On one hand, everyone is talking about an AI bubble. That it has gotten too big, that it will burst, that the party is over.
On the other hand? Intel, which some people had literally written off, is discussing a deal that could bring memory production to the United States. And Apple is preparing to re-enter a market it abandoned 15 years ago, with Nvidia by its side.
THE INTEL THAT EVERYONE HAD WRITTEN OFF
Remember how many times we read that Intel was finished? That it had missed the train, that the Ohio factory was just a construction site that would never produce chips.
Then Reuters came out with a very different story. SK Hynix, the Korean memory giant, is reportedly in talks with Intel about a deal that would have it produce memory chips on U.S. soil. For the first time in its history.
And there are two scenarios on the table.
In the first, SK Hynix would lease part of the factory Intel has been planning for years in Ohio. In the second, they would create a joint venture involving Intel, SK Hynix, and major cloud companies that want to lock in memory supplies before someone else gets them.
And the market? It responded immediately. Intel’s stock jumped 5% in premarket trading and closed up 9%.
But there is a big “but.” Actually, there are three.
First, SK Hynix itself walked the story back. In an official statement on Wednesday, it clarified that nothing had been finalized, either regarding cooperation with the companies mentioned or regarding memory production in the United States. It said it was exploring various options to strengthen its competitiveness. Nothing more.
Second, there is Seoul. HBM and DRAM are considered sensitive technologies in South Korea. So a deal like this could face resistance from the Korean government itself. Sources describe the discussions as “exploratory,” with no decision having been made.
And third, there is Intel. It held the groundbreaking ceremony for the Ohio project in 2022, with then-President Biden in attendance. Since then, however, the company has slowed the project down and last year pushed back the start of production to 2030.
So why is there so much pressure?
Because there is a global shortage of memory. U.S. Commerce Secretary Howard Lutnick has already publicly pressured SK Hynix and Samsung to increase memory production in America. And Micron announced that it would spend $250 billion to expand its capacity domestically.
SK Hynix is already building a $4 billion packaging facility in Indiana. But it has not announced a memory fab in the United States. The wafers are manufactured in South Korea and shipped to Indiana for assembly and testing.
APPLE REMEMBERED SERVERS
Now let’s move to the second story, which is equally unexpected.
According to The Information, Apple is considering building enterprise servers. With its own chips, from the M Ultra family. And with networking equipment from Nvidia.
And who would it sell them to? AI developers, businesses, and governments. In other words, the company that sells us iPhones wants a piece of the compute market.
The details in the report are specific. There would be two versions. One with two of the upcoming M8 Ultra chips, and another, more powerful version, with four. The connection between them would use Nvidia’s NVLink Fusion technology, along with switches, chiplets, and software.
But there is an important caveat here. The product would not launch before 2029. It is not certain that Nvidia would ultimately be involved, and the entire plan could still be canceled.
“And didn’t Apple try this before?” you might be wondering.
It did. From 2002 to 2011, Apple sold the Xserve, using Intel and IBM chips. And it didn’t go well. It failed to find an audience.
The only difference is that things are completely different now.
And to understand why, let’s look at Apple’s fiscal third-quarter 2026 results.
Mac sales increased 29% year over year, reaching $10.3 billion. And the impressive part isn’t just the percentage.
It was the largest percentage increase of any Apple product category. Higher than the iPhone. Higher than Services as well.
“And who is buying that many Macs?” you might be wondering.
AI labs and developers. They have accumulated tens of thousands of Mac minis and Mac Studios to run their workloads.
So what does that mean?
It means Apple is already selling into this market without having specifically designed a product for it.
And now it is saying, “Wait a minute, let’s build a product specifically for these customers.”
In fact, the new CEO, John Ternus, has supported the plan for an entire year. He sees it as an opportunity to bring Apple’s custom silicon into the AI compute market.
Put the two stories together.
Two different stories, one common element.
Nobody is planning for the next quarter.
They are planning for 2029 and 2030.
