The scholars had been right. The pace of development in the information age really was fast.
But who would have thought it would become this fast...?
Roughly fifteen years ago, when smartphones became commercialized, humanity seemed to have reached the pinnacle of civilization.
But in less than ten years, AI appeared, and the AI born that way swept the Nobel Prizes in a mere five years.
The scholars had been wrong. They had clearly said low-skilled laborers would be replaced first and that creative fields like the arts would become dominant, but the reality was... the smart ones were replaced first.
Someone once said, “Studying is the easiest thing,” and apparently that wasn’t just a promotional line from a self-help book. It really had been easy.
The reason AI seized the world stage as soon as it appeared could be said to be the GPUs that advanced day by day. As computing speed improved, computers’ ability to acquire and distinguish information became vastly better. Behind AI’s dazzling development were Envidi GPUs that improved every year, and the chips they announced—A100, H100, H200, and so on—were upgraded almost threefold every year.
The A100, considered a cutting-edge chip just five years ago, was now on the level of a farm tractor... In other words, even the H200 that people couldn’t buy because there weren’t enough of them could soon become a farm tractor too.
[Breaking News — Gogul Announces Gemini 3.0!]
[Is the Envidi Era Coming to an End? Gogul Reveals It Was Developed with Its Own TPU, Without GPUs!]
And yet that didn’t seem like such a distant story anymore.
Early morning, on the road up to Icheon.
News that had crossed over from the other side of the globe struck our KOSPI.
The new version of Gemini announced by Gogul displayed innovative technological prowess in its “image generation and editing” functions... but what was more astonishing than anything else was that this technology had been created not with Envidi’s H200 chip, but with Gogul’s self-developed TPU chip.
[Breaking News — KOSPI Opens with a 2.5% Surge!]
[An Opportunity for Hoam and Hynix Too?!]
It was truly astonishing.
Wasn’t the next thing after the H200 naturally supposed to be Envidi’s H300? Was it the TPU era now?
Global rating agencies focused especially on the economic viability of TPUs. Unlike GPUs, which had no choice but to consume enormous amounts of electrical energy, TPUs boasted considerable energy efficiency. Just as humanity had given up internal combustion vehicles and moved toward the electric vehicle era, if the energy efficiency gap between the two chips continued to widen, no one could guarantee which chip would become the main computing chip.
“...”
The entire time I watched the news, my fists were clenched tight.
Thanks to frequent late nights at work and repeated drinking sessions with politicians lately, my memory wasn’t intact. I definitely remembered that AI-related businesses had gone through a singularity, but since that singularity hadn’t happened just once or twice, I couldn’t quite remember exactly what had become the mainstream.
Still, one thing was certain: soon, America’s hegemony over the AI market would grow so strong that even second place would become meaningless...?
“...”
By the time I arrived at Icheon Station, I was already utterly exhausted.
Could Nave really protect our search market share against monsters like these?
*
“Hello, Chairman. Have you been well?”
“Welcome, Team Leader. It’s been a long time.”
I wiped the gloomy look off my face and entered Hynix’s headquarters in Icheon.
Chairman Kim Jeongjin, whom I hadn’t seen in a while, greeted me with a bright smile.
The chaebol heads I had met recently had all looked like they were at a funeral, but this side alone had an extremely bright expression. That was because the sales performance of our prime contractors in America had improved day by day, and as their subcontractor, our semiconductor industry had received a considerable share of the crumbs.
“First, I’d like to offer my condolences. Regarding this Sovereign Project.”
“Condolences? I’m the one who should be apologizing. We made it all the way to second place by making use of Hynix’s strengths, but Nave’s competitiveness was outstanding. I’m sorry we couldn’t live up to the great expectations placed on us.”
Hynix’s advance in this Sovereign Project had been substantial.
They had defeated one formidable company after another and reached the final round. Although they had remained in second place and had to swallow defeat, it seemed to have been a major change in atmosphere for Chairman Kim Jeongjin.
After Sovereign ended, he announced “Technology Leap 2.0” and declared that Hynix’s AI development would not stop, regardless of the result.
“But we learned a great deal. Before, it felt vague, but as expected, once we smashed our heads against it directly, I started to get a feel for it.”
“I heard you decided to increase R&D further.”
“Yes. This is only the beginning. We plan to use the current semiconductor supercycle as seed money and reinvest a significant portion of the profits from it into AI development.”
I smiled and nodded.
“Yes, you can do it. You will make it happen.”
Once the pleasantries ended, he spoke to me as if he had been waiting.
“Team Leader, you came today because of GPU development, didn’t you?”
“Yes. I’m curious about the current progress.”
One year ago, when the Sovereign Project began.
The pension fund had invested ten trillion won into Hynix.
At the time, I had put forward one condition when making that ten-trillion-won investment: use that money not on your current mainstay, memory, but on proprietary GPU technology.
“I didn’t come expecting visible results in just one year. I just want to hear about the current situation.”
“I hate to say this, but so far, there haven’t been any meaningful results. We also tested the computing capabilities of several models we developed ourselves, but... in terms of performance, functionality, and price, it isn’t easy to narrow the gap with the frontrunners.”
“I see. Don’t worry too much about it. I didn’t think you would succeed from the start.”
“But, Team Leader... I think I need to ask you for a favor.”
“A favor?”
“...We need more investment.”
He continued with a troubled expression.
“As you know, the GPU landscape is swaying once again.”
“TPUs?”
“Yes. The industry had been uncertain about them for some time, but Gogul finally succeeded in developing the technology. Of course, internally, our strategy office does not assess TPU technology as having caught up to Envidi’s GPUs yet. It’s true that they’re at least one generation behind. However...”
“However?”
“The energy efficiency possessed by TPUs has major implications. Electric vehicles have no efficiency advantage at all compared to internal combustion vehicles, and yet countries are pouring subsidies into them solely because of environmental issues, aren’t they? If the same phenomenon occurs in the graphics market, I believe the GPU market landscape will undergo a very significant change going forward.”
Tensor Processing Unit, TPU.
Gogul had begun offering the service in 2018, but because its versatility was not high, it had not received much response from the market.
To explain this difference in versatility, I would have to start with the von Neumann bottleneck... but it was better to simply understand it as a chip specialized solely for AI calculations, unable to handle graphics, encoding, simulations, and the like. A mathematician who was only good at mental arithmetic, so to speak.
But the efficiency of that mathematician who was only good at mental arithmetic was tremendous.
A TPU’s energy efficiency per watt was four times that of a GPU. Data centers had long since become power-guzzling behemoths, and TPUs compensated for that weakness.
What was even more frightening was the speed at which TPUs were developing.
When Gogul first developed a TPU chip in 2013, it was not a chip with such excellent efficiency. At the time, it actually consumed more electricity than a GPU.
But Gogul had invested for ten years in that computing chip that seemed to have no prospects, and the current seventh-generation TPU born from that had developed into technology capable of saving roughly thirty times more power than the initial model.
“We need money... money.”
And so Chairman Kim’s shameless appeal for money sounded to me like the pitiful crying of a baby searching for its mother’s milk.
We needed money... If we were going to smash our heads against this and take it on, we needed money first. We needed the chance to use that money to meet the price of top talent and experience countless failures.
“How much do you need?”
“Gogul has not disclosed it directly, but based on our own calculations, we estimate that raising TPU performance to the seventh generation must have cost at least several tens of trillions of won.”
My head went dizzy.
Thanks to the recent boom in global stock markets, the pension fund’s earnings in the last quarter alone had reached thirty trillion won.
And yet we had to invest all that money into our semiconductor industry again...
“Chairman. They say if a wren tries to follow a stork, it will split its legs. To be honest, neither our pension fund nor the Ministry of Science and ICT has that kind of money. Frankly, I’m not hoping for Hynix to develop technology that surpasses Envidi GPUs. Would it be impossible to aim for niche markets that Envidi chips can’t reach with a strategy of cheap, mass-market products?”
“It is absolutely impossible.”
“Absolute...ly?”
Kim Jeongjin was firm.
“Team Leader, people have recently been saying that Big Tech companies are inflating each other’s sales because of Envidi’s credit transaction issues, but that is a very ignorant statement about the market. On the surface, it looks as if they’re joining hands, but if you look behind the scenes, it’s nothing short of a den of demons.”
“...A den of demons?”
“Yes. MS’s Azure Maia, Amazon’s Trainium, Meta’s MTIA... Right now, all the Big Tech companies are pouring themselves into developing their own chips. I believe Big Tech’s efforts to ‘break away from Envidi’ will only intensify in the future. Gogul’s TPU announcement is just one part of that phenomenon.”
I smiled bitterly.
In truth, the contest for the GPU market had already begun. The Big Tech companies were each developing proprietary technologies to reduce their dependence on Envidi, and they were spending trillions of won on it.
More than anyone, those who had mastered money seemed to already know. If you let a single component hold your leash, you would be dragged around for the rest of your life.
“Very well. Then let’s give it a try.”
After a long moment of thought, I opened my mouth.
“The amount you just mentioned will be invested in the form of R&D subsidies and equity purchases after discussion with the government. However.”
“However...?”
“For the next ten years, reduce dividends to around one percent, and sell off the loss-making businesses within Seongyeong Group as soon as possible.”
“...Are you telling us to restructure?”
“Yes. We have to squeeze out every won we have and don’t have to challenge ourselves for the next-generation growth engine, don’t we?”
“U-understood. But dividends around one percent will provoke no small backlash from shareholders... It also diverges from the shareholder return policies the government is advocating.”
“Big Tech companies are also notorious for stingy dividends. Instead, their stock prices rise by more than the dividends, so shareholders don’t complain. Considering Hynix’s recent stock price, I don’t think the backlash will be that large. Please reduce expenditures however you can and devote yourselves to reinvestment.”
Kim Jeongjin thought briefly, then answered.
“Understood. In fact, the recent family court ruling worked out favorably, so I’ve been able to set aside one major worry regarding management rights. As you said, from now on, we will reinvest all the money we earn into future growth engines.”
I clenched my fist.
Even so, in the coming AI era, Korea was in a somewhat better position. Even if we hadn’t found the gold mine, didn’t we have a firm grip on at least one pickaxe called memory semiconductors?
Now all that remained was to continuously reinvest the money earned from semiconductors into the next-generation growth engine. Although it was something absent from my memory, I felt Korea could sufficiently set its sights on that bowl of rice too.
“Send me the funding plan within this week. I’ll contact you after reviewing it. Then, if you’ll excuse me.”
After leaving my seat, I hurriedly picked up my phone.
Before the ringtone could even finish once, an excited voice came through.
—Oh my, Sejun!