Lu Mingzhou appeared at the door on time and knocked.
"Old Lu, you seem to be in a good mood. What's the good news?" Han Luyi pointed to the chair opposite him.
"The November financial data has just been compiled, and it's quite a pleasant surprise." Lu Mingzhou smiled, sat down, opened his laptop, and showed Han Luyi a table filled with numbers.
"BugKiller's November revenue was 1.12 million, more than previously estimated. This was mainly due to the impact of the previous code theft scandal. After our open-source Public Relations, we signed several large enterprise clients. The standard enterprise version couldn't meet their needs, so they all opted for customized solutions. This part accounted for approximately 300,000 in additional monthly revenue."
Han Luyi nodded. BugKiller is the company's first product line. Since its launch, it has been growing steadily. Although it is not eye-catching, it is increasing every month and has also gained large enterprise clients. Everything is going well.
"The main event is Open Things." Lu Mingzhou tapped the screen, "Including Pro members and independent deployment rebates, the November income is 3.46 million." Han Luyi was both surprised and delighted to hear this.
When Kaiwu Open Things in October, the estimated monthly revenue was only 800,000. 800,000 is certainly not bad, but it more than quadrupled in one month, which can be said to have exceeded his most optimistic expectations.
Half of this benefited from the collapse of All Things Alive, and the other half from the surge in traffic following GAIS.
Open Things has replaced BugKiller as the cash cow for Source Code companies, and a large part of this credit goes to Su Niannian for her intuitive product judgments that served a wide range of non-technical users, not just programmers.
"The two combined"
"Four million five hundred and eighty thousand," Lu Mingzhou finished for him. "That's a little more than the upper limit of the range estimated last time. After deducting Computing Power, manpower, and maintenance, the company's net operating profit for November is just over one million two hundred thousand."
Han Luyi's lips curled slightly.
Two months ago, the company had only eight people, crammed into a small shared office in Zhangjiang Park. At that time, the profit was 600,000 yuan. After financing, the company expanded rapidly, growing from eight people to fifty people in less than two months. We moved into the Qiantan Center office building, and the labor cost increased more than tenfold. Salaries alone cost more than one million yuan every month.
Even under these circumstances, it was still profitable, and the profit was double that of before, Han Luyi himself did not expect.
"And there's even more growth in December." Lu Mingzhou turned to the next page. "Qianfan White Label integration is live. The first week's data just came out, and the average daily usage is almost four times that of the beta testing period. Q4 is peak season. Many merchants on Thousand Sails are creating promotional pages, and many medium and large e-commerce companies have year-end budgets that they need to spend quickly, so the demand is huge. Based on the trend from the first week, the estimated revenue from Thousand Sails in December is 3.87 million, the Computing Power cost is 1.64 million, and the net profit is 2.23 million."
One external connection, almost catching up with all of Open Things's users.
"The peak season will pass; what are your expectations regarding its impact on revenue?"
"It will definitely drop a bit, but Thousand Sails has a large customer base. Even in the off-season, the monthly income is conservatively estimated to be between 1.5 and 2 million." Han Luyi looked at the numbers on the screen.
BugKiller has a base of over one million, Open Things has over three million as its main source, and Thousand Sails adds another layer. Based on this calculation, the company's monthly revenue in December could potentially exceed eight million.
"Operations are completely stable." Lu Mingzhou closed the folder, his excitement still hard to hide. "I've been in this industry for over ten years, and it's rare to see a company turn a profit in its first year. Revenue like ours, with this profit margin, is truly unique." "Seeing how excited you are," Han Luyi smiled. "I have some more good news."
"What?"
"I just had a meeting with Zhao Wenyuan. The model team has made a new breakthrough in post-training, improving accuracy from over 60% to nearly 85%. Previously, Open Things might require four or five interactions with the model, but with the improvement, two are enough."
This means that the cost of reasoning can be reduced by another half.
Lu Mingzhou reacted quickly, his eyes widening in surprise. "The cost of reasoning can be reduced too? By how much?"
"Reduce it to half. Both Open Things App and Thousand Sails have reduced their reasoning costs by that much."
Lu Mingzhou couldn't help but gasp; the number was truly astonishing.
Haha, Old Lu, I love seeing you get scared, Han Luyi thought to himself.
Lu Mingzhou quickly typed a few words on his computer, then looked up with a changed expression. "The combined reasoning costs of 'December Open Things' and 'Thousand Sails' can be reduced by over 1.2 million, directly boosting the company's net profit to over 3.5 million."
He tapped the keyboard a few more times, then chuckled as he did his calculations, "Since the Heavenly Craft project was launched, we've spent a little over twenty million on Computing Power leasing, data cleaning, and retraining. With improved inference efficiency, the Computing Power savings alone will bring back about ten million a year. We'll break even in two years, not even counting the user growth resulting from improved accuracy. This is an absolute steal."
Han Luyi said, "Old Lu, you were against me investing in model making back then, but now you know how good R&D is, right?"
"But," Han Luyi changed the subject, "That's not enough."
Lu Mingzhou looked at him.
"Regarding the model's future development, Zhao Wenyuan outlined three technical routes. The most valuable one, the self-training base model, will cost 30 million RMB." Han Luyi held up three fingers. "The starting price... every time it crashes and needs to be rebuilt, that's another 30 million."
Lu Mingzhou remained silent.
A monthly profit of 3.5 million would require nine months of saving without spending a penny, and failure wouldn't be an option.
"We have enough food to keep us alive, but not enough to fight a war," Han Luyi said.
The two of them were quiet for a while.
"Do you know how Large Model companies burn through money?" Lu Mingzhou suddenly asked.
Han Luyi glanced at him. Old Lu had worked in finance for over a decade and had many friends in the investment circle; his understanding of the industry's financial models was deeper than Han Lu's. "I recently talked to some friends who invest in AI," Lu Mingzhou leaned back in his chair. "The financial models of Large Model companies are completely different from ours. Their logic is 'burn first, then reap'—the first three years are almost entirely unprofitable, with Computing Power expenses starting at several hundred million yuan a year, and revenue entirely dependent on API call volume to slowly climb. The top companies burn through over a billion yuan a year just on training fees, relying on investors willing to keep pouring in round after round."
"We are different," Han Luyi said.
"Completely different." Lu Mingzhou nodded. "The Source Code follows the logic of application companies—the product makes money, users pay, the cash flow is positive, and it generates revenue every month. The logic of Large Model companies is to lay a solid foundation first. No matter what you build on top of it, whether it's an API or a to-C product, as long as the foundation is in my hands, the ecosystem is in my hands. That's why they dare to lose money, because they're betting that in the end everyone will have to stand on their foundation."
He glanced at Han Luyi: "We're building on someone else's foundation. If the building is well-constructed, we'll have many customers and make money quickly. But the foundation isn't ours."
The problem is, you want to build your own foundation now.
Han Luyi remained silent.
"The advantages of building your own foundation are obvious: you're not controlled by others, costs are manageable, and performance can be optimized for specific scenarios," Lu Mingzhou said. "But the disadvantages are also clear. Once you step into the Large Model race, financially you become an amphibious warrior: on one hand, you have to maintain the revenue generation capacity of the application side, and on the other hand, you have to withstand the money-burning intensity of the model side. Balancing both is much more difficult than focusing on only one side."
"You think it's not feasible?"
Lu Mingzhou thought for a moment, then carefully chose his words: "With the current profit level, feeding it once is possible. But training isn't a one-time thing; the model needs to be iterated, the data needs to be updated, and Computing Power will only become more and more expensive."
"What if we also raise funds? Like those Large Model companies, we can fill in round after round," Han Luyi said. Of course, he knew the answer couldn't be that simple, but he wanted to hear Lu Mingzhou's opinion first.
"We can raise funds, and this story is easy to tell," Lu Mingzhou nodded. "But you need to think carefully about the costs. Why can Large Model companies burn through billions? Because from day one, they've followed that Luyi— high valuations, rapid dilution, and by Series C, the founding team might only have single-digit percentages of shares left. Investors are betting that this company will become infrastructure, a winner-takes-all scenario, so they're willing to accept temporary losses."
He glanced at Han Luyi: "Boss He knows we're making models. You mentioned it to him during the Series A funding round, and he didn't object to the 30 million post-training budget. But that was 30 million, fine-tuning someone else's foundation. What you're talking about now is a self-trained Large Model, 30 million each time, starting over if it fails, and iterating continuously. That adds up to hundreds of millions, which is completely different from what we agreed on back then."
Lu Mingzhou continued, "Ultimately, the business models of application companies and model companies are different. Investors won't understand overly complicated stories; they'll just ask, 'What kind of company am I investing in?'"
Both of them?
"Both mean that conditions on both ends must be met: the application side needs to maintain its revenue, and the model side needs to withstand the cash burn. Profit margins are lowered, the growth story becomes more complicated, and in the next round of financing, you either have to accept higher dilution to raise money, or convince investors that you can win two battles on two fronts simultaneously," Lu Mingzhou said. "The first path will result in a significant dilution of your shares; the second path has almost no precedent."
Han Luyi sat there in silence for a while.
What if... we don't do it in the Source Code?
The thought suddenly popped into his head, and he was taken aback for a moment.
The model business was spun off into a completely new entity, with separate financing and accounting. Source Code Technology focused on developing applications, becoming the company that "makes money standing up"; on the other hand, a different logic was used to run the model, burning the fire that might lead to AGI.
Han Luyi instinctively swallowed his words and didn't say them aloud.
A cruise ship sailed by on the Huangpu River outside, and the distant sound of its horn drifted over.
Lu Mingzhou was still waiting for his response.