Ryan is still waiting for an answer on the other end of the internet.
Han Luyi had already figured out his thoughts.
To fool others, you must first fool yourself.
The time to showcase your acting skills has arrived.
"Making base models is not the direction of Source Code Technology," Han Luyi said.
Ryan raised an eyebrow slightly, waiting for what was to come.
"Heavenly Craft is a code-specific model," Han Luyi said. "It ranks among the top in the industry for Chinese code generation, not because of the number of parameters, but because of the closed loop of product behavior."
He moved forward slightly, his gaze fixed on the camera.
"The Source Code aims for an ultimate closed-loop experience from requirement to delivery. We created Heavenly Craft because there wasn't a suitable code model for our scenario on the market at the time, so we had to build our own. But creating Heavenly Craft isn't the goal of the Source Code; the product is."
Ryan didn't respond immediately, but instead added, "The upper limit of a product's closed loop depends on the upper limit of the underlying model's capabilities. Your code model is very strong in Chinese scenarios and also in small-scale commercial scenarios; but what about English scenarios? Multilingual scenarios? How about handling complex business logic? Once your user base grows to a certain scale, you will definitely need to expand to more scenarios and more regions. At that time, will the model still be able to handle it?" This question was direct, sharp, and hit the nail on the head.
"We can't hold on any longer." He was quite frank. "Heavenly Craft does have its limitations in English-language scenarios, and you know that better than I do. But that precisely illustrates my judgment: the model isn't Source Code Technology's barrier; the product is. If a better model emerges, Source Code Technology's products will integrate with that better model. Heavenly Craft is currently Open Things's best option, but Source Code Technology doesn't have that kind of arrogance; we won't be tied to any single model."
Ryan nodded, but didn't let the topic drop.
So you mean, the Source Code should always be for the application layer, and the model should always be someone else's?
Han Luyi shook his head and held up one finger.
"It's not about 'always using other people's models,' but rather that the model itself is not the end result of the competition."
"A few years ago, there was only one top-tier pedestal model in the world, the GPT-3. Now there are five or six, and next year there may be more than a dozen, and the gap is rapidly narrowing."
"When infrastructure becomes a commodity, value moves up the value chain."
"I'm afraid I can't completely agree with that assertion," Ryan said. "The model isn't fiber optic."
"My judgment is based on my own reasoning," Han Luyi said calmly.
Ryan gestured for him to continue.
"That's how the internet was in its early days." Han Luyi said, while opening his Sight to check some data. Data would make people more convinced that he had done research.
"In the late 1990s, Wall Street's wildest money was poured into infrastructure: fiber optics, routers, and switches. AT&T &T spent over $100 billion acquiring cable television networks, Global Crossing spent billions laying submarine fiber optic cables. The biggest consensus at the time was: whoever owns the pipeline owns the internet."
He held up his hand and made a five.
In March 2000, Cisco, a company that sells routers and switches, surpassed Microsoft to become the world's most valuable company, with a market capitalization of $550 billion. That year, the global telecommunications industry raised more than $1.5 trillion in total funding.
"What about website developers? Their valuations are insane too. Amazon was over $30 billion at its peak, and Yahoo over $120 billion. But Wall Street's logic is—these companies are valuable because there are people laying the infrastructure underneath. The infrastructure is the real asset."
Ryan kept looking at the screen without interrupting.
He knew what happened next, but his thoughts were drawn in by Han Luyi's logic.
"People say the bubble of 2000 was the dot-com bubble, but in my opinion, it was the ISP bubble."
"Looking back now, the so-called bubble in the application layer was just a mismatch in timing. Amazon's $30 billion market capitalization back then is practically a gift today at $2.5 trillion. What truly burst and never recovered was the infrastructure sector."
"After the bubble burst in March, WorldCom went bankrupt due to fraud, wiping out $107 billion in assets; Global Crossing went bankrupt, and dozens of telecommunications companies collapsed."
"But after the ruins were cleared, who survived? The companies that laid the pipelines collapsed and never rose again. On the other hand, application layer companies: Amazon rose from the ashes after its market value plummeted by 90%, Google started making profits at the lowest point of the bubble, Facebook grew wildly on the fiber optic cables laid by others. Today, the world's ten most valuable companies are almost all application layer companies. And Cisco? Twenty-five years later, its market value is over two hundred billion US dollars, and it hasn't even returned to half of its peak."
"Looking back now, the combined market capitalization of Apple, Microsoft, Google, Amazon, and Meta—these five application layer companies—exceeds ten trillion dollars. All global telecom operators and equipment manufacturers combined are worth just over two trillion. That's a five-to-one ratio; the infrastructure has become a utility, and the value lies entirely in the application layer." "The same applies today. The foundation model is important, but the greatest value lies in the application layer."
"This is my judgment."
Han Luyi concluded, then looked at Ryan on the screen.
Ryan paused for a moment before finally speaking, "AI models are different from internet service providers. Internet service providers don't have the flywheel effect, but models do. The more people use them, the more data they generate, and the better the model becomes."
"You're right," Han Luyi said, "But the flywheel isn't a new concept. It's been used at the application layer for many years. The more user behavior data, the tighter the product loop, the better the experience, and the more users. Once this flywheel starts spinning, the underlying model can be changed, but the user habits and user acceptance at the top cannot be changed."
He looked at Ryan on the screen.
"Whoever can translate the model's capabilities into what users truly need will capture the value. It's not about who has the biggest model, but who understands users best. Your advantages in infrastructure are clear, and I won't use the Source Code resources to do things where I have no advantage. The goal of the Source Code is to maximize the product experience at the application layer, enabling everyone without technical expertise to solve their problems using AI."
This was a very aggressive statement. To put it bluntly: you're just providing me with services by making models, while I'm on the right track by developing applications. If Han Luyi really thought that way, he probably wouldn't have dared to say it like that. But it was precisely because he knew Ryan had a different opinion that he used these words to lower his guard.
Ryan fell silent again on screen, as if frozen.
"That's a good answer," Ryan finally said.
He did not agree with Han Luyi's judgment, but Han Luyi's logic was completely autonomous, which shows that he was not just saying it casually, but had really thought about it seriously.
The conversation yielded more results than he had expected.
First, it confirms that the CEO of Source Code Technology is indeed in a different direction than he predicted, posing no threat and making cooperation more reassuring. Second, the depth and breadth of this thinking itself demonstrates the capabilities of a CEO.
When looking for a partner, he naturally prefers someone with strong capabilities.
"I appreciate your thinking, but there's something I must share."
Ryan canceled screen sharing and looked directly at the camera.
"We're not the only ones working on automated workflows. I have information that Google internally announced a similar project last week. I don't know the codename, but the content is almost identical to the direction we've shown. Meta is also making moves in this area."
"Whether or not to do it is no longer the question; the real question is who will do it first."
He paused for a moment.
Time is tight.
"If you'd like to collaborate, we can release an overseas version of Open Things based on our model within a month, ensuring localization and compliance, and quickly achieve meaningful user growth." Simultaneously, both parties can collaborate on developing automated workflow products and launch them as soon as possible.
"But if you want to create a generic model from Source Code— " Ryan's tone remained unchanged, "You'll need to invest at least a billion dollars, and take a year—maybe more."
It could be more time, or it could be more money.
Upon hearing this, Han Luyi knew that Ryan naturally believed that if the Source Code wanted to break through its limitations, the next step should be to move towards a more general approach. This judgment pointed in the same direction as what Han Luyi was currently doing.
Ryan simply assumed Han Luyi wouldn't choose him—he didn't know Zhao Wenyuan was already training the prototype.
"Thank you, Ryan," Han Luyi concluded in standard business English. "This was a very constructive conversation. We need to conduct an internal evaluation, and we'll get back to you once the founding team is aligned."
"Of course, I understand, no rush." Ryan nodded with a smile, then added, "But whether you agree or not, our pace won't slow down."
Aren't you in a hurry?
"Looking forward to your message, Louis."
"Me too, looking forward to our next conversation."
The screen went black.
The meeting room was suddenly unusually quiet. The ambient noise that had been playing on the screen for almost an hour had all disappeared, leaving only the soft hum of the central air conditioning.
Su Niannian turned to look at Han Luyi: "Is that really what you think?"
Han Luyi laughed heartily, his shoulders relaxed, and he turned to look at Su Niannian.
"Of course it's a lie. What, you believe it too?"