Exploring the frontiers of Technology and AI
The last seven days has been undoubtedly the most consequential series of events in tech history.
It started about a week ago with the leader of the most important company in
the world, Jensen Huang of NVIDIA.
NVIDIA worth $5.8 trillion, the number one most valuable company in the world,
being removed in the most brutal fashion because he had an affair with an alleged Chinese spy.
And then that snowballed into affair number two and then affair number three.
And at this point, you're thinking, well, obviously you do need to remove him.
But it's not quite that simple because the alleged spy accusations,
the evidence actually has come up empty. There has been no obvious evidence
and the intelligence services have reported the same. That's number one.
And then number two, on the same day of his ousting, Anthropic and OpenAI,
the leading AI labs, pushed for a bipartisan bill to ban chip exports to China.
Which smells like a lot of this could have potentially been coordinated.
Putting all of that aside, we have to answer a few questions.
Number one being who fills the gaping hole that's at NVIDIA,
the company, and who fills the gaping hole that is left in the rest of the entire
economy, which indirectly is the economy itself.
You see, a lot of the investing landscape right now relies on a lot of investments
that Jensen Huang made himself via NVIDIA.
You're looking at neoclouds that rely on backstops and banking NVIDIA effectively
became a bank to all these companies.
You then have all the AI labs and hyperscalers who have bought hundreds of billions
of dollars worth of commitments for NVIDIA GPUs. Can they still deliver on that?
And then you have the rest of the world who have either directly or indirectly
invested as NVIDIA, whether through their retirement accounts and 401ks,
or whether that is through personal investing.
The idea here is NVIDIA has typically been seen as a safe bet,
and yet they've shed around $2 trillion off their market cap since the start of October.
And then there's the question of competition, NVIDIA's competitors can smell
blood in the water. AMD, Meta, Microsoft, Google, Amazon.
OpenAI, Anthropic, they all have their own chip teams that are racing to try
and eat into NVIDIA's market share, Google being the closest to it,
and they're going to seize the opportunity.
And finally, there's the conversation around open source.
It's no secret that Jensen Huang was one of the biggest advocates for open sourcing
AI as much as he can to get AI accessible to anyone and everyone.
There is an obvious risk here, especially with OpenAI and Anthropic pushing
their recent bill, that the concentration of AI power could go to one,
two, maybe three companies.
And what does the world look like in a world where that concentration of power
exists amongst very few people, whilst the rest of the world doesn't really
get to have a say in how AI rolls out?
So I want to give you my thoughts on how things are going to play out from an
investing perspective, but also the technology perspective.
The king at the end of the day has been removed from his throne and that's going
to have massive ramifications on everyone.
So let me first start by unpacking some new bits of evidence which kind of suggests
that this was sort of a coordinated event.
The affairs exist and Jensen Huang for all intents and purposes has admitted
to at least one maybe two of them.
I don't know about the third one but the idea is he has done that and those
acts are reprehensible.
The way that this has played out is sort of weird. The first is this statement
from a government representative, John Molinar, where he goes,
Jensen Huang's firing from the company he founded is an unfortunate end that
could have been avoided.
It is in our national interest for NVIDIA and our other AI champions to succeed.
But American companies should never jeopardize our nation's security by giving
Chinese companies access to their most advanced chips.
NVIDIA must now change its corporate practices, and appoint leadership that
makes national security its top priority.
Now, I don't necessarily disagree with any of these statements.
I think it makes sense, but it's sort of the way that it's stated makes it seem
like this was an intended function of the US government.
And indeed, if you actually look at when the investigation into Jensen Huang's
affair actually started, it wasn't last week, it wasn't the last couple of months,
it was almost over a year ago.
And it wasn't just the US government that had surfaced these rumors,
it was the rest of the internet. I remember actually seeing tweets at the end
of last year of Jensen Huang with Alice Huang, the alleged Chinese spy.
And they're out together and people were like, this seems kind of weird.
They're very close and affectionate.
And apparently rumors has been spread in the company and in the ecosystem in
general about Jensen's alleged infidelity.
Now that doesn't justify the cause, but it points to the fact that people were
kind of already aware of this, especially the government themselves.
Then the question then becomes, was Jensen selling illegal chips to China?
Because it's kind of a gray area. You see, the government has flip-flopped between
allowing NVIDIA to sell chips to China and then stopping them from selling chips.
And then more recently, it's like, okay, you can sell some chips as long as
they're kind of poor performing bad chips.
And keep the best chips for America. So the policies have been changing quite
frequently, and Jensen has been a pretty public advocate as to being pro-selling
certain chips, poorer-performing chips.
And his whole thesis here is that if you sell China poorer chips,
they can get hooked on the NVIDIA or American ecosystem, and that prevents them
from building their own infrastructure, which could arguably be independent,
more performant for Chinese manufacturing setup, which could end up beating America.
So the idea here is to keep them hooked on the American infrastructure so that
we can keep an eye on our enemies. Now, the second bombshell bit of evidence
is a review of whether Alice Huang was actually a Chinese spy.
And it seems like every government and intelligence authority has said that
there's inconclusive evidence that this is even the case. This tweet here from
David Sachs, he kind of runs through the list. He says, AG, I think that's Attorney
General, says there's no conclusions reached.
CIA says not yet convinced. DNI says no, not entirely accurate and overtalked.
In other words, saying that this has been blown massively out of proportion.
And that leads me to the third bit of evidence, which is Anthropik and OpenAI
are reportedly bonding and forming an alliance to push forward a regulation
or law to impose a ban on chip exports to China.
Now, the initial and obvious reaction to this Huh.
That's probably good, right? Like if we ban them from getting access to our
chips, they won't be able to build breaking edge AI and we will end up winning the race.
And certainly there is an argument for this, but it seems pretty coincidental
that they're accelerating this right when the key advocate for open source and
selling chips to anyone and everyone, not just China, is hurting.
And make no mistake, the outcome from this would be there's an increasing concentration
risk being placed in America itself. And guess who the two biggest labs are
that are buying these chips?
Synthropic and OpenAI. So it would disproportionately benefit them.
Less competition out there. Remember, Chinese AI labs were the biggest advocates
for open models. And a lot of American companies and founders actually use Chinese
models. So I can assume they're looking at this and thinking no one wins either way.
And the last thing I'll say is, if Alice Huang was actually a Chinese spy,
she did an absolutely terrible job of doing it.
She made no effort to try and hide the fact she was flying to China every other week.
She was being close to Judson Huang in public. She wasn't exactly trying to hide it.
She was posting social media photos on WeChat, which is the Chinese public social
media site of her and Judson Huang flexing.
She was publicly attending investing meetings with Chinese AI labs to be like, hey, can I invest?
And the idea here is, if she ended up not being a Chinese spy,
that wouldn't surprise me. But let's now move on to what the state of affairs
is right now, which is the leader of NVIDIA, Jensen Huang, has been removed.
What happens to NVIDIA, the company? Now, there's an interim CEO.
Her name is Colette Grass, and she is the CFO at NVIDIA. She has been at NVIDIA for.
Almost as long as Jensen. Not quite. I think she joined in 2012 or 2013.
And listen, my initial reaction to this is, it makes sense.
It's sensible. She is the only other person at NVIDIA that kind of knows the
ins and outs of everything. If there's anyone that can kind of calm the waters
over the last recent couple of weeks and months, it's going to be her. So it's a good pick.
But she can't be a permanent replacement.
And that brings into the question, who replaces Jensen Huang? And in my opinion...
There isn't really anyone, if I'm being honest. Jensen Huang was a one-of-one founder and CEO.
But let me plead my case here. Jensen Huang started this around 33 years ago
from a Denny's. This is like a fast, casual pancake restaurant in the US.
And he was a dishwasher there.
And he would sit in a corner side table every single day after work or before his shift.
And he would plan and try and figure out how he can kind of enter the chip industry.
And back then, there was no AI.
There wasn't even any modern day gaming. There was no Call of Duty.
He was just a man with an idea. He saw the computing kind of wave come over
and he was like, we're going to need hardware for this.
So he went from that to building a $5.8 trillion company.
Now, there are many things that get seeded in a singular founder if you're able
to do this. And I'll give you a few examples here.
Number one, he takes crazy risk and bets when every single sane person would tell him not to.
And because he's done that successfully over the last three decades,
he's been able to be a key figure in gaming, crypto, as well as AI.
Now, I'll give you a few examples.
Back when NVIDIA was just building these GPUs, graphics processing unit,
they didn't really have an idea of why or when or who they should serve.
And then gaming exploded and they oriented their entire bit of hardware to focus on that sector.
And then when crypto happened, NVIDIA and Jensen Wang specifically made an early bet.
And he said, I think crypto is going to be a huge thing. I think blockchains
are going to be a very important technology.
We need to orient our hardware to fit those things because it already is pretty good.
He did the exact same thing for AI. And I want to show you this crazy image
here where he's giving the first set of NVIDIA GPUs at the time.
To none other than OpenAI, who are building or training their first model.
This is before they even got off the ground, before they were even a legitimate
company. No one knew them at the time.
But also look at who's signing the GPU.
It's none other than Dario Modi, who worked at OpenAI before he then left years
later to set up Anthropic and build one of the leading AI research labs.
So there's a lot of law behind this. And you could argue that OpenAI wouldn't
exist today if Jensen Huang hadn't made a major bet.
But then there's this final crazy example where back when AI just wasn't a thing
at all, when it was just a few developers experimenting on things,
Jensen did a crazy thing.
He purchased this company that had no influence on the business that earned
NVIDIA 90 to 95% of its revenue. At the time, they were focused on gaming.
And he purchased this company for so much money because he was like,
this company will enable us to shove more transistors and run more processors
in our GPUs, which I think will help AI achieve its goals.
And back then that was a crazy bet to make, but he still did it.
And it ended up being one of the greatest enablers for allowing NVIDIA to build
its monopoly and market share and become one of the leading AI providers in the world.
You can't describe NVIDIA. You can't describe the AI semiconductor space.
You can't describe the AI space without having Jensen Huang there.
I guess that's my entire point.
He was there at every single part of the sector, at every single part of the
industry's development, and he had a say and an influence into how it materialized today.
And you have to sort of have that visionary scope.
Of how things are playing out in order to influence what the tip of that spear
is going to look like when you build an industry from zero to one.
Without Jensen Huang, we wouldn't have had the AI world that exists today.
And that's not saying that we don't have so many other technical geniuses that are out there.
I'm just saying for NVIDIA, for GPUs, for hardware, of which the entire AI economy
and economy in general relies on right now, it's a big loss.
So as for who can fulfill his new CEO role.
There's a few names being floated around. In an ideal world,
people are saying Lisa Su of AMD, who also happens to be Jetson Fong's cousin,
that's going to be a really awkward Thanksgiving meal that's coming up, should take the helm.
But I don't think she's going to leave AMD. AMD is a publicly traded company
and also a major competitor to NVIDIA. I don't think that's going to happen.
There's rumors of Nikesh Arora, who leads Palo Alto Networks.
He has good experience in building hardware, but not to the scale of NVIDIA.
His company is like, Like, you
know, sub $1 trillion, NVIDIA is multiple trillion dollars, or used to be.
And then there's the floating of maybe Apple's head of hardware or AI hardware,
Nick Sruji, who has maybe the experience of spending all this money and being
able to understand this.
There's internal suggestions, such as Ian Buck, who is basically the most technical
guy that understands how to build bleeding edge GPUs and maybe,
maybe has the risk appetite. Again, it's going to be hard to replicate.
And then you have other suggestions, which are the head of AWS,
Amazon Web Services, cloud computing platform, which is kind of like the closest parallel analogy.
It's not quite the same. He doesn't actually build the chips,
but maybe if he has experience deploying hundreds of billions of dollars worth of CapEx.
In Amazon alone, maybe he could kind of mimic or do the same kind of deal-making.
That's the other thing, by the way.
The deal-making that Jenshuang was able to pull off required him flying to Taiwan,
where there's this company on this tiny island, TSMC,
who is responsible for 99.9% of all semiconductor chip creations.
I'm not just talking about NVIDIA, by the way.
I'm talking about AMD. I'm talking about Google's TPUs. all of them gets it
manufactured in this tiny company that exists on this tiny island in the middle of the ocean.
And so what Jensen Huang would do is he would go there and he would shake hands
with all the heads of the supply floor manufacturing of all the different heads
of TSMC to agree on deals to acquire capacity.
And it wasn't just as simple as like, hey, you'll give me the capacity, right?
He would create reasons for TSMC to build new business lines.
This is my whole point. He would go in and he would say, hey,
I have engineers which discovered a new way to re-architect this GPU.
And it doesn't exist at scale yet, but I think you guys can create the machinery
that will allow me to do it at scale.
I want you to trust me and take a bet on doing this.
And so they would discuss over, I don't know, eating fried chicken on the street
markets where they would drink beer and discuss this into the early hours of the morning.
And then they would make a commitment. They would say yes.
And this commitment was a handshake deal. Now let's move on to what this means for the competition.
This now presents a really exciting opportunity for companies like AMD.
Like Google, like Intel, like Amazon, like Microsoft, like Meta,
like OpenAI and Anthropic, who are all building chips.
And they have really impressive chip teams to encroach onto NVIDIA's market
monopoly around 80 to 90% of the entire chip market.
Now, if I were to stack rank them in my head, I would put.
AMD and Broadcom near the top. Now, Broadcom are the guys that basically build Google's TPUs.
Google doesn't build it in-house. They kind of design the blueprints of what
the TPU is going to look like.
And then they send that to Broadcom. Broadcom are the guys that actually,
you know, put it together, kind of layer it out. And then they either send it
to TSMC and say, okay, here's the thing, build it out.
But Broadcom plays a huge role in kind of like architecting that.
Then you look at AMD and they kind of have more of an in-house thing with the
same endpoint being TSMC, go build the thing, please, right?
So if you move kind of below that, you have Google, obviously, themselves.
Now, when you look at GPU market share, the number two player after NVIDIA is Google.
They're selling hundreds or tens to hundreds of billions of dollars worth of
TPUs over the next couple of years.
So they're the next major competitor. And then beneath that,
you have some of these smaller chip companies like Etched, which we've mentioned
before on the show, which is like a smaller startup, they have a longer way to go.
And then you have the likes of open-air and Anthropic, which,
listen, they don't have the experience or expertise or the resume to say,
listen, we can build chips as fast and as to a larger scale as NVIDIA.
But I would say two things to that.
Number one, they don't need to build it as fast because the chips that they
build aren't going to anyone but themselves.
And it's going to be hyper-specialized for their models, which means that they'll
be able to jump up in intelligence levels for every single model because they
have their own specialized hardware. They've optimized the software as much
as they can, and now they need to do the same for hardware.
But the second thing is, they have AI that can just be pointed at the chip and
say, hey, can you build me an NVIDIA quality chip? Actually,
can you build me an NVIDIA quality chip that's 2x better than their recent model?
And so it's this weird world where they can kind of like point smarter models
as they get successively smarter to build a better chip that can then be fed
to build and train the next model. And it's this whole recursive self-improvement thing.
Now, what does the removal of Jensen mean for the rest of the AI ecosystem?
Because it's no secret that NVIDIA and Jensen specifically backstopped a lot of the AI economy.
They backstopped a lot of the neoclads, which are smaller cloud or GPU providers.
And Jensen was like, listen, if you default, I'll cover you for up to 25 to
50 to maybe even 100%. The bad alternative being if you remove the CEO in a
black swan event, which is exactly what's happened, a lot of these neoclads may lose that funding.
And so you might see a kind of collapse of cards, at least in the near term
over all these different deals, which NVIDIA has promised to finance.
Now, there's one person right now, which will determine which way this flips.
And that is the interim CEO, Colette Grass.
She now has a really hard task of settling and calming the waters.
Please no affairs, Colette, and making sure that the ship keeps chugging along,
that all the deals that exist today get served.
Now, where this could end up being a bad thing is if she re-reviews the deals
and she reneges on it, or she pulls back from deals that have been handshaked
on and hasn't officially been committed to because she wants to kind of like.
Conserve resources or whatever that might be.
And she's infamously been known to be very risk-averse, which isn't a bad thing,
by the way. If you're the CFO, you kind of should be, and you should have a
Hail Mary CEO that kind of wants to go for it, and they balance each other around.
If she's both the CFO and the CEO, this might kind of pull back on a lot of
the NeoCloud deals that we see today. And then if you're an investor like.
Me, for example, that has bought NVIDIA and has held NVIDIA for a while,
you might be questioning, well, where do we go from here?
The positives, the pros are they backstop 80% of the entire chip market.
They are the chip market.
And they have all the talent that is required and needed to continue building
and delivering on these GPUs.
And thankfully, they have the systems and the handshake commitments that NVIDIA
or Jensen Huang has already made.
Now, the question then becomes, what happens after it? What happens to building
the next generation of NVIDIA GPUs, which are pegged to release,
I think sometime next year?
What happens to the generation after that? Remember, Jensen Huang as leader
was always 10 steps ahead of the competition.
He was always thinking about several generations ahead, what that GPU might
look like. I think the most furthest ahead is Feynman of 2028.
I think that was the name of the GPU.
What happens to those projects? If those end up getting stalled,
if those end up getting kind of pushback or re-architect it,
that might bring into question whether NVIDIA as a long-term investment is a good bet or not.
And of course, just instinctively, if you have the safest AI stock in the entire
market crash around 20% to 30% over the last month and lose around $2 trillion
of market cap at one point,
you'll obviously be questioning all your other AI investments,
which kind of in some way depend on NVIDIA as a whole.
All right, the final two things I want to cover are open source models and what
this means for their future, and then how this plays out with China going forwards.
Now, on the open source model front, there has been no bigger advocate for building
AI in open source than Jensen Huang.
And his whole reasoning behind this is intelligence shouldn't just be concentrated
in the hands of a few very rich, powerful companies or individuals that run
these companies. And he was alluding to the likes of Anthropic and OpenAI.
NVIDIA invested a lot in open source. Most recently, they acquired Hugging Face
for, I think, $13.6 billion.
Hugging Face being one of the largest distribution platforms for open source.
They host open source models.
And open source is great because you are able to own your own AI.
But it's also cheaper, it's also more efficient, and it levels the playing field.
And most recently, ironically, you can see that being done primarily by China,
who all their AI labs, once they've been building their models,
have in large part open sourced or open weighted the entire thing.
Now in this new world without Jensen Huang, a lot of these open source initiatives
and companies will lose the wind beneath their wings because Jensen Huang was
investing and backstopping a lot of that.
And Dario, CEO of Anthropic, has famously been opposed to open source.
His argument, which I think is a conceivable one, is if you put these powerful
models in the open source world, it means that anyone and everyone can just
re-architect it to potentially perform malicious events.
Now, when it comes to China, it's a coin toss, if I'm being honest.
On the one hand, NVIDIA and Jensen Huang's whole thesis of sell them.
Cheaper, lamer GPUs, keep them hooked on the American GPU system so that we
always know where they are, where our competition is, and make sure that we
always get the best GPUs and we can stay ahead.
The evidence so far has suggested that that has been the case because Chinese
morals have consistently been behind.
And actually, more recently, over the last couple of months,
they fell even more behind because they haven't been able to distill all American
frontier labs. So they're losing on the hardware level and they're losing on
the frontier intelligence level.
Now, you don't have to trust my word for this. You can look at China itself,
whose government has actually decreed that their AI labs should not purchase
NVIDIA or American chips.
And It should rely more on Chinese-made chips, primarily from Huawei.
And this then unpacks the gray murky area, which is, there was this viral tweet
that went out, I think, on Thursday or Friday of last week, where Will DePue,
who was an ex-OpenAI researcher,
did a forensic analysis of six of the top Chinese AI labs or their models, and he determined.
Each and every one of those models, or the majority of them at least,
were trained on NVIDIA Blackwell GPUs. This is crazy because the NVIDIA Blackwell
GPUs are banned from China.
On the other side, I think what's going to play out now is China is going to
double down on their own hardware infrastructure, their own chips,
and they're going to invest. And so that means for a while, let's call it a
year, China will be behind in AI.
And we're all going to think, we have the lead. That's great.
But I think what's actually going to end up happening is after a year,
or maybe it's a year and a half, I don't know how long this is going to take,
we'll suddenly see a Chinese model that is super intelligent,
but is also cheap as hell.
And so far, we've seen that, but we haven't seen a Chinese model that has beaten the frontier labs.
I think if we go down this path, there is a non-zero chance.
I don't know if it's actually going to happen, but there's a non-zero chance
that China can not only catch up, but beat us.
Whereas prior to that you had a conceivable argument with Jensen's strategy that they would never
catch up that they would never beat frontier American intelligence and I guess
that is the entire digest of what's going on with NVIDIA with Jensen Huang and
how all of this might play out so please keep engaging and yeah I'll see you on the next one.