Is America in Trouble? Yes.

Welcome to the No Longer A Nincompoop with Nofil newsletter.

Here’s the tea 🍵

• Why NVIDIA crashed 📉

• Why NVIDIA will matter more than ever 🧠

• Why China might render NVIDIA useless 🧨

• Would I use DeepSeek to code? 👨‍💻

• How the US might kill open source 🔫

• Why DeepSeek became so popular 📈

What on Earth is going on?

I didn’t think after my newsletter last week so many crazy things would happen.

Before I get into the details, a few things I need to clarify.

NVIDIA

Last week, I posed the question:

If DeepSeek is able to replicate OpenAI’s models at a fraction of the cost, what is the moat?

1) The only thing we know was a fraction of the cost was the final training run. The ~$6M figure was not for the entire creation of the R1 model.

I’m seeing so many people online say this and it’s annoying me.

Now, it is impossible for us to say how much the building of the entire model cost.

Was it less than OpenAI’s o1?

Probably.

Was it a tiny amount like $6M?

Definitely not.

2) Following my questions around moat, literally the next morning, NVIDIA stock, as well as chip stocks, tanked.

At first, people were emailing me confirming my thoughts that DeepSeek’s success is bad for NVIDIA and the market was reflecting that.

I thought I was smart for a second lol.

It turns out it wasn’t related to DeepSeek at all.

Trump said he will impose tariffs on all chips coming out of Taiwan a few days after the stock crash [Link].

The US government is also considering tightening export controls to China, so they can get even less NVIDIA chips (legally).

(We already know China gets heaps of NVIDIA chips through Singapore, considering Singapore accounts for ~20% of NVIDIA’s revenue. I wrote about this months ago.)

It seems like the preemptive crash was caused by people behind the scenes realising this was coming and selling (Pelosi style); basically insider trading.

I don’t think the release of DeepSeek really had anything to do with it, and even if it did, the effect didn’t last long as the stock bounced back up the next day.

The real question that I have is:

Why is Trump doing this?

Why is he placing tariffs on Taiwan?

I have absolutely no idea.

If someone does, please, email me and let me know. I am very curious.
Email: [email protected] 

The way I see it, this does two things:

1) It pushes Taiwan closer to China

2) It advances China’s push to become less reliant on NVIDIA chips (more on this below)

This being a response to the open sourcing of DeepSeek seems like a monumental overreaction. But, this isn’t even the worst of it as I’ll discuss further below.

More importantly, however, is the idea of moats.

When I questioned sky high valuations last week, I did so without knowing about something called Jevons Paradox.

Jevons Paradox

It basically says that when technology makes a resource more efficient to use (meaning we can get more economic output per unit of the resource), this typically:

• Reduces the cost of using that resource

• Makes processes depending on that resource more economical

• Opens up new uses for that resource

• Increases overall economic growth

As a result, total consumption of the resource actually increases despite the improved efficiency.

William Stanley Jevons first observed this in 1865 when studying coal use in England.

He noticed that more efficient steam engines, rather than reducing coal consumption, led to increased coal use as steam engines became economically viable for more applications.

GPUs might be the next coal.

By understanding how we can use them in an extremely efficient manner and essentially getting more output from a single unit, we can open the doors for more use cases.

Simply put, if we can do more with less GPUs, more people will use GPUs to do things for them because of how much can be done with so few.

This will lead to an increase in demand for GPUs and therefore more usage of them.

Improvements in resource efficiency will lead to increased consumption of said resource, rather than a decrease.

Considering the fact that companies use such a massive amount of GPUs for inference rather than training, this makes the likelihood of Jevons paradox even more convincing.

Side Note:

Training: Using chips to train AI models. This doesn’t necessarily require the bulk of chips that companies are hoarding.

Inference: This is serving the actual AI model for people to use. The bulk of chips bought by companies is used for inference - to actually serve the model to the public.

If more people are using AI then we will need more inference compute, meaning we will need more GPUs.

So, this means NVIDIA is even more important now than ever right?

Buy the dip?

Yes, I think so.

There is, however, one other problem.

China.

China might screw up everything

DeepSeek’s success is a lot more important than you think; more than most people realise.

There are a number of things to go through here so I’ll try and cover them in an order that makes sense.

Chips

Since we’re already talking about NVIDIA, let’s talk about chips.

Did you know that DeepSeek trained their AI models on NVIDIA chips?

…

I know what you’re thinking.

Is this guy stupid?

Of course I know that.

Well then, let me ask you this.

Did you know that DeepSeek uses Huawei chips to actually run the inference of the model?

Meaning, when you go to their website and ask a question, the AI is running on Chinese chips, not NVIDIA.

This is a VERY big deal.

Inference, that is hosting the model for people to use, is the main use of a large number of chips.

When OpenAI buys thousands of chips, most of them will go towards hosting their models so users can actually use them.

Chinese companies are figuring out how to host and run these models without the need of NVIDIA chips.

> 5% performance loss
> 70% cost reduction

These figures are only going to get better.

Chinese companies are already serving R1 for pennies.

The only company serving R1 cheaper than that is DeepSeek themselves, who are also using Huawei.

American providers are a tad more expensive.

The chips aren’t quite good enough to actually train models… For now.

Either way, I can’t overstate how important this is.

And this isn’t just DeepSeek working with Huawei.

Following DeepSeek’s success, Li Qiang actually met with the founder of DeepSeek about the future of AI in China.

This is the #2 guy in China.

DeepSeek now has the governments full support.

A monumental overreaction

DeepSeek R1 is a good model. No doubt about it.

If I wanted to code a project, would I use it?

No.

I’d still use Claude.

Like any tool, I’ll use what I think is the best tool for the job.

Right now, Claude is still the best coding AI one can use for only $20/month.

Is R1 good at coding?

Yes, it is. But, using these models is more complicated than a simple better/worse comparison. I will write more about this soon.

For example, R1 got a 2x speed boost and the code written for the boost was written by R1 itself [Link].

But, I digress.

Does the release of R1 mean China is ahead of the US in terms of AI?

No, it doesn’t.

DeepSeek is playing catchup.

OpenAI is still the leader in the space, make no mistake about it.

But, the reaction of the US community to R1 has been nothing short of insane, and I truly mean insane.

At first it was funny (and a bit sad), US investors and politicians who clearly know nothing about AI were fear mongering, calling it a copycat.

It is no longer funny.

You know why?

The US tightened export controls on China and imposed tariffs on Taiwan.

This wasn’t enough.

Congress has actually created a bill that would kill open source AI.

I’m not only talking restricting releasing AI models generally like Meta, which would come with a penalty of 20 years imprisonment.

I’m talking even downloading the weights of an open source AI model from China.

Sharing research online that could be viewed and used by China?

The penalty for violating this?

• $1M for an individual

• $100M for a company, plus 3x damages.

But, the bill also makes it an "aggravated felony" to do these things, meaning any noncitizens (e.g. CAN / FR / UK) involved can be deported.

It’s because of open source research that OpenAI, ChatGPT and this newsletter even exist.

Google released so many important research papers over the last decade that have led us to where we are now.

Without open source research and collaboration, humanity would not be where it is today.

You can read more about this bill from Ben Brooks [Link].

The level of delusion at play here is hard to put into words.

The leaders of OpenAI, Anthropic and morons like Alexandr Wang and Vinod Khosla, are going to destroy open source AI with their fear mongering and lust for power.

They don’t want anyone but themselves to have control over AI.

DeepSeek has done more for the “little guy” over the last two years than any American lab bar Meta and their Llama releases.

Competition is a good thing.

Just look at what OpenAI has done since the R1 release.

• Released o3-mini

• Updated the free version of ChatGPT to use o3-mini

• Released “Thinking mode” to free users

They’ve got bloody Eric Schmidt talking about open source and he was one of the biggest advocates for government control last year.

They’ve even got Sam Altman considering open source, something he hasn’t done in years.

Can you imagine if OpenAI had open sourced even some of their research first?

Not even the models, just the research.

We wouldn’t even be talking about DeepSeek. It wouldn’t even matter!

OpenAI had the chance to herald in a new age and be loved by the people.

But, in their greed for monopolisation and money, they lost their chance.

Now, everyone is flocking to a Chinese app.

Mind you, banning imports has nothing to do with letting China get US tech.

It seems it has everything to do with making sure people only use OpenAI or an American lab.

Gee, I wonder why.

DeepSeek has gone mainstream at a level only seen by ChatGPT.

It’s #1 in the app store.

This makes you wonder.

How did this even happen?

How DeepSeek became so popular

DeepSeek is a household name now.

My barber is asking me about it. My dad is asking me about it.

Tens of millions of people around the world are using it, especially in Asia.

But, how did it get to this?

How did DeepSeek, essentially unknown to the average person, become so popular?

A few things to consider:

• It’s a very good model

• It’s open source

• You can see it’s Chain of Thought (CoT)

Finally, there is one point that is not very apparent.

A lot of people really hate OpenAI.

Like, seriously hate the company.

If you’ve gone into the comments section of TikTok and see how people talk about OpenAI, you’ll understand what I mean.

Not only did they go from somewhat open to completely closed source, they have done nothing but advocate for government control and the monopolisation of AI.

To most people, they’re the company that stole everyones work and are profiting off of it.

They also just did something nigh inconceivable.

They accused DeepSeek of stealing from ChatGPT.

Yes.

The company that built itself by using everyone else’s data without repercussions, is now accusing someone else of stealing from them.

Talk about hypocrisy.

There is a very large part of non-tech social media, be it Twitter, Instagram or TikTok, that share two things in common.

1. They hate AI and in turn hate OpenAI

2. They hate anything from America because “capitalism bad”, and so they like DeepSeek because it’s bad for America

Check out some of these for more info [Link] [Link] [Link].

Once again, I feel like I’ve barely begun and it’s already way too long. More newsletters will be out very soon and premium readers will get them first of course 🤗.

Please consider supporting this newsletter or going premium. It helps me write more :).

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}