Data Center (1)
The CEO’s office at JH Research.
Knock knock.
A soft rap sounded against the door of the executive suite.
“Come in.”
Chief of Staff Jang stepped through, something cradled carefully in his hands—a neatly wrapped paper box. My eyes went wide the moment I saw it.
“Wait… is that—?”
“Yes. The JH-100 sample has arrived from Ohsung.”
“Oh! Finally!”
This was the first finished unit since we had commissioned Ohsung for the packaging work.
JH-100
I tore into the box. The JH-100 was the LPU module that carried JH Research’s name. Its overall form was a solid rectangle, the surface finished in a deep metallic sheen so dark it was nearly pure black. A cooling fan sat flush along one side of the top face, engineered to yank heat out of the interior as fast as it was generated.
“It opens if I press here, right?”
“That’s correct.”
Click.
I opened the metal casing. Inside, a green board was packed with dense microscopic circuitry; at the center, the LPU die and HBM memory stack had been packaged with exquisite precision. The HBM rose in stacked cubic layers, pressed close against the LPU die—linked so tightly they looked like a single living organism built for high-speed data transfer. Gold terminals gleamed at the power and connector sections, while the case edges were finished with ventilation channels that guided airflow. The whole design radiated the raw force of industrial hardware fused with the precision of high-end electronics.
Was there anything more that needed saying?
“It’s gorgeous.”
Jang’s expression was quietly proud.
“The first batch of five hundred units is scheduled to arrive in a week, and shipments will keep coming steadily after that.”
The production loop was complete.
LPU design and mask fabrication in the Nano World → mass production of the LPU at TSNC → DRAM purchased and turned into ten-layer packages at Baekryong → finished as HBM back in the Nano World → packaging at Ohsung.
“Chief Jang. Let’s hold a meeting. Gather the team leads.”
“Understood.”
The company had grown large enough that we finally looked the part. JH Research alone now employed over a hundred people. We had gone from a firm that auctioned a single product to one that exchanged finished goods with TSNC and Ohsung; we needed people who matched that scale.
When I reached the conference room, Dr. Han Woo-seok was already there.
“Dr. Han. I heard your dissertation went through a few days ago?”
Color rose in his face.
“Yes. Thanks to you. I’m grateful.”
“Come on—that was all your hard work.”
“Not at all. The paper passed quickly because it was JH. Anywhere else it would have taken forever.”
Dr. Han Woo-seok had once been the PhD candidate I hired as a private tutor. Strictly speaking he had already finished every required course; only the dissertation stood between him and the degree. Everyone casually called him “Doctor,” but there was a real gap between candidacy and the actual diploma. Plenty of people finished the coursework yet never managed the thesis. A doctoral dissertation had to present knowledge that did not yet exist in the world—and creating something that did not exist was never easy.
“As it happens, I brought a bound copy of the dissertation.”
He handed me a thick volume.
Design of the Blue Tower HBM Architecture for Next-Generation Ultra-High-Speed Computing and Optimization of the Integration Process
“You allowed me to document everything, so writing it was straightforward.”
“Don’t mention it. I planned to publish the basic specs anyway.”
We had to sell the product; how much of a problem could basic disclosure be? Besides, the Blue Tower was more than just its design. There was something even more important.
Mithril.
No other material approached mithril’s electrical conductivity. Without it, we never would have drawn this much attention.
“Because you let me use it for the thesis, the degree came easily.”
“Please. That was your effort. Anyway, now that the paper’s done, you’re all-in on JH, right?”
Dr. Han bowed his head.
“I will bury my bones at JH.”
“Haha. Planning to go that far? In any case, Doctor—you received the performance sheet for the JH-100 from Ohsung?”
“Yes. And we finished our own performance tests as well. Allow me to present the JH-100’s specs.”
He brought up a projector and threw the specifications onto the screen.
Model: JH-100
Architecture: Next-generation parallel compute core based on LPU
Memory: 10-layer HBM stack (1 TB+, bandwidth 30 TB/s+)
Packaging: 2.5D interposer integrated structure, LPU + HBM monolithic
Compute Performance: Multi-thousand TFLOPS class at FP8 (500–1,000× existing)
Power Draw: Roughly half of comparable units (~250 W for equivalent performance)
Cooling: Dual hybrid fans, high-performance heat pipes & heatsink
Process Node: Sub-2 nm ultra-fine process
Expected Lifespan: Stable operation 7+ years
Form Factor: Standard LPU module, all-black metal housing
The staff stared at the monstrous specs and murmured in awe. When we had first revealed the LPU alone, inquiries had poured in from every direction—even the Vice President of the United States. Now their faces asked the same silent question: what would happen once this went public? Excitement lit every expression as they looked at the world’s highest-grade mass-producible LPU module.
“Whoa. That bandwidth is no joke.”
“Power consumption is only half of current top-tier units.”
“The heat and power load are revolutionary.”
“Performance is a full decade ahead of this generation.”
“Heh. In a word—it’s a monster.”
“How much are we going to rake in this time?”
Team Lead Ran Zhao-ming looked at me. We had hired her as our China specialist, yet she had proven sharp in every direction—excellent work habits, deep semiconductor knowledge, and a keen commercial sense. A plump, warm-hearted woman with the presence of a field commander.
“What will you do, Boss?”
I turned the question back on her.
“If we sell these, how much should we charge?”
What was the market price of the world’s finest LPU module?
“At this moment there is no product that can keep up with it.”
“Right.”
“So in practice we have a monopoly.”
“Exactly.”
“Then of course we name our price! Market value is whatever we say it is.”
After a brief silence the staff began cautiously floating numbers and debating.
“Fifty million won per unit?”
“Why that?”
“If you go by performance, it feels about right. Stack enough current GPU modules to hit this level and you’re looking at roughly fifty million.”
Ran shook her head.
“Nope. That’s the same as saying ten units of performance-1 equal performance-10. You can’t calculate it that way.”
“Then what?”
“…Three hundred million?”
“Hic!”
“Something smaller than a book costs three hundred million?”
“Is size the issue? This is monopoly and first-mover advantage. The only product that exists is ours. Real monopoly. And the volume we can supply is limited. Whoever receives the world’s best hardware seizes the first-mover edge.”
Dr. Han nodded in agreement.
“She’s right. Especially in fields that must lead the early market—AI, autonomous driving—our LPU module will have an extremely powerful effect.”
“Exactly. The performance will be so overwhelming that the AI built on it will be overwhelming too.”
Ran’s small eyes gleamed.
“There are plenty of AI services in the world, but consumers actually use only a handful. Every one of those was first to dominate its market.”
“So obtaining our LPU module means you get to dominate the market?”
She nodded.
“Of course.”
She grinned and spread all ten fingers.
“In that spirit—let’s go with one billion won per unit!”
The other staff members’ eyes went round.
“Wow, Team Lead Ran! Aren’t you swinging a little hard? If the price is that high, buyers will just walk away.”
But Ran never lost her smile.
“It’s fine. Only number one needs to buy.”
Han sided with her.
“I can’t guarantee the final price, but places that need ultra-high-performance data processors are everywhere—AI, autonomous vehicles, robotics, cloud, virtual reality, drug discovery…”
“Exactly. If you want to sit in first place in any field that processes massive data, you absolutely need our product. In my view we shouldn’t even be talking corporate deals—we should be talking nation-level transactions.”
The phrase “nation-level” made me ask for details.
“Oh? What do you mean by nation-level?”
“Normally the company in first place is also the one that understands value best. But the party willing to pay the highest price isn’t always number one.”
“Interesting.”
“If number two sees a path to becoming number one, the wallet opens.”
“That makes sense.”
“Right now the United States is number one in the data market. If America buys our LPUs, it stays number one. But our product is simply too good. If a second- or third-place nation acquires it, the odds of overtaking first place shoot up. So if we talk to number two or three, they’ll pay even more.”
“Are you talking about China?”
“Not necessarily. China would be fine, but Korea works too.”
“Korea?”
“There isn’t another country as serious about AI as the Republic of Korea. If we build our own data center with LPUs, the government will cheer. Sweet-talk them a little and subsidies will appear.”
“Oh-ho. Subsidies.”
“But if it’s a domestic data center, we still have to decide whether to outsource the build or do it ourselves. That becomes a question of how far we want to expand the business.”
Listening to the staff debate, a warm satisfaction settled over me. Jang Dae-shik, my ever-reliable right hand. Dr. Han Woo-seok of Hankook University. Ran Zhao-ming, rich in both theory and practical sense. The company wasn’t just accumulating money—it was accumulating outstanding people.
“I’ve heard the discussion.”
They waited for my decision. I lifted the LPU module from the table in front of my seat. Clean, elegant lines. Still magnificent no matter how many times I looked at it.
“So whoever holds this LPU module becomes the leader of data. That’s the claim?”
“Correct.”
“Let’s assume we put the JH-100 on the market. Right now Amazon is number one in cloud, yes?”
“Yes.”
“If Amazon buys our product, it fortifies its cloud lead and advances its logistics business as well?”
Heads nodded around the table.
“Highly likely.”
“And Joke from Facebook keeps contacting me. As you know, he’s completely obsessed with virtual reality.”
“True. There are even rumors he’s renaming the company.”
“I hear he’s changing the name from Facebook to Metaverse.”
“Metaverse?”
“It means he’s going all-in on VR. Anyway, if Joke buys our LPU modules he’ll pour them into virtual reality. Or, since it’s an SNS company, into SNS.”
More nods.
“If OpenAI buys them, obviously they’ll go into AI. If Merks buys them, they’ll go into autonomous driving or robotics. Right?”
“Correct. Big Tech is big—they can’t suddenly pivot their existing businesses.”
An LPU module the size of two palms. The price was still undecided, but the staffer who had suggested fifty million won had fallen silent, so the unit was clearly in the hundreds of millions. Buying them by the tens of thousands would require trillions of won. The only parties who could write that check were the American Big Tech firms I had just listed—or, as Ran had said, entire nations.
I rolled my wheelchair back a little.
Creak.
The wheels pressed against the floor.
“As you can all see, I am disabled.”
No one in the room needed the reminder. Their faces asked why I was bringing it up now.
“And my mother is still living with the aftereffects of a cerebral hemorrhage.”
Mother and son, side by side, both disabled. I looked down at the LPU module.
“In simple terms, if ten thousand—or a hundred thousand—of these modules gather in one place, they become the One Ring.”
The comparison might have sounded absurd, yet the staff nodded as if it were the most natural thing in the world.
“We have a data-processing module that rivals the One Ring. If we sell it for money, it becomes the tool that realizes someone else’s dream.”
Somehow the conversation had turned into a discussion of the company’s direction. The staff listened in silence.
“But among the people who would buy this, is there anyone who would build a product for me? Or put real effort into something for my mother?”
There were medical companies that cared about medicine, but none that combined Big Tech-scale data expertise with a genuine commitment to healthcare.
“This One Ring… we’ll use it ourselves.”
Eyes widened around the table.
“We build a data center, then we build an AI. Plenty of places are already open-sourcing their models, so catching up on the software side shouldn’t be that hard. And right now our hardware is the best in the world. Let’s take open-source AI and grow it on top of that.”
I glanced sidelong at Dr. Han. He was far more of an expert than I was.
“Excellent. Honestly, I was debating whether to suggest it myself. But selling would bring in so much money I couldn’t bring myself to say it. Let’s do it ourselves!”
Energy sparked in every pair of eyes.
“Yes! Let’s take it for ourselves!”
“Right—don’t sell a single one. We do everything.”
A faint guilt pricked at me.
“If we sell, trillions of won will land in our accounts within months. You would all see performance bonuses. But if we don’t sell—if we build a data center and create an AI instead? You know even OpenAI, the world’s leading AI company, is still operating at a loss, right?”
Sell now and bonuses arrive. Don’t sell and no one knows when bonuses will come. The question hung unspoken: Are you sure you want me to decide this on my own?
“Boss, bonuses are nice, but we’re not exactly a poor company.”
“Exactly. I already feel like my salary is more than I deserve.”
“Salary matters, but so does vision.”
“You’re basically becoming Korea’s Aaron Merks. Hell yes!”
Their reaction was nothing like the worry I had felt.
“Then shall we run with it?”
“Yes! Let’s run!”
“Yes! Our company is jumping into the global AI war! Man, this is exhilarating!”
I was doing this for myself, and they were calling it vision.
“All right. Look into two things for me. First, a site where we can build a data center. Second, AI companies worth acquiring.”
Creak.
I gripped the handrims of my wheelchair and gave them a shake.
“You said we should run, right? But as you can see, I don’t have legs. So run for me.”
“Yes, sir! Hop on! We’re running!”