# Intel Should Second-Source Nvidia

In August 2025, the Wall Street  Journal reported that the United  

States Government is considering  to take a financial stake in Intel.

A bit later, Bloomberg reported that the  

US Government is looking to take  a 10% stake via CHIPS Act money.

The Intel situation is fluid. And my style  of work is not suited for these strange,  

24-second news cycle. Things change on a dime.  But I have an idea that I want to write about.

One of Intel's problems is that its old monopoly  is dying. So to revive it, I think Intel should  

get access to Nvidia's crown jewels. In today's  video, a modest proposal regarding Intel.

## Waving the Bully Stick

By the way, I just want to let you guys  know that I'll be in Arizona for SEMICON  

West in early October. If you're going,  shoot me an email and let's meet up!

So what seems most likely about to  happen is that the US Government  

and a few aligned investors  first take a stake in Intel.

The government puts up money to finish the  fab in Ohio. Then raises tariffs to a high  

number with some carveouts to satisfy  the crowd. And then they will apply  

pressure to the tech giants to get them  to throw some orders at Intel Foundry.

There's a comment on the WallStreetBets thread on  the story that succinctly lays down the sentiment:

> Dude is 100% gonna "nudge" Apple  and Nvidia to use their 14A foundries,  

this is gonna skyrocket by EOY 2026 in the 100s

Doug O'Laughlin of Fabricated  Knowledge agrees and adds that  

pressure might also be applied onto the  semiconductor equipment companies too:

> Additionally, forcing  semicap companies like KLAC,  

Applied Materials, and Lam Research to  invest and give resources in exchange  

for approved licenses is another  example of a carrot and a stick.

A few companies have already started to turn  on this. For instance, just as I write this,  

Intel has announced that SoftBank agreed  to make a $2 billion investment into them.

Considering how Masa invests, I am not surprised  he's first through the gate. There is news that  

he was in talks to buy Intel's fabs. Maybe  that will happen by the time you see this.

## Coalition of the Unwilling

As I said before, this seems to be what  the plan is. A Girl Scout cookie drive  

to raise money for Intel. Raise  tariffs. Shake out some orders.

I have no doubt that the US Government will  speak loudly and waggle a stick. But as Wei  

Zhou of SemiAnalysis put it in a recent episode of  the Transistor Radio podcast, the result would be  

a "coalition of the unwilling". There because  they have to be, not because they want to.

Bullying the seven American tech giants  - Meta, Google, Amazon, Apple, Microsoft,  

Nvidia, and Tesla - to come together would be a  monumental effort. When have we seen all seven  

or even a majority of these seven work together  on a business initiative that has gone anywhere?

I am reminded of US Memories. In the late 1980s,  the Semiconductor Industry Association tried to  

assemble an American coalition to fund an  effort to make DRAMs in the United States.

Seven American companies nominally agreed to join  including three major computer manufacturers.

But less than a year in, DRAM prices started  falling. With supply more than fulfilled,  

the DRAM buyers like Apple and  Sun lost interest or worse,  

ran out the clock - hemming and  hawing until things fell apart.

Intel has yet to produce a leading-edge chip for a  third-party customer. A foundry is like a bakery.  

Right now, Intel's cake menu is limited because  its kitchen has for so long been optimized just  

for Intel x86 CPUs. Expanding a menu takes  years, and many skills, and many mistakes.

Now we suddenly want them to take on orders ...  from seven of the world's biggest, most demanding  

customers ... for hundreds of thousands of the  most sophisticated cakes ever made ... vastly  

different cakes too ... and deliver those cakes  on time ... and in excellent working condition?

## The MP Materials Scenario

I want to discuss a few other scenarios that  has come to my attention in recent days.

Doug mentioned MP Materials' rare  earths deal with the Department of  

Defense. Rare earths have many  uses, but the most significant  

are powerful permanent magnets used in  electric motors, hard drives, and more.

The People's Republic of China  famously monopolizes the supply  

of rare earths through superior refining capacity  

and market-distorted low prices. The  whole market is worth about $3 billion.

The US Government understands this. So classic  industrial policy from Europe and Asia,  

we back a national champion. MP Materials  owns the Mountain Pass mine in California.

In July 2025, the US Department of Defense  announced a $400 million investment into MP.  

DoD also agreed to acquire 100% of the output  from a magnet refinery that MP is building,  

with the price set at a cost-plus basis.

This will probably work for rare earths -  especially if we accept the environmental  

damages that we know will come with refining  them. But rare earths are commodities dug out  

of the ground. They are not chips imbued with IPs  provided by third-party companies and customers.

Moreover, the numbers involved are just  phenomenally larger here in the Intel  

scenario. People estimate that Intel  needs something like $40 billion to  

finish its fabs. Moreover, the fabs  must also be upgraded so to stay at  

the leading edge. So continual  capital influxes over the years.

## The TSMC Scenarios

There have been a few proposals that  involve TSMC, the market leader.

A few weeks ago, the Taiwanese news media reported  that one of the conditions that the US wanted from  

Taiwan in exchange for a lower reciprocal tariff  rate was that TSMC take a 49% stake in Intel.

And a few months ago, before Lip-bu Tan  became the CEO, there was news in the New  

York Times that TSMC could take control of  Intel's plants as part of a larger split.

A third proposal also comes to mind. Writer and  friend of the channel Tim Culpan of Culpium has  

long predicted that TSMC will someday spin off its  Arizona assets as a publicly traded subsidiary.

So how about we merge TSMC's US assets with  Intel's? We can also fold in GlobalFoundries,  

Texas Instruments, and all the other  semiconductor companies too! Yeah! Let's do it!

On a technical level, TSMC and  Intel's process nodes are different  

and incompatible with each other. TSMC  wants nothing to do with Intel's nodes.  

If they got into an Intel fab, all  they would want is the equipment.

And the culture clashes would make all  the toxicity of the Arizona build look  

like a walk in the park. I asked a TSMC manager  before about a TSMC-Intel merger, and he said  

that Intel'ers might be afraid of TSMC, but  TSMC managers are afraid of Intel employees.

So no, I don't think these three scenarios  will work. So here goes my new idea.

## What Funded Intel's Fabs

The thing that I want to hone on is what I believe  to be the source of Intel's massive profits.

Those profits do not come from semiconductor  manufacturing excellence. Rather, it is the  

opposite. Intel leverages its profits to fund an  advanced semiconductor-manufacturing division.

Intel has certainly before integrated a few  impressive technologies into its process nodes,  

but they didn't originate those technologies.

Like the High-K Metal Gate  was first explored by IBM,  

Sematech and other companies in  the industry. Though I am aware  

that Intel pioneered the Gate-Last method  that properly integrated High-K metal gate.

Two other iconic semiconductor technologies  were copper interconnects and EUV. Again,  

IBM invented the damascene techniques  that made copper interconnects possible.

And as we know, a Japanese  researcher first proposed  

EUV. And the core technology pioneered by  the US Government's Department of Energy.  

Intel did help fund and move that  effort towards commercialization.

So I just want to say that I give  Intel full credit for executing  

on these technology integrations. But  research is a terrible business - that  

is why it's done by nonprofits. And it's far  easier to execute when money is no object.

So where does Intel massive profits in the 1980s  and 1990s come from if not manufacturing? They  

came from Intel's aggressive sales,  slick marketing, and above all,  

IP lock-in. Wilf Corrigan, founder and former  CEO of Fairchild, recalls in an oral history:

> One company controls the  whole industry like DeBeers  

controls the worldwide diamond business ...

> this is an IP control, it's not because  necessarily Intel can manufacture better  

than anybody else though they're a  very good manufacturing company today.

> It's really the control of the IP,  the intellectual property, the patents,  

the legal position, the copyright control  and so on, which nobody's been able to break

Wilf has it right. Throughout American history,  

control over special IPs has consistently  built monopolies. At first through patents,  

like in the case of AT&T and Marconi, and then  later industry standards like the Wintel PC.

Intel funds its fabs through its  grip on IP. That special IP being  

its x86 instruction set architecture -  that set of special instructions that  

a software can tell a CPU's hardware to  perform on data - first implemented with  

microcode. When x86 lost its place in  the computing world, thus so did Intel.

## A Second Source

So what can we do? If we can't bring  back the x86 monopoly for Intel,  

how about we gift them a franchise  that is just as profitable?

At an AI summit in July 2025, Trump told the  audience that he considered breaking up Nvidia,  

to give them a little competition. But aides told  him it was not so easy or something like that.

So no breaking up Nvidia. But if Jensen Huang so  respects and wants to work with Trump, then how  

about sharing some of Nvidia's consumer surplus  with Intel for the sake of national security?

I harken back to the 1960s and 1970s, when  the US military and large companies demanded  

that their semiconductor suppliers  strike second-sourcing deals with  

"second-sourcers" - often another large  semiconductor company. So the original  

owner licenses over their technology to the  second-sourcer, done via a transfer process.

One of the most famous such deals  took place in 1976 and then 1981.  

As a condition of putting an Intel CPU  inside their PC, IBM demanded that Intel  

license and share their chip technology to  a second source. That turned out to be AMD,  

and helped make AMD the critical second  source for x86 chips in the 1980s.

Nvidia's edge over other AI chip companies  like AMD is its proprietary software framework,  

CUDA. CUDA exhibits exceptional robustness,  maturity, and backwards compatibility as  

compared to competing frameworks. And it  is the basis of Nvidia's 80% gross margins.

So the idea is Nvidia issues Intel a license  for the IPs and patents in the Nvidia CUDA  

and AI ecosystem, and does a technology  transfer. With the goal being that Intel  

can produce "Nvidia-compatible" chips in their  American fabs to be sold as a second source.

That Intel-AMD deal was structured as a mutually  beneficial exchange of products based on those  

products' complexity and design effort as  defined by a formula. So if they'd like,  

they can have Intel issue an  x86 license to Nvidia in return.

Nvidia sells a CPU of their own called  Grace, and they apparently bundle it  

together with their GPUs. If you buy  the GPUs, you have to buy the CPUs too.  

Grace isn't an x86 chip and I highly  doubt that Nvidia will want to convert  

it over to x86. But perhaps they  do it to expand the x86 ecosystem.

I don't know how much of the AI accelerator  market that Intel can expect to get with its  

Nvidia-compatibles. Nvidia made $150  billion in revenue and $89 billion  

in profits over the past twelve months.  Let's assume that's the current market.

Go back to the IBM PC, the PC-compatibles took  roughly about 30% in their first year and 40%  

their second as guys like Compaq, Columbia  Data and others entered. So let's say the  

compatibles get 20-30% of the AI market to be  conservative. That's $30-45 billion of product  

revenue potentially each year. Enough to fund  Intel's factories so long the AI boom continues.

Nvidia's gross margins and revenues will  of course fall. But margins were pushing  

80%. So I think they remain one of the  world's largest companies by market cap.  

So I reckon they'll be fine. Maybe they  can get a tariff credit or something.

## Conclusion

I want to emphasize that this is just  a crazy idea. There will of course many  

legal and technical objections. So most likely,  

the government just does the obvious  thing. The stuff we talked about earlier.

But there are a few reasons why I like this idea  

so much. For one thing, it exercises both the  company's product and manufacturing sides.

With all the talk about national security,  Intel's product half has sorta faded out  

of the conversation. But the products  still matter. It was one of the first  

things mentioned by the Board at Gelsinger's  retirement announcement: Products lead the way.

An Nvidia second-sourcing agreement allows  Intel Product to develop an in-demand,  

leading-edge product, but do  it in the way that they are  

familiar with - by working closely  with their folks at Manufacturing.

And Intel Manufacturing can get comfortable  working with its internal customer on a  

technically demanding, high volume product to  clear the pipes, and get its kitchen ready for  

opening to outside customers. No dealing  with 7 very mean tech giant customers.

Another thing I like to add is that  second-sourcers can help improve the  

overall ecosystem to everyone's benefit. If AI  is going to create an intelligence explosion,  

then it would be nice to have  more than just Nvidia in there.

This is more than by just cutting  prices. They can also contribute,  

technically. AMD, the once second-source  copycat, has contributed several things  

to the x86 framework that were  licensed back to Intel like AMD64.

And finally, the idea has this sort of "teach  a man to fish" thing. Intel's problems are  

so big that just repeatedly giving them  fishes - i.e. cash infusions - will not  

solve the core issue at hand. We help them  by teaching them a productive business. That  

has been the problem all this time.  So let's make Nvidia share theirs.