Rebirth 08: Rise from copycat phones

Chapter 557 The Foundation of Artificial Intelligence: Trillions of Watt-Hours of Electricity!
After gaining a preliminary understanding of the basic operating conditions of the various business units within the group over the past year, Xu Shenxue...

We also gained a more comprehensive understanding of the situation within the Zhuyun Group.

Overall, Zhuyun Group's smart terminal business remained stable over the past year, while its smart home business was barely surviving.

The PC and server businesses are also just so-so.

In the software and internet business, mobile content services, primarily delivered via smartphones and tablets, are showing steady growth. The virtual content industry has great potential, but due to the limited number of virtual devices, this potential has not yet been fully realized.

The cloud computing and server businesses are mixed. Regular cloud computing is just so-so; the business itself isn't very profitable. It's mainly about absorbing CPU production capacity, shifting from selling server CPUs to selling computing power! GPU server sales are quite good, but this has little to do with the server business itself; it's essentially still selling APO graphics cards…

The financial investment business focused on maintaining stability, with no major highlights or mistakes.

Zhiyun Satellite and Zhiyun Aviation primarily serve the group and its sister companies. Although they generate revenue and profit, the amount is small and negligible. The combined profits of these two businesses are less than what Zhiyun earns from selling phone cases, data cables, Bluetooth keyboards, mice, and other small accessories!
Over the past year, Zhiyun Group has experienced significant growth, especially in terms of profits, primarily driven by various new businesses centered around artificial intelligence technology.

This includes various computing chips essential for artificial intelligence technology, such as APO graphics cards, PX chips, EYQ chips, ZY chips, and LC chips.

At the same time, the development of artificial intelligence has driven the development of computing chips and high-performance storage chip business. High-bandwidth memory used in computing cards such as HBM2 and HBM3 is technologically advanced and expensive. This business has become an important source of revenue and profit growth for Zhiyun Storage.

In an APO graphics card, HBM2/3 memory accounts for 60-70% of the cost.

Virtual devices and intelligent robots are also based on artificial intelligence technology. Among these two businesses, intelligent robots are growing very well and have reached a very large scale, becoming a new source of revenue and profit growth for Zhuyun Group.

Virtual devices have enormous potential, and although the virtual business itself is not yet profitable, it provides a lot of profit for Zhiyun Semiconductor. To be honest, Zhiyun Group can make a lot of money from virtual devices, but these profits are not reflected in the virtual device business, but in Zhiyun Semiconductor's chip business.

Looking at the group's overall operations, it can be seen that artificial intelligence technology, or big data modeling technology, has become the core driving force for the continued high-speed growth of Zhiyun Group!

Relying on artificial intelligence technology, Zhiyun Group has gradually transformed from a smart terminal manufacturer and semiconductor manufacturer into a comprehensive high-tech manufacturer with artificial intelligence technology as its core link, focusing on the development of semiconductor design and manufacturing, smart terminals, smart robots, and virtual industries.

This demonstrates the importance of artificial intelligence technology...

In late January, Xu Shenxue convened a meeting of senior management at Zhiyun Group. Senior vice presidents of the group and several core scientists attended the meeting. In addition, several other executives who were vice presidents of the group but were in charge of the group's core business subsidiaries also participated in the meeting.

The meeting focused on summarizing the company's strategy over the past year and several years, while also outlining the development strategy for the next year and several years, especially the direction of technological development!
During the meeting, the terms "artificial intelligence" and "computing power" were among the most frequently used words!
Professor Lin Andong is the only W15-level scientist in the Zhiyun Group, the Chief Scientist of Zhiyun Group, the Dean of Zhiyun Artificial Intelligence Research Institute, and also holds the title of "Senior Vice President of the Group"!
During the meeting, Lin Andong stated that the group's existing intelligent computing power is severely insufficient, which has hindered the research and application development of artificial intelligence technology.

Currently, even including the intelligent computing arrays of its sister companies Hailan Auto, Fairy Mountain Holdings, and Yihai Technology, the total computing power of the entire group is only about 200 EFLOPS (FP32, 100 quadrillion floating-point operations per second).

Although this computing power accounts for more than 70% of the world's total AI computing power, it is still not enough!

Incidentally, intelligent computing power and general-purpose computing power are not the same thing. General-purpose computing power mainly relies on CPU chips...

In the statistical definition of Zhiyun Group, intelligent computing power is classified as "artificial intelligence computing power," which is the computing power used to train and run artificial intelligence or big data models.

Represented by GPUs in electronic chips and superconducting quantum computers in new technology directions... the new large-scale intelligent computing center built by Zhiyun Group now uses superconducting quantum computers and GPU chips together.

Professor Lin Andong hopes that in the next three years, the group will strive to build 2,000 EFLPOS of intelligent computing power to meet the needs of training ultra-large models and various artificial intelligence business training and operation!
Of course, building such a massive intelligent computing power within three years is extremely difficult... because it will be limited by the production capacity of AI/APO graphics cards.

Meanwhile, the superconducting quantum computer, a new and important component of the intelligent computing power of Zhiyun Group, also has certain production capacity limitations. For example, the production capacity of the MMK34 room-temperature superconductor is currently relatively limited, and there are also some process problems in the processing of advanced quantum chips, making it difficult to achieve low-cost large-scale mass production.

After all, Zhiyun Group's superconducting quantum computer operates with tens of thousands of qubits, unlike those quantum computers outside with only a few dozen or a few hundred qubits that can be easily compared to it.

The YCL-51 superconducting quantum computer used in Zhiyun Group also requires a 'chip'... Although this quantum chip is somewhat different from traditional electronic chips, it still has very high requirements for manufacturing processes.

Its assembly and processing precision is also very high, especially the precision of device stacking, which requires reaching the nanometer level.

In order to automate, scale up, and even make quantum chips cheaper, scientists at Zhiyun Group used their experience in electronic chips as a reference to try to develop and design a complete set of automated production equipment, which applied a large number of traditional semiconductor manufacturing equipment.

They even came up with a branching technology route that allows existing semiconductor manufacturers to produce it directly. However, the quantum chips produced this way are not yet up to par and can only be developed as a technology reserve.

The fifth-generation quantum chips on the YCL-51 superconducting quantum computer are now produced in a laboratory setting using specially manufactured equipment... Not only are they expensive, but more importantly, the production capacity and yield are too low!
Unfortunately, this fifth-generation quantum chip has tens of thousands of qubits, making it extremely difficult to manufacture.

Zhuyun Group has not been able to achieve large-scale, low-cost mass production of the YCL-51 superconducting quantum computer in the short term... This is one of the reasons why the YCL-51 superconducting quantum computer is so expensive.

Quantum chips are all handcrafted by scientists in laboratories using top-of-the-line scientific equipment... In addition, the MMK34 room-temperature superconducting material is also very expensive.

Add to that other costs, especially research and development costs.
As a result, the internal procurement price has reached 300 million per unit!

Professor Lin Andong announced plans to achieve 2,000 REFLPOS of intelligent computing power within three years, while the people in charge of the semiconductor and superconducting quantum computing departments looked grim...

This Professor Lin speaks without thinking, making it sound like intelligent computing power is as common as cabbage, something you can easily get by messing around with.

If you need 700 or 800 EFLPOS within three years, we can grit our teeth and make it for you, but 2,000 EFLPOS, and you want to make it within three years... you're asking too much of us.

Currently, the most advanced super intelligent computing array in the Zhiyun Group is still composed of a large number of YCL-51 superconducting quantum computers and a massive number of AI6000 graphics cards, with a single unit having a computing power of only one hundred EFLPOS.

Two thousand EFLPOS would require twenty such computing arrays...

This has exceeded the production capacity limit of Zhiyun Group!

Unless they use all the computing power themselves and don't sell it to others... but that's obviously not going to work. If they don't sell it to others and earn high gross profits, how can they support the construction of their own computing centers?

Do you think building a computing center is free?!

The strategy of Zhiyun Group is to sell a portion of its computing chips at extremely high prices, obtain huge profits, and then use these profits to build its own computing centers.

If they don't sell, they'll have to bear all the costs themselves, and those exorbitant costs are something even Zhiyun Group can't withstand!
This quantity is too large; it has exceeded production capacity limits.

Besides, even if we manage to develop the intelligent computing power of these two thousand EFLPOS, the power supply won't be able to keep up.

Computing power is extremely power-intensive; behind every computing center lies a massive amount of electricity!

The existing 200 EFLPOS computing power of Zhiyun Group and its sister companies consumes approximately 200 billion watt-hours of electricity annually.

Of course, the existing computing power also includes a large number of APO4000, APO4500, and APO4600 graphics cards... These outdated GPU chips consume too much power, resulting in very high computing costs...

This is also why the APO5000 graphics card with its new technology and manufacturing process is more popular with enterprise customers, because it has lower computing power costs!
The new intelligent computing centers currently under construction at Zhiyun Group, especially those using superconducting quantum computers, consume far less electricity and have lower computing costs, approximately half the cost.

Even so, if subsequent computing centers uniformly adopt this type of superconducting quantum computer plus AI graphics card, based on an intelligent computing power of 2,000 EFLPOS, it would conservatively require trillions of kilowatt-hours of electricity annually.

Currently, the country's annual power generation is only over eight trillion kilowatt-hours... This is already the world's largest power generation, and it is far ahead of other countries. The United States only has four or five trillion kilowatt-hours!

Lin Andong plans to add over a trillion kilowatt-hours of electricity within three years, which is a quarter of the annual power generation in the United States. Where will Zhiyun Group find so much electricity? And electricity isn't something you can just conjure up like cabbage.

The power grid has to build a power plant before it can supply electricity to Zhiyun Group!
Currently, in order to supply power to the computing centers of Zhiyun Group and other domestic enterprises, the power grid is rapidly launching various large-scale power generation projects one after another, creating a bustling scene. Even so, this is barely enough to keep up with the growing electricity demand from domestic industry, civilian use, and data centers.

If you suddenly increase the electricity demand fivefold in three years, there will be an additional demand of trillions of kilowatt-hours. Even the grid workers would be cursing!
Electricity constraints are also a major problem. Why can't many small and medium-sized countries develop artificial intelligence? Because even if Zhiyun Group were to sell them APO graphics cards at full capacity, they still wouldn't have enough electricity…

Following Lin Andong's suggestion, what does it mean to build 2,000 EFLPOS? It's roughly equivalent to the annual power generation of countries like Japan and Russia!

If you want to play with artificial intelligence, you need electricity first!
Many people felt that Professor Lin Andong's suggestion was unrealistic, but Xu Shenxue's suggestion was indeed very necessary... When it comes to artificial intelligence, computing power is the foundation of everything!

Many of the group's businesses, such as Yun AI, which provides a wide range of AI services, consume a significant amount of computing power every year.

For example, in the virtual device business and intelligent robot business, a lot of computing power is needed every year to train big data models... Even the superconducting quantum computer itself needs to use top-level big data models to control qubits, and this data model also needs long-term continuous training for improvement.

Hailan Auto, Galaxy Pharmaceuticals, and Galaxy Life behind it also require a lot of computing power.

Even the various space activities carried out by Nanmen Aerospace, especially the subsequent unmanned spacecraft and lunar bases, require the support of artificial intelligence technology, which means that a lot of computing power is needed.

If artificial intelligence technology has become the core underlying technology of Zhiyun Group, and even the various companies under Xu Shenxue!
Therefore, computing power has become an indispensable infrastructure for all companies under Xu Shenxue's control!

Therefore, Xu Shenxue paid close attention to this issue and pointed out: "Since the existing hardware facilities are insufficient to meet the requirements, what about adopting new technologies?"

"What about AI series graphics cards that use 5nm or even 3nm processes?"

"Technological advancements can significantly reduce computing costs. Compared to 7-nanometer chips, 3-nanometer chips can achieve double or even several times the performance increase with the same power consumption. Doesn't that mean saving 60-70% of the power requirement?"

"Furthermore, the superconducting quantum computer, the sixth generation of superconducting quantum computers has been under development for some time, when will it be able to be built?"

"If this is successfully developed, a 50,000-qubit superconducting quantum computer will be five times more powerful than a fifth-generation quantum computer with 10,000 qubits. This also means that the power consumption will be reduced by several times for the same computing power."

"We're in the high-tech field, so we shouldn't limit ourselves to the present; we should look at problems from a developmental perspective!"

"Two thousand EFLPOS is a lot. According to outsiders, it would require at least two trillion watt-hours of electricity. However, with our new quantum computing array, the power consumption can be reduced to ten thousand watt-hours!"

"So, what if we go a step further and use a sixth-generation superconducting quantum computer with GPU chips made using five-nanometer or even three-nanometer processes? Wouldn't that allow us to control power consumption to 500 billion watt-hours or even lower?"

"An additional two trillion watt-hours of energy consumption is impossible, and meeting one trillion watt-hours of energy consumption is also quite difficult, but is it possible to overcome five hundred billion watt-hours?"

"Achieving 500 billion watt-hours of electricity in three years isn't too much!"

Xu Shenxue's statement brings the issue back to the research and development of semiconductor manufacturing processes and superconducting quantum computers.

Ye Xiangbo, the group's vice president and head of the superconducting quantum computer project, said: "Chairman Xu, our sixth-generation quantum computer project has been basically completed in terms of research and development. In particular, in the key technology of controllable qubits, we have been able to control 50,000 qubits by using big data models for analysis, prediction and intervention!"

"The main obstacle now is the mass production process of the sixth-generation quantum chip. Our sixth-generation quantum chip is quite extreme in terms of technology and process. Its alignment precision has been reduced to fifty nanometers, making it very difficult to stack and process, and the yield is also very low!"

"In order to process this sixth-generation quantum chip, we collaborated with Fairy Mountain Holdings to develop a complete set of ultra-high precision manufacturing equipment. The manufacturing difficulty of some of the core equipment, such as the specially developed 'focused ion beam cutter' and nanoscale micro-manipulation alignment machine,' is not much less than that of EUV lithography machines!"

"However, according to our technical progress expectations with Fairy Mountain Holdings, we expect to obtain the first set of sixth-generation quantum chip processing equipment by next spring. Although the production capacity of this set of equipment is still very limited, it can be designed to produce 1,000 chips per year in a semi-manual laboratory environment. If several sets of equipment are deployed at the same time, it will be enough to initially meet the production needs of the sixth-generation superconducting quantum computer!"

Xu Shenxue said, "So that means I'll be able to see a mass-produced sixth-generation superconducting quantum computer next year?"

Upon hearing this, Ye Xiangbo did not answer immediately. Instead, he pondered for more than ten seconds before saying, "Next autumn, after we obtain the equipment, we will still need to conduct complex testing. The sixth-generation quantum chip also needs to be fabricated. Quantum chip technology is complex, and fabrication is difficult. Give us six months for fabrication and testing, and we should be able to produce the sixth-generation superconducting quantum computer around next autumn!"

At this point, Ye Xiangbo took a deep breath and said to Xu Shenxue and the others, "The performance of the sixth-generation quantum computer will be greatly improved compared to the fifth-generation quantum computer. If used for computing in the field of artificial intelligence, its computing power cost will be reduced by at least five times!"

"We have that confidence!"

When Ye Xiangbo spoke, he had made a huge decision...

There was nothing he could do; Xu Shenxue was being pressured from above…

Xu Shenxue himself has made it clear that artificial intelligence is the core and computing power is the foundation... If current technology cannot achieve this, then we need to look at more advanced computing chips and quantum computers.

As the person in charge of the quantum computer project and other cutting-edge technology projects, Ye Xiangbo had to stand up and express his support for his boss's ideas and strategic plans!
While it remains uncertain whether it can be produced by next fall, Ye Xiangbo is confident that with hard work, they can achieve a decent level of progress and produce some significant technological breakthroughs that are sufficient to explain their achievements, such as developing an experimental prototype. As for the mass-produced model, even if it cannot be produced by next fall, it is highly likely that it can be produced the year after.

Even if we take a step back and say that they really can't develop it by the year after next, it's not his fault. His team has overcome all the technical difficulties of the sixth-generation quantum computer. He and his team have developed the difficult sixth-generation quantum chip and the control of 50,000 qubits.

The inability to mass-produce the sixth-generation quantum chip in time will not be his problem, but rather the problem of Fairy Mountain Holdings... It's their fault for developing equipment so slowly!

As for the problems with Fairy Mountain Holdings, Xu Shenxue, the big boss, will naturally go to Fairy Mountain Holdings to put in a good word; he doesn't need to worry about it.

At this moment, Fu Zhengyang, Senior Vice President of Zhiyun Group and Executive President of Zhiyun Semiconductor, said: "We have completed the design of the APO6000Y/120GB version of the APO6000 graphics card with 5-nanometer process. It is expected that we can start the initial tape-out next month. However, mass production is limited by the capacity of 5-nanometer process. It is not possible in the short term. The earliest mass production will be this fall!"

"The APO7000 graphics card using the 3nm process has no major design issues, and we are actively working on it. However, when it will be mass-produced depends on the technological advancements of Zhiyun Microelectronics' 3nm process and the progress of factory construction!"

Seeing that everyone's eyes were on him, Ding Chengjun of Zhiyun Microelectronics said helplessly, "You all know the situation at Zhiyun Microelectronics!"

"This production capacity isn't something I decide, but rather something everyone decides... It's all built according to the investment funds set in previous years. For the 5-nanometer process, this year's production capacity is 50,000 wafers."

"We at Zhiyun Microelectronics will not comment on how this production capacity is specifically used... Whether it's used to manufacture the S1103 chip or the GPU for the APO6000 graphics card, everyone can decide how to allocate the capacity; we can do whatever we want!"

"The second 5-nanometer factory will not be put into production until next year, at which time the production capacity will be more abundant!"

"As for the equivalent 3-nanometer process, our factory for the 3-nanometer process has just been completed, and the equipment has not yet been installed. It will not be able to start production until the year after next at the earliest, and the initial capacity will only be 40,000 wafers!"

"Of course, if the group is willing to allocate more funds, such as an additional $30 billion, then if we at Zhiyun Microelectronics urgently start construction on a 3-nanometer process factory, working overtime, we expect to be able to rush to complete another factory the year after next, adding another 40,000 to 50,000 wafers of production capacity. But that's not enough, because the production capacity of the HEUV-300C lithography machine that can be used for double exposure at Gulf Technology is just that!"

These words caused everyone in the conference room to fall silent!

Don't expect Gulf Technologies to increase the production capacity of its HEUV-300C lithography machine in a short period of time...

This involves more upstream core cutting-edge technology industries, where the production capacity of each core component is limited, making it extremely difficult to increase production capacity.

This made Xu Shenxue sigh: if you want to develop intelligent computing power on a large scale, funding is just the least of your worries... Zhuyun Group has plenty of money!

The problem is that each replacement of various parts and core equipment involves a complete chain of steps... To boost computing power and production capacity, it is necessary to accelerate the entire top-tier, cutting-edge industrial system, which is extremely difficult.

(End of this chapter)