concentrated

"If you can identify six wonderful businesses, that is all the diversification you need. And you will make a lot of money."

[concentrated @wrtconcentrated]

@JasonrShuman No way 89% cracks the top 10

[The Inflection Economy @inflectionecon]

$MSFT losing exclusivity to OpenAI getting read as bearish.

Feels backwards. If models are commoditizing, exclusivity matters less. Control shifts to the system around the model: distribution, workflows, enterprise integration. This selloff feels like a misread.

[concentrated @wrtconcentrated]

@WagsCap Think the Q4 guide with Fairwater live should be the defining factor. Market will be able to shrug off weak Q3 if guide is >40%

[concentrated @wrtconcentrated]

https://t.co/E1DewwhA6z

[roon @tszzl]

“the future of work” there is no future of work. we are going to systematically remove the burden of the world from atlas’ shoulders

[concentrated @wrtconcentrated]

> AI can automate work
> Is really good at software engineering
> Software companies spend lots of money on software engineering
> Sell software companies?

[concentrated @wrtconcentrated]

@deedydas Hey Deedy are you sure on the Manus $500m ARR?

[concentrated @wrtconcentrated]

Clean break from China likely enough to get the deal past regulatory scrutiny

[concentrated @wrtconcentrated]

Manus = systems level engineering.

Interesting that manus will keep operating independently. Manus capabilities can improve Meta AI but Meta can also grow an AI business that isn't reliant on users opening it through Instagram / Facebook.

[concentrated @wrtconcentrated]

While LLMs will continue to improve and beat new benchmarks, the real progress in '26 and beyond will be from constructing agentic systems.

The unlock for meta users and end-to-end ad campaigns could be a major accelerant to Meta revenues.

[concentrated @wrtconcentrated]

Meta <> Manus is about systems engineering. Agents turn strong models into highly capable systems when done right.

Truth is that most of the best AI tools we are interacting with today are systems under the hood (GPT-5 is actually a router and a number of different models)

[Stock Surgeon @TheStockSurgeon]

This is a strategic acqui-hire plus capability grab, not a headline M&A flex.

Meta Platforms buying Manus isn’t about revenue today — it’s about speed, talent, and architectural advantage in the next phase of AI.

Manus is known for building agentic AI systems — models that don’t just respond, but plan, execute, and chain actions across tools. That matters because Meta’s next bottleneck isn’t compute or data, it’s usable intelligence at scale.

Here’s how Manus helps Meta in real terms:

First, AI agents across Meta’s ecosystem.
WhatsApp, Instagram, Facebook, and Messenger all sit on massive user intent data. Agentic systems can turn those platforms from passive feeds into active assistants — booking, searching, shopping, moderating, summarising, and executing tasks autonomously.

Second, Llama goes from model to system.
Meta’s Llama models are strong, but models alone don’t win. Manus brings orchestration — multi-step reasoning, tool use, memory, and autonomy. That’s how Llama becomes competitive with closed ecosystems without closing itself.

Third, ads and commerce efficiency.
Agentic AI can optimise campaigns end-to-end: creative generation, A/B testing, budget reallocation, and conversion optimisation in real time. That directly hits Meta’s core cash engine.

Fourth, developer leverage.
Manus-style agents lower the barrier for builders. If Meta packages this well, it strengthens its open-AI positioning versus OpenAI/Microsoft and Google — more builders, more lock-in, more network effects.

Why the China angle matters:
Chinese AI teams are often exceptionally strong at applied systems, efficiency, and fast iteration under constraints. That’s exactly what Meta needs as compute costs rise and AI moves from demos to deployment.

Bottom line:
This isn’t about “Meta buying a Chinese startup.”
It’s about Meta buying agentic intelligence to turn attention platforms into action platforms.

If executed well, this pushes Meta closer to owning the interface layer of AI for billions of users — and that’s far more valuable than any single model.

[concentrated @wrtconcentrated]

Susan Li on how Meta is balancing flexibility with ensuring they don't get caught off guard by an AI inflection.

I think many don't appreciate the lead times and bottlenecks to spinning up datacenter capacity.

[concentrated @wrtconcentrated]

Its interesting that most people can agree on AI being the most transformational technology of our lifetimes, have genuine fears that it will replace their own job, yet can't grok a company investing slightly above the amount that grows their FCF every single quarter

[concentrated @wrtconcentrated]

@kakashiii111 Depreciation has no link to the external value of the asset being depreciated. The only relevant question is whether they will be using the GPUs in 6 years? They will no longer be leading edge, but there will be workloads you can route to older GPUs.

[concentrated @wrtconcentrated]

Building AI products that don't work today is buying lottery tickets on AI progress. For now - maybe brand and distribution matters more.

[concentrated @wrtconcentrated]

@SleepwellCap There are obvious ways to poke holes in this guys analysis, but you can be be more thoughtful than using trailing FCF x to dunk when the biz is going through a massive capital cycle to capitalise on the next platform transition in tech.

[concentrated @wrtconcentrated]

@realroseceline Good bait

[concentrated @wrtconcentrated]

@Cookery_God VC backed players burning cash, PE doing PE stuff, incumbents and a potential way of achieving the same thing with sw from a different category. This is like every established software sub-category haha

[concentrated @wrtconcentrated]

A very good summary of everything that is going wrong with the Microsoft/OpenAI partnership

[concentrated @wrtconcentrated]

What's exciting about MongoDB at <7x rev is that they have made a number of highly positive strategic steps to consolidate workloads on the platform (time series, vector, re-ranking) and are looking very well positioned for AI Inference.

[concentrated @wrtconcentrated]

The biggest disservice that Dev has done MongoDB is try to convince the market that they are the superior database and can go after all workloads. When the market sees Postgres traction, it gets interpreted as MongoDB "losing" - hence Blair fighting the bear case here.

[concentrated @wrtconcentrated]

The reality is that each database caters to different needs, and *most* workloads are better catered to by relational. BUT both are major share gainers in a $100bn+ database market.