Rebirth: Starting from Lighting Up the Tech Tree

Chapter 123 The Supply Chain War

The supply chain situation deteriorated faster than Zuo Cheng had anticipated.

Five days after the contract was signed, Fang Ze walked into the office with a list in his hand, his expression colder than usual.

"SMIC has officially notified us that the next batch of chip packaging orders will be delayed due to capacity adjustments." He placed the list on Zuo Cheng's desk. "Hanwei Electronics has also sent an email saying that the price for signal conditioning chips needs to be renegotiated, with the price increase raised from 30% to 45%."

Zuo Cheng glanced at the list but didn't say anything.

"And then there's Jingrui Technology," Fang Ze continued, "they rejected our order outright, saying their entire inventory was pre-ordered by a major client. I asked them which major client it was, but they wouldn't tell me."

Three people, three suppliers, three different methods. But all leading to the same result: choking off 402.

"It's not a production capacity issue," Zuo Cheng said. "It's because someone is controlling the supply at the source."

Fang Ze revealed the equity structure of the three suppliers. Tianxing Capital invested in them successively in the first half of this year, holding between 15% and 30% of the shares, which is enough to influence production scheduling and customer priority.

"Zhao Tianxing, the actual controller of Tianxing Capital, and Lin Jianhua were college classmates," Zuo Cheng said.

Fang Ze was silent for a few seconds, then said, "Yin."

Zuo Cheng calculated the inventory: Cambricon had used 12 of its 50 sample wafers, leaving 38 for testing. The Sky Dome Phase IV project required aerospace-grade chips, but they didn't have any. The customized MLU270-T version would be available in two to three months at the earliest.

The development progress at Blue Bay is urgent; once the contract is signed, it must be started. If the chip supply can't keep up, all delivery milestones will have to be postponed.

"How long can the current inventory last?" Zuo Cheng asked.

Fang Ze calculated: "At the current rate of consumption, including chips used for testing and development, it will take about two weeks. If the development of Sky Dome Phase IV officially begins, the consumption will double, then it will only last for one week."

A week.

This is not a comfortable number.

Zuo Cheng instructed Han Lu to immediately implement the alternative plan. Han Lu acted swiftly, contacting six potential suppliers that afternoon, covering chip packaging, signal conditioning, and power management.

The results are not optimistic. Of the six companies, three are already operating at full capacity, two are offering prices 40% higher, and only one is willing to accept the order, but delivery will take six weeks.

"Tianxing Capital's reach extends further than I imagined," Han Lu said. "Large companies dare not offend Huaxin, and small companies, once controlled by Tianxing, have almost no alternatives."

Zuo Cheng paced around the office twice. This wasn't a simple business dispute; Tianxing Capital had a planned strategy of strategically positioning itself along the industry chain, specifically targeting startups reliant on upstream suppliers. Moreover, someone had provided precise intelligence behind the scenes. Lin Jianhua knew which suppliers 402 used and that the Cambricon customized version hadn't been released yet—information that wasn't publicly available.

At 3 p.m., Zuo Cheng called the core team to the conference room.

Shen Yiming sat in the corner, Fang Ze leaned back in his chair, Han Lu held a notebook, and Chen Hao stood next to the whiteboard. Liu Wei was away on a business trip and participated in the meeting via voice message.

"Everyone knows the situation," Zuo Cheng said. "The supply chain is blocked, and there's no alternative in the short term. Cambricon's customized version will take at least two or three months. Blue Bay is pressing us hard, and we can't delay."

He paused, then asked, "Does anyone think of another way?"

Chen Hao spoke first: "If we can't get around the hardware issues, can we compensate at the software level? For example, by optimizing the algorithm and reducing the performance requirements of the chip?"

Zuo Cheng glanced at Shen Yiming. Shen Yiming, who had been lost in thought, suddenly looked up when he heard this.

"Brother Cheng," he said, "could you explain in more detail?"

Chen Hao paused for a moment, then said, "Exactly. If the chip's performance isn't enough, we can compensate with better algorithms. The same effect requires less computing power, thus reducing our reliance on the chip."

Shen Yiming's eyes lit up.

"This direction is feasible," he said. "Our previous approach was to run the compressed model on a high-performance chip. But if we think about it the other way around, making the model itself lighter, we don't need such a powerful chip; we can run it directly on an off-the-shelf, general-purpose chip."

"What do you mean?" Fang Ze asked.

"Software defines hardware." Shen Yiming walked to the whiteboard and drew a simple diagram. "The core idea is to move some computing tasks off the satellite and distribute them to surrounding satellites via inter-satellite links. Each satellite only runs the lightest decision-making layer, while the complex calculations are handled by the cluster collaboration."

Zuo Cheng stared at the diagram on the whiteboard, his mind racing.

The essence of this solution is to bypass the chip bottleneck. If each satellite only needs to run a lightweight model, a regular general-purpose chip is sufficient. There is no need for a high-performance AI chip like the MLU270, nor is there a need to wait for Cambricon's customized version.

But there is a key issue.

"Is the delay in inter-satellite communication acceptable?" Zuo Cheng asked.

Shen Yiming stopped writing and turned to look at him.

"If the latency is too high, the collaborative decision-making between satellites will not be able to keep up, and the entire system will be rendered useless."

Shen Yiming thought for a moment, "Theoretically, the latency depends on the bandwidth and protocol of the inter-satellite link. Currently, the Tianqiong satellite uses a traditional communication protocol, with a latency of about tens of milliseconds. This should be sufficient for simple decision-making interactions. However, for complex real-time collaborative calculations, this latency is insufficient."

The meeting room fell silent for a few seconds.

Zuo Cheng looked at Shen Yiming, and Shen Yiming looked at him. They both understood that the problem lay in the communication protocol.

"If we could develop a low-latency inter-satellite communication protocol," Shen Yiming said slowly, "this solution would be entirely feasible."

Fang Ze frowned. "Low-latency inter-satellite communication protocols aren't readily available. The communication layer for Project Sky Dome was developed by Blue Bay Communications themselves; any modifications would require coordination with them."

"It's not about modifying the existing one, it's about creating a completely new one. A communication protocol specifically optimized for satellite-based distributed AI computing."

Zuo Cheng leaned back in his chair, his fingers tapping lightly on the table.

Low-latency inter-satellite communication protocol.

He knew this area. The system panel had two main branches: communications engineering and aerospace communications, with many related components. If he could integrate a few of them, he might be able to create what he needed.

The problem is that the technology radar is still cooling down, and it will take at least another day.

But the urging from Blue Bay won't wait for him.

In the meeting room, everyone was lost in their own thoughts. Han Lu was writing rapidly in her notebook, Chen Hao was staring blankly at the whiteboard, and Fang Ze adjusted his glasses.

"Give me a day," Zuo Cheng said.

Everyone looked at him.

"Han Lu, keep looking for the supply chain. Don't stop; find one that suits you," he said. "As for the software solutions, Yi Ming, today you need to outline the technical framework and clarify the latency requirements and protocol specifications. Fang Ze, assess which models could be used if we were to replace them with general-purpose chips."

He stood up. "We'll have a meeting a day later to finalize the plan."

Everyone nodded.

Zuo Cheng walked out of the conference room and stood in the corridor for a while.

Sunlight streamed through the glass, the autumn light already carrying a winter chill. He glanced at the system panel; though he couldn't see anything there, he knew what was inside.

Technology radar. Cooling down. More than thirty hours remaining.

He needs to be fast.