Why Is SK Hynix Considering Making Memory Chips in America?
The world's most advanced AI chips need something most people rarely see.
Memory.
More specifically, they need enormous amounts of fast memory to keep expensive AI processors working.
South Korea's SK Hynix has become one of the most important companies in that system.
So here is a strange question.
Why would a Korean memory company consider making more chips in the United States, where manufacturing can cost more?
That question became more interesting on September 16, 2026.
Reuters reported that SK Hynix is in exploratory talks with Intel about potentially manufacturing memory chips in the United States.
One possibility could involve using part of Intel's massive semiconductor site in Ohio.
Another could involve SK Hynix, Intel and major cloud companies looking for more secure chip supplies.
Nothing has been decided.
But the talks reveal something bigger about the AI boom.
The cheapest place to make a chip is not always the most valuable place to make it.
Start With the Customer
Imagine you run a small bakery.
You need flour.
The flour mill can be hundreds of miles away.
A truck arrives every week.
No problem.
Now imagine you run the largest bakery in the world.
Your factories consume enormous amounts of flour every day.
If deliveries stop, production stops.
Suddenly, you care about more than the price of flour.
You care about:
Where it comes from.
How much is available.
How quickly you can get more.
And what happens if the supply chain breaks.
Something similar is happening with AI memory.
Memory used to look like another component inside a computer.
Now certain types of memory have become critical infrastructure.
AI Changed the Importance of Memory
A powerful AI processor can perform enormous numbers of calculations.
But it needs data.
And that data has to arrive extremely quickly.
That is why HBM — High Bandwidth Memory — has become so important.
Think of an AI processor as the world's fastest chef.
Making the chef faster is useful.
But not if the ingredients arrive through a tiny door.
HBM helps widen that door.
It moves enormous amounts of data quickly enough to keep expensive AI processors working.
That means companies building huge AI systems do not only need GPUs.
They also need enough fast memory to keep those GPUs busy.
And when everyone wants the same critical component at the same time, the location and security of its supply begin to matter.
America Has Many of the Biggest Customers
The United States is home to many of the companies spending enormous amounts of money on AI infrastructure.
Amazon.
Microsoft.
Google.
Meta.
And Nvidia, whose processors sit at the center of much of the AI boom.
Those companies are building increasingly large computing systems.
Those systems need processors.
The processors need memory.
The servers need networking.
And the entire system needs enormous amounts of electricity.
SK Hynix already has major manufacturing operations in South Korea.
But it is also building something new in America.
SK Hynix Is Already Spending More Than $4 Billion in Indiana
In August 2026, SK Hynix broke ground on a new semiconductor facility in West Lafayette, Indiana.
The planned investment is more than:
$4 billion.
SK Hynix says the facility will become a U.S. production base for next-generation HBM.
The company expects mass production to begin in the second half of:
2029.
It also expects the project and surrounding semiconductor ecosystem to create about:
7,000 direct and indirect jobs.
The company is working with more than 100 partners and Purdue University as part of the project.
So the idea of producing advanced AI memory in America is no longer hypothetical.
Construction has already started.
The new question is whether SK Hynix could go even further.
Now Intel Enters the Picture
On September 16, Reuters reported that SK Hynix is in exploratory talks with Intel about potentially manufacturing memory chips in the United States.
One possibility under discussion could involve SK Hynix leasing part of Intel's semiconductor complex in Ohio.
Another could involve a joint venture with:
SK Hynix
Intel
and potentially
major cloud companies.
The discussions are still exploratory.
No final agreement has been announced.
And an important detail remains unknown:
What memory would actually be produced there?
SK Hynix makes several types of memory, including DRAM, NAND and HBM.
So it would be wrong to say that Intel's Ohio site is definitely going to make HBM.
It may not.
The deal itself may never happen.
But the fact that these structures are being discussed tells us something about how the AI supply chain is changing.
Intel Has Something SK Hynix Could Use
Intel has been building an enormous semiconductor manufacturing complex in Ohio.
The company says its investment in the site exceeds:
$28 billion.
Two leading-edge semiconductor factories are planned there.
But the project has taken longer than originally expected.
Intel's current schedule calls for the first manufacturing module to be completed in 2030, with operations beginning between 2030 and 2031.
The second is expected to begin operating in 2032.
That creates an interesting situation.
Intel has:
land
factories under construction
U.S. semiconductor infrastructure
and
decades of manufacturing experience.
SK Hynix has something different:
advanced memory technology
strong HBM capabilities
and
AI-driven demand.
Put those pieces together and an unusual possibility appears.
The company that owns the factory does not necessarily have to be the company whose memory eventually comes out of it.
The Customer Could Become Part of the Factory Too
This may be the most interesting part.
Reuters reported that one possible structure could involve major cloud companies.
Think about how unusual that sounds.
Normally:
A cloud company buys chips.
A semiconductor company makes chips.
A factory produces them.
Simple.
But AI is making those relationships more complicated.
Imagine a cloud company expects to spend tens of billions of dollars building AI infrastructure.
A shortage of memory can delay servers.
Delayed servers can delay data centers.
And delayed data centers can leave billions of dollars of infrastructure waiting.
At that scale, the customer may care about more than getting the lowest possible chip price.
It may care about:
getting the chips at all.
The relationship can start changing from:
supplier → customer
into something closer to:
supplier + manufacturer + customer + capital.
This pattern is appearing across the AI economy.
AI Customers Are Starting to Secure Their Own Supply Chains
Large technology companies increasingly sign long-term agreements for things they cannot afford to run out of.
Electricity.
Computing capacity.
Data-center space.
Networking capacity.
And increasingly, critical semiconductor supply.
The buyer does not always simply wait for a product to appear.
Sometimes the buyer helps create enough certainty for the infrastructure behind that product to be built.
That changes the economics.
If memory becomes a bottleneck, a giant cloud company has a reason to care about the factory producing it.
Not because the cloud company suddenly wants to become a memory manufacturer.
Because:
an empty server rack cannot run AI.
But Isn't Making Chips in America More Expensive?
Potentially, yes.
Reuters reported that producing memory in the United States could involve higher costs than manufacturing it in South Korea.
That makes the story more interesting.
If production cost were the only consideration, the decision would be much easier.
Build where manufacturing is cheapest.
But companies now have to consider other things.
Supply security.
Government incentives.
Tariffs and trade policy.
Access to major customers.
Technology restrictions.
National security.
And perhaps most importantly:
the cost of not having enough chips.
Imagine making a component in America costs more.
Normally, that sounds inefficient.
But imagine a shortage of that component delays a multibillion-dollar AI data center.
Now the calculation changes.
The cheapest chip can become very expensive if it arrives too late.
Governments Are Now Part of the Supply Chain
Semiconductors have become too strategically important for governments to ignore.
When SK Hynix announced its Indiana advanced-packaging project in 2024, it reached preliminary terms with the U.S. Department of Commerce for up to:
$450 million in proposed direct funding
and access to:
$500 million in proposed loans
under the CHIPS and Science Act.
Why would a government help finance a memory facility?
Because advanced semiconductors are no longer treated as ordinary commercial products.
They are increasingly treated as strategic infrastructure.
The United States wants more semiconductor production and advanced packaging inside the country.
Shorter supply chains.
More domestic capacity.
And less dependence on overseas production for critical technologies.
South Korea Has an Interest Too
But SK Hynix is a South Korean company.
And advanced memory technology is strategically important to South Korea as well.
Reuters reported that a potential U.S. manufacturing arrangement involving sensitive memory technology could require scrutiny in South Korea.
That means a memory factory can involve much more than two companies negotiating a commercial deal.
It can involve:
SK Hynix
Intel
American cloud companies
the U.S. government
and
the South Korean government.
A tiny piece of silicon can become part of industrial policy.
This Is Bigger Than SK Hynix
For decades, the semiconductor industry became more global by dividing the work.
One company designed a chip.
Another manufactured it.
Another made the equipment.
Another packaged it.
Factories were built where companies believed they could operate efficiently.
That system is not disappearing.
But AI is adding another question.
Not simply:
Where can we make this cheapest?
But:
Where does critical capacity need to exist?
Those are very different questions.
When a component is easy to replace, price matters enormously.
When a component becomes a bottleneck for infrastructure worth billions of dollars, reliability starts to have a price too.
So does distance.
So does political risk.
So does guaranteed capacity.
AI Is Changing the Map of Semiconductor Manufacturing
The most visible part of the AI boom is still the processor.
GPUs get the headlines.
But underneath those processors sits an enormous physical supply chain.
GPU
↓
HBM
↓
Advanced packaging
↓
Semiconductor factories
↓
Manufacturing equipment
↓
Power
↓
Data centers
Each layer can become a bottleneck.
And when a bottleneck becomes important enough, companies start spending money to expand it, move it or secure more control over it.
That is what makes SK Hynix's U.S. expansion worth watching.
The company has already broken ground on a more than $4 billion HBM production base in Indiana.
Now it is exploring additional U.S. memory manufacturing possibilities with Intel.
No Ohio deal has been announced.
We do not yet know what kind of memory would be produced there if a deal happens.
But the reason these discussions exist is already visible.
AI has made memory too important to think about only in terms of price.
The question is no longer simply:
Where can we make memory cheapest?
It is increasingly:
Where do we need memory to be available?
In the AI era, those can be two very different places.
BEYOND THE OBVIOUS.
Related on OVIQQ
Before asking where AI memory should be made, there is a simpler question:
Why does AI need so much memory in the first place?
Why Does AI Need So Much Memory?
Sources
Reuters — SK Hynix in Talks With Intel About Deal to Make Memory Chips in U.S. for the First Time
SK hynix — Groundbreaking Ceremony for HBM Production Base in Indiana
https://news.skhynix.com/en/groundbreaking-ceremony-in-indiana/
Intel — Ohio One Construction Timeline Update
https://newsroom.intel.com/corporate/ohio-one-construction-timeline-update
Intel — Intel Invests in Ohio
https://newsroom.intel.com/press-kit/intel-invests-ohio
SK hynix — Preliminary CHIPS Act Terms for Indiana Advanced Packaging Facility