Tech Giants: Starting from Bankrupt Companies

271: Chapter 271 Zhulong’s Basic Artificial IntelligenceThe lights continued to dim.

In the center of the stage, the display cabinets were removed, and the robots exited.

In their place was a beam of light.

The big screen lit up, and in the void of darkness, a swirling halo of light emerged.

It was not a regular circle, but a constantly shifting, nebula-like structure, reminiscent of the ancient myth where ”the Zhulong opens its eyes to bring day and closes them to bring night.”

No physical form, no logo, no brand identification.

Only light.

Xingchen Zhulong · Primary Artificial Intelligence

”What we are about to release next has no physical object to display,” Su Chen’s voice rang out in the darkness. ”Because it is not hardware; it is a method.”

The big screen popped up a technical white paper:

Zhulong V1.0 Technical Specifications

Parameter Scale: 175 Billion

Training Data Volume: 45TB

Multimodal Capabilities: Text / Image / 3D Model / Gene Sequence / Molecular Structure

Inference Speed: 0.8 seconds / task

Deployment Method: Local Privatized Deployment + Federated Learning

Price: $10 Million (¥70 Million)

”This is not a chatbot,” Su Chen emphasized. ”It is a scientific research productivity tool.”

Offstage, those technical executives from multinational pharmaceutical companies, semiconductor giants, and automobile manufacturers simultaneously set down their notebooks.

Live Demonstration One: Biomedicine

On the left side of the big screen was an input box, and on the right was the real-time output.

Su Chen dictated the instructions, and the technical team entered them on the spot:

”Targeting the KRAS G12C mutation, design five candidate small-molecule inhibitors, requiring a blood-brain barrier permeability > 30% and oral bioavailability > 50%.”

KRAS G12C—one of the most difficult mutation targets to treat in lung cancer, known as an ”undruggable target.” Amgen spent 20 years before launching the first inhibitor in 2021.

Zhulong running.

18 seconds.

Output on the screen:

Candidate Molecule 1

Structural Formula: 2D Rendering + 3D Conformation

Synthesis Pathway: 4-step reaction, starting materials commercially available

Predicted IC50: 12.3 nM

Predicted Blood-Brain Barrier Permeability: 37%

Predicted Oral Bioavailability: 62%

Patent Assessment: Uncovered

Candidate Molecule 2

...

5 molecules, all with complete data provided.

Comparison on the side: The traditional computer-aided drug design process requires 3 weeks—3 weeks of time used for searching literature, building databases, running molecular docking, and optimizing lead compounds.

In the VIP area offstage, Pfizer’s Global President of R&D, Emily Watson, sat up straight abruptly.

She turned to her assistant and said, ”Within ten minutes, I want to see the background information on the Xingchen AI team.”

”Are we going to buy it?”

”Not buy,” Emily stared at the screen. ”Collaborate. If they are willing to open the API, Pfizer’s early drug discovery cycle could be shortened by 80%.”

Live Demonstration Two: Industrial Digital Twin

The second task.

Su Chen signaled the engineer to connect to the real-time data stream of the Xingchen Huizhou chip factory.

The right side of the big screen switched to Zhulong’s task interface.

Instructions:

”Analyze the current operating status of Lithography Workshop No. 3 and predict potential equipment failures in the next 72 hours.”

Lithography Workshop No. 3—the core area where the Sui-Guang 16 lithography machine is located. Any unplanned downtime results in losses in the millions of dollars.

Zhulong running.

20 seconds.

Screen output:

Lithography Machine No. 3: Electrostatic Chuck Temperature Anomaly

Current Temperature: 23.7 ℃

Historical Baseline: 22.5 ± 0.3 ℃

Trend: Slowly rising over the past 6 hours

Prediction: After 72 hours, the temperature will rise to 25.2 ℃, exceeding the specification limit

Failure Probability: 87%

Recommended Action: Perform electrostatic chuck cleaning and maintenance within 72 hours

The synchronously generated 3D digital twin model rotated on the screen, marking the location of ”Machine No. 3 Electrostatic Chuck” with a red heat map.

Offstage, the representative from TSMC turned pale.

The representative from Samsung began taking notes.

Zhang Rujing from SMIC nodded slightly—he had signed the 28nm technology license, and now seeing Xingchen’s AI capabilities, that decision looked more and more like the right choice.

The screen cut to a live feed of the Huizhou factory.

An engineer walked toward Machine No. 3 with a tablet, opened the casing, and inserted an endoscope probe into the gap of the electrostatic chuck.

The screen displayed: visible dust particles attached.

The entire venue was silent.

Then it erupted.

”This is not AI,” Klaus Richter, President of Digital Industries at Siemens, Germany, stood up, his voice dry. ”This is an industrial oracle.”

He did not use the word ”Oracle”—he used the German word: Orakel.

Everyone was chasing ”the larger the parameter count, the more advanced.” DeepMind Gopher’s 280 billion, Google Switch Transformer’s over a trillion—but these models could not be commercialized, could not be deployed independently, and training them once cost tens of millions of dollars in electricity.

The definition of ”Zhulong’s” advancement: not ”the most parameters,” but ”the highest output per unit of computing power,” ”the lowest deployment cost,” and ”the most complete multimodal integration.”

Compared to GPT’s pure text, Zhulong’s advantage lies in its tri-modal model, which can directly process real business data such as product images, customer service recordings, and user reviews.

Zhulong’s source of profit is not selling hardware, but teaching pharmaceutical companies, industrial giants, and government clients how to use this $10 million supercomputer to solve actual problems.

Subsequently, there was another 15 minutes of ordering time, with major pharmaceutical companies placing orders first.

The big screen refreshed:

120 global companies/institutions subscribed to the Zhulong AI system

Pfizer (USA): 5 sets

Novartis (Switzerland): 3 sets

Roche (Switzerland): 3 sets

SMIC (China): 2 sets

CATL (China): 2 sets

Airbus (France): 2 sets

Siemens Medical (Germany): 2 sets

Samsung Electronics (South Korea): 1 set (pending approval)

TSMC (Taiwan): 1 set (pending approval)

...

Total amount: 120 sets × $10 million = $1.2 billion

$1.2 billion, not for selling products, but for selling decision-making capabilities. This was just the first batch of orders; once they tasted the benefits, they would only become more dependent on Zhulong, eventually becoming loyal users of Zhulong.

At 2:10 PM, the stage lights lit up again, returning to a neutral white.

The Chixiao robot was pushed back onto the stage, standing to the right of Su Chen.

Su Chen did not look at the teleprompter, nor did he look at the big screen.

He turned, facing the audience, as if communicating with students in a classroom.

”Humans spent millions of years evolving hands and brains,” he said. ”We used these two things to create civilization.”

He paused.

”But over the past two hundred years, the Industrial Revolution turned humans into extensions of machines—repeating the same action on an assembly line, processing the same type of spreadsheet in an office, searching for the same batch of literature in a laboratory.”

”This is not technological progress; this is technological alienation.”

The Chixiao robot turned its head slightly, its LED eyes facing Su Chen.

”Chixiao is not here to replace humans,” Su Chen said. ”It is to liberate humans from repetitive labor, to do things that require creativity, empathy, and aesthetic judgment.”

”Zhulong is not here to replace scientists,” he continued. ”It is to pull scientists out of the ocean of literature to think about truly important problems—disease, energy, climate, interstellar civilization.”

He was silent for three seconds.

”This is Xingchen’s understanding of the intelligent era.”

Augmentation, not replacement.

The applause lasted for 30 seconds.

Not a polite 30 seconds—no one left, no one checked their watches, no one pretended to applaud.

It was a true 30 seconds.