Matakats

Founder | AI × startups × GTM
Building in public. Sharing what I learn.

[Matakats @matankatzzzz]

https://t.co/lynwPhL4KE

[Matakats @matankatzzzz]

@nvidia stopped being a chip company. It’s now the central bank of AI.

This week: in talks to put billions into @perplexity_ai at $30B+.
Last week: $6B licensing deal + $1B equity in @poolsideai .
December: bought Groq’s tech and hired the team.

The playbook: sell you the compute, take your cash, buy your equity with it. Every dollar spent on GPUs flows back to Nvidia twice.

Perplexity went from ~$20B to $30B+ in 12 months. Nvidia already backed them 3 times before this.

Whoever wins AI search, Nvidia collects the toll.

[Matakats @matankatzzzz]

@Tristanrhee3 Also books can answer any question so your point does not make sense.

It’s about how much you can memorize, and subsequently how much of it can you put into practice that leads to positive outcomes.

[Matakats @matankatzzzz]

@currentbitsNET Validation is the key. Has always been.

[Matakats @matankatzzzz]

@ayoub_el77927 Hard disagree.

It depends on who / what you follow.

Just like with friends. If your circle is a certain way, you’ll end up that certain way.

[Matakats @matankatzzzz]

@TTrimoreau This is your chance to figure out :

1. Why did the 3 pay?
2. Why did the rest not?

[Matakats @matankatzzzz]

Talked to 50+ companies this month.

Zero had a strategy problem.

All of them had the same problem: decided X in March, somehow doing Y in August, and nobody can tell you when it flipped.

AI won’t fix this. It doesn’t even see it.

The gap between deciding and doing is where quarters go to die.

And nobody owns that yet, until very soon.

[Matakats @matankatzzzz]

@uriel_builds @huggingface Insane today is the new normal

[Matakats @matankatzzzz]

@huggingface said no to Nvidia at $7B last year because they didn’t want one dominant investor.

This week they reportedly sold to Nvidia for $12.9B.

~85x revenue.

Same week OpenAI published benchmarks claiming its own chip beats Nvidia’s.

This isn’t an acquisition. It’s insurance.

When your biggest customers start building their own chips, you stop selling picks and shovels and start buying the mine.

Every founder building “infrastructure” should ask: am I the pick, or am I the mine?

[Matakats @matankatzzzz]

@neerajjj6785 The ones saying building the product have no clue.

Maybe 5 years ago that was the case. Today, getting your product in front of users is far more valuable and difficult.

[Matakats @matankatzzzz]

@RamanRjha4647 Nonsense

[Matakats @matankatzzzz]

@saaayyyaaan You can have the best landing page out there and still won’t generate revenue.

What it comes down is the messaging . If a potential client can resonate with something you are looking to solve, he will make sure to give you a shot.

[Matakats @matankatzzzz]

https://t.co/aCOwAribmI

[Matakats @matankatzzzz]

Everyone is building “the company brain.”

YC put it in its RFS. Microsoft, Glean, and a dozen startups are racing to be the one memory layer every agent reads from.

I think the premise is wrong.

There is no company brain. There are layers.

Your brain doesn’t have one department for everything. Vision, motor control, memory, language. Separate systems, one organism.

Companies are the same:

•Data intelligence: what is true
•Business intelligence: what happened
•Execution intelligence: what we decided vs what we’re actually doing

Dump all three into one memory and you get a very expensive search bar that can’t tell you why the launch slipped.

Gartner says 40%+ of agent projects get cancelled by 2027. Not because the models are bad. Because “remember everything” is not a job. Catching drift is a job.

The winners won’t build the whole brain. They’ll own one lobe and do it better than anyone.

[Matakats @matankatzzzz]

https://t.co/uQSiIFCbzg

[Matakats @matankatzzzz]

@OpenAI is launching a phone.

Here is why:

1.Q2: $6.7B revenue, $12.3B loss. Consumer subscriptions don’t cover the compute bill. Apple and Google take 30% of every ChatGPT subscription sold through their stores.

2.They paid $6.5B for Jony Ive’s team. You don’t buy the iPhone’s designer to build a smart speaker.

3.This week they published benchmarks claiming their own inference chip beats Nvidia. First-party silicon is step one of every hardware company.

4.Ming-Chi Kuo says an OpenAI “AI agent phone” could hit mass production in H1 2027. 30M units by 2028.

The screenless “third device” story is the decoy. The real move is owning the OS, the app store, and the 30%.

Apple built the iPhone to sell hardware. OpenAI will build one to escape the App Store.

[Matakats @matankatzzzz]

https://t.co/DdwWuJr9Jw

[Matakats @matankatzzzz]

https://t.co/hHs1cbtXtz

[Matakats @matankatzzzz]

@ben_sage Are you really going to achieve what you intended to achieve?

[Matakats @matankatzzzz]

@suni_code What do you mean by regular? Also generalizing your argument is not defending it either.