca3]

@TheLongInvest $TMDX

[Domenico Mosca @DomenicoMosca3]

@RihardJarc They need to change Ceo and lean the company. With right management Amazon will be highest valued company on earth

[Domenico Mosca @DomenicoMosca3]

🎪

[Antonio Linares @alc2022]

We’ve gone from AI is a bubble to AI is disrupting everything

[Rihard Jarc @RihardJarc]

$GOOGL's new Gemini 3 DeepThink has ARC-AGI-2 at 84.6%.

WOW.

[Rihard Jarc @RihardJarc]

I have increased my $META position substantially following these earnings results.

Here is my take on the quarter.

1. Revenue from the core business is accelerating faster than expected, with 25% YoY revenue growth this quarter and guidance for Q1 of 30% YoY on the high end. The guide suggests the fastest growth $META has seen in 5 years. On the call tho $META CFO did mention expecting lower growth rates for the full 2026 than the one in Q1.

2. $META is showing quarter after quarter, clear ROI from their AI investment. AI and GenAI are affecting $META by driving higher engagement rates and better targeting, translating into more time spent and better ROAS for advertisers.

3. $META finally »unloaded« the CapEx number burden that was hanging over the stocks since the last earnings. $META expects CapEx for 2026 to be $115B-$135B, compared to $72.2B in 2025. While this is a true "bomb", it was somewhat expected, but $META backed it up with real ROI from the past AI investments in the form of revenue acceleration, which investors will approve.

4. An important part was $META, hinting that Reality Labs losses in 2026 would be similar to those in 2025, and said they expect this to be the peak in terms of losses from this segment.

5. AI Glasses are the next form factor: »Sales of our glasses more than tripled last year, and we think that they're some of the fastest-growing consumer electronics in history.«

6. $META is being effective in managing its infrastructure by using SLMs, not just LLMs, and having hardware compute specialized for different workloads (not just $NVDA). On top of that, on the follow-up call, we got clarity that the increase in OpEx from using public cloud providers is not a long-term thing, but more of a short-term thing until $META can scale up its own infrastructure sufficiently.

7. $META hinted at potential new revenue lines from AI in the future, from e-commerce, agents, and subscriptions that will unlock a new revenue growth driver for the company.

Given the guided growth rate and my understanding of $META's business and long-term margin profile (as investments and losses in AI & RL will normalize), $META would, without those excess losses, make close to $100B in net profit in 2026. Even after the 8% jump in the stock price, the stock still trades at an 18x multiple to those earnings, on top of 25-30% core top-line business growth, with new revenue opportunities still to come.

A rare breed.

[The Long Investor @TheLongInvest]

The good news is that there is always an opportunity in the markets

China is outperforming the US and has been the perfect hedge

Metals are out performing everything and also a hedge against the US market and crypto

Your job is to have Cash ready, have your safe havens in place and know where the rotation is going

Everyone should be licking their lips at the prospects of being able to buy Mag 7 positions testing their 200 WMA.

$NVDA, $GOOG and $AMZN are my main targets.

[Rihard Jarc @RihardJarc]

$META has just opened the floodgates for the AI agentic application layer. $META just acquired Manus, an AI-agentic startup, for more than $2B because Manus excels at agentic real-world tasks. It is also the startup that has reached the $100M ARR market the fastest in history (8 months).

While many might think Manus is just an LLM wrapper, a closer look at the company reveals some interesting aspects and values:

Manus, unlike ChatGPT, was built to execute tasks rather than provide text answers. The goal is to assign it a high-level task so the agent can navigate different tasks autonomously to complete the job (e.g., navigating the web, writing code). Foundation models predict the next word, while Manus predicts the next action.

The unique part about Manus is that it uses an architecture called Code-Act. Instead of just talking about a problem, it writes a Python script on the fly to solve it, executes that script in a secure »sandbox« (virtual machine), and looks at the result. The Sandbox means that Manus provides every user with a dedicated, cloud-based Linux environment. They now have more than 80M virtual computers. This means the upfront costs are higher than with ChatGPT. You pay for the virtual computers even when they are idle, but because of their specific architecture, they are cheaper per task completed. Manus is model-agnostic and uses multi-model routing. For scraping a website, it can use a low-cost model; for summarizing content, it can use a more expensive model within the same task. It also caches previous task executions, so if, for example, it has already solved how to log in to a site, it saves the script from previous tasks and, with it, the tokens. Because it runs in the background with the virtual machines and uses a 3-agent orchestration system of a planner, an executor, and a verifier, it also often doesn't need »retries« like other models need (as they often lose focus or connection to do long-hour tasks). The combination of those things makes it cheaper per solved task.

Manus at launch was higher than OpenAI Deep Research across levels 1, 2, and 3 of the GAIA benchmark. The GAIA benchmark is a reality check for AI agents on how well they perform on real-world tasks that are routine for humans but typically a challenge for AI (humans score 92% accuracy).
How does this fit into $META? Well, $META is looking to build a superintelligent AI assistant for every person, so this aligns with what they are trying to do. They are already working on the foundation model at $META (if they fail, they can use open-source or partner with an AI lab). Still, you also need to have agentic infrastructure, which is everything that Manus has done and has achieved serious traction for it (so it works).

This best fits into $META's WhatsApp as an assistant that they can offer both to consumers and businesses, and a strong play for their $META Ray-Ban Smartglasses, where you need an autonomous, agentic system to run those glasses. $META will also help Manus with its infrastructure, allowing Manus to stop renting virtual machines from cloud providers and use $META's own infrastructure.

With this move, $META also gains an extremely capable team focused on agentic workflows and building the »body« of the agent, as the »brain« is already being built by $META through its foundation model work.

In honesty, $2B-$4B doesn't seem like much given the cost of AI talent today and the traction and growth Manus has had.

Zuck is also showing you where the puck is going. $META might have opened up the floodgates for the application layer of AI with these moves. I expect others will follow, so in hindsight, this deal will look cheap.

[Domenico Mosca @DomenicoMosca3]

@BourbonCap #tmdx

[The AI Investor @The_AI_Investor]

Jonathan Ross built Google’s TPU and later Groq’s LPU.

Listen to his takes on OpenAI’s compute deals and the differences between Groq’s LPU and Nvidia's GPUs.

[Shay Boloor @StockSavvyShay]

$GOOGL enters the AI economy with three advantages OpenAI doesn’t have:

1. Global distribution surfaces (Search, YouTube, Android, Gmail)
2. The world’s most efficient ad stack
3. End-to-end control of infrastructure from chips to data centers

That means AI usage for Google is margin-accretive by default and not something that needs to be figured out later.

[Shay Boloor @StockSavvyShay]

$AMZN is launching an $META Instagram-style video feed for TV that's debuting on Fire TV devices.

It turns TV into a scrollable discovery feed by blending short-form video, creators & ads directly into the living room & plugging into an ad business already doing about $65B in revenue growing ~24% with ~35% margins.