Discussion keeps the world turning.
This is Roundtable.
The digital gold rush is on, and this time it's not about data, it's about computing power.
Discover why China's tech scene is betting billions on this new frontier.
And step into relaxation with us.
Foot therapy and traditional Chinese medicine wellness practices are turning tired feet into pure bliss for China's busy workers.
We'll talk about why folks are lining up for a soak and a squeeze.
Coming to you live from Beijing, this is Roundtable.
I'm He Yang. For today's program, I'm joined by Steve Hatherley and Yu Shan in the studio.
First on today's show, did you know that you can rent computing power just like office space?
With China's AI industry booming, computing power leasing is poised to become a 260 billion yuan, or 36 billion US dollar market by 2026.
Advanced AI technology isn't just for tech giants anymore.
China's computing leasing, well computing power leasing boom is empowering businesses of all sizes to scale smarter and faster.
But is this a gold mine for everyone or just a select few?
So guys, break it down for us.
Non -techies, what is computing power leasing?
I mean, this is the first time for me to get in contact with this concept as well.
So bear with me if I'm getting a bit slow.
But first, let's just take a look at this computer power leasing business model.
So this is according to a Science Direct article from this August that computer computing power leasing is this business model that allows some enterprises to first of all build their own computing power centers by purchasing computer power servers.
And once that is done, they have their own servers already equipped, they can rent that computing power to other companies and collect to rent.
It's like building a house yourself and then rent it to other people.
Yeah, like instead of buying a hotel then you just rent a space to stay.
Yeah, this is a mouthful, isn't it?
To talk about a topic like this because it's not something that we see pop up in the news every day.
But I found a description that I thought was pretty easy to understand.
So as you explained it, Yushan, these companies, if they have the resources, it takes a lot of money to build a center like this where you can house all of these computers.
And once they do that, if they do that, then you can think of that as kind of like a giant global computer, if you will, or a national computer where individuals, companies can rent pieces of that on a user basis, meaning day by day if that's what they need it to be.
And it's a smart way for businesses to save money getting these super powerful computations done on demand.
And it also allows them access to special hardware that they might not be able to buy themselves.
Because a lot of businesses, especially these days with AI, AI is just one example.
They need really powerful computers to do things like train AI models or run scientific simulations or create 3D animation.
Those are other examples as well.
But buying and taking care of those machines is really, really expensive for companies.
So they look elsewhere.
And a decentralized GPU network is a very clever solution for companies that need it.
So we're talking about both sides.
And also, it's kind of like confusing, especially when you're talking about a complicated tech matter in English, because there's the lease and rent.
So you've got all right, you've got these service providers, or let's say landlords, they have the resources to set up these digital data centers with massively powerful GPUs or graphics processing units to provide all this horsepower.
And then you have these smaller companies who are like, oh, we're too poor to, we don't have the resources to set up this hardware ourselves.
So why not do we rent it from someone?
Another big company.
And these are the two sides of companies in this huge business industry.
Correct. And you said that some companies, they might not have the money to operate these computers, buy these computers themselves.
But there's another side as well.
They might not need that level of computing power on a daily basis.
So when they need it, they can pay for it.
Remember, this is a pay per use system.
Pay as you go. So some companies, on one particular day, they might need a tremendous amount of computing power.
But for the next three or four days, they might not need that much at all.
So this is a really great solution for companies who find themselves in that boat.
Right. And apparently nowadays here in China, the market of this business is getting bigger as well.
So well, this is according to a report by Zhong Yuan Puhua Industrial Research Institute that is based in Shenzhen in South China.
So the report revealed that the potential revenue of the domestic computing power leasing market is expected to reach 260 billion Yuan, which is roughly 35 .8 billion US dollars by 2026.
And it will grow rapidly at a rate of over 20 % per year.
I think that's like a pretty high rate for growth for a relatively niche market that's just been introduced in recent years, I think last year in 2023.
Yeah. Mid 2023 was the time when this became a buzzword in the tech world.
But we also need to sort of put things into perspective that is in tech world, in the tech pace, then what's that?
An hour? Sorry, a year and a half is like a century.
So much happens so rapidly.
That's a fair point.
Not in real life as much.
Fair point. Yeah. Well, and with this growth, we're seeing a lot of big players now entering the game.
Internet giants, including China Mobile, Alibaba, Baidu, Tencent, Inspur Group, they're laying out the computing power leasing industry, or they've entered into it.
And there's a stock software 0033 .cn that said since the introduction of the concept of computing power leasing in, as you mentioned, mid 2023, the number of listed companies that have added the concept of computing power leasing has reached 114 as of a year ago.
And that number might even be bigger now.
And there are also many small enterprises in the industry besides these big giant players.
So we're seeing a lot more companies getting involved now.
Yeah. And if they're not already incorporating the service into their company plan, they're doing it in very soon time.
That's what we mean by adding this concept into their company strategies.
So yeah. So how do you start this business?
I read that there are three ways.
First of all is how we mentioned it earlier.
So this company who has the power and the strong financial and technological capabilities, they can choose to build their own or cooperate to build a computing power center themselves and then rent that out.
And the second way, which is a bit easier, is less costly, is that the company can utilize the existing cloud service platforms to expand business and provide customers with more convenient and efficient computing power services.
And thirdly, there are also some companies that were not originally in the computing industry.
And for them, they can rely on their advantages in certain fields to differentiate computing services across borders.
So it's not like they have to be the centralized master of this huge advanced technological hub themselves, but that they can just be specialized in one particular area.
Right. Yeah. The question as to whether this is profitable or not is an interesting one.
There's an industry insider engaged in computing power leasing businesses in Shenzhen.
And he or she said last year that when computing power resources were the most scarce, the prices of the computing power leasing market continued to rise.
But it took two to three years to recover hardware costs.
Now we're seeing changes in supply.
We're also seeing changes in demand.
And when we see more players enter the game, and we see this in other industries as well.
Think about it. If there's one restaurant on a street for an entire apartment complex, that restaurant is going to do really well.
But if a second and third and fourth restaurant opened up, well, then you're going to see some price competition there.
And that's exactly what's happened here.
They've seen some prices falling since the beginning of this year.
Nvidia, for example, the U .S.
chipmaker, their core products reached peak prices last year.
And in October this year, they saw some prices drop.
So we're seeing now with more players in the game that it's becoming more competitive and we're seeing prices go down.
But I mean, the future is still looking bright on that side, even though there are more competitors.
I mean, more companies joining this market now.
So this is according to a staff member from a domestic tech company who said that their company's total leasing business has reached a hundred and seventy five million yuan upon the time that this guy was interviewed.
So that's roughly twenty four million U .S.
dollars with a contract period of five years and a gross profit margin of around 20 percent.
So if there are long term leasing customers, then the current computing power leasing market has great profit potential in the future.
And doesn't this depend heavily on the future of AI application in the country, because you're essentially talking about, well, supply and demand.
And you've got we've talked about the supply side of these companies getting their hands dirty in this arena.
And then you've got to have enough renters who have this demand of massive amounts of computing demand.
Yeah, but it seems like and I mean, the evidence, the numbers we just talked about, the growth year on year is impressive.
So there's evidence there.
But we see evidence around us all the time and in the news and on Roundtable, we talk about training AI models and the number of companies that are using AI now.
We've even gone so far as to talk about how much AI should be used in the workplace and things like that.
So AI is one example.
But what about things like creating 3D animation now?
Animation. AI animation now is insane.
The quality is almost human like.
So you need a lot of really powerful computers to do that.
And again, these companies either don't need it on a daily basis, that level of power, or they don't have the money themselves, or they just don't want to invest their company money in that way when they have the option to rent.
It seems like, and there's no way perhaps to predict the future, but it seems like there will be enough, a big enough market in place for companies like this to continue to do quite well.
And also the better configuration and also AI modules are adding to that, say, power or computing standard, you're trying to rent to other people, the higher the price is going to get.
And because better computing configurations, including not only AI, but also high performing GPU, CPU, and other kinds of software softwares can bring higher computing efficiency, of course, and AI is like, like Steve mentioned, among many other softwares that's, you know, assisting on that as well.
So the more buff you're adding to this server, the more expensive is getting as well.
And I'm not sure if that's all in all so friendly to smaller businesses who's just trying to rent a little bit of computing power to do their product, let's say.
But this is a great thing on the other side of the coin, if you can see it, for the smaller businesses, because otherwise they can't even, they don't even have the ticket, they don't even have sort of the horse and the spear to go into the battlefield.
But now if you can pay some money to rent this computing power, then at least you have a way to train your tiny AI model or whatever it is.
Otherwise you can get the chance.
Yeah, exactly. So that's a really great point because it allows a lot more players into the game that otherwise wouldn't have had opportunity to play.
That was a perfect example.
In some examples here, I see that well, even though this company that I'm reading about, Jiangsu Little Electronic, even though it's a domestic company, it's also doing its overseas business.
So domestically it's mainly operating through leasing its own GPU server computing power or say subleasing the computer power it rents in.
So what this company is doing is that they rent the, sorry, they get the computer power from other bigger companies and they sublease it.
Sublet. Yeah, yeah, yeah.
To other smaller companies.
It's like a third party upon a third party.
So you're like a middle man in the...
In a sense, yeah. I don't understand.
What about the smallest company or the customer in the downstream?
Is there some kind of information asymmetry here why can't you just circumvent the middle man and go straight to the provider of the computing power?
I mean, they need to listen to Roundtable.
I'm sure they have all kinds of sources for information, but you can see also you have to just give it to these entrepreneurs in China.
They're very smart when they see an opportunity, they'll go in and they'll make a quick buck there.
And look at these giant companies that we just talked about as examples.
They see an opportunity here.
They see growth here and they're getting involved as well.
Oh, definitely. And we're also seeing that this has been touted as a major opportunity in the tech world.
This is seen as new territory for growth.
We've talked about how these companies are getting involved.
And could you also give us an overall picture of how fast developing computing power has turned out to be in China?
And bear in mind, it's only a year and a half or just a little bit more than that.
But look at this new territory.
I mean, so far, even though it's just been a year and a half, this whole business or this whole industry is strongly supported by governmental policy as well.
And this is according to a 2023 Xinhua report that Chinese authorities unveiled an action plan for the high quality development of computing power infrastructure.
So according to this plan, China aims to achieve a total computing power of over 300 F -lopes by 2025.
So this term F -lopes is a unit of measurement for a computer's speed.
So one F -lopes computing system can complete one quantillion floating point operators per second.
Oh my gosh, this whole sentence is filled with new vocabularies.
Really powerful computers.
Really fast. Very fast computers.
Yeah. So by August 2023, the Ministry of Industry and Information Technology said that China's computing power had reached 197 F -lopes, ranking second globally.
And the nation is making steady progress as well in achieving this goal that, well, according to Zhao Zhuguo, spokesperson of the Ministry of Industry and Information Technology, this October, China's computing power has reached 246 F -lopes.
So almost a hundred just in a year's time.
Second globally. Pretty impressive.
And also there's another aspect to this because if you're going to have these computer centers, if you're going to have this type of industry, well, then you also need a supply chain.
We talked about the supply chain expo on Roundtable, didn't we?
And we learned a lot of wonderful things there, but that exists here too.
And the computer power, computing power industry chain in China, it's formed now.
There's upstream suppliers, including chip manufacturers and server manufacturers and other computing power resource suppliers.
There's midstream providers of computing power, leasing service providers downstream, covering various application fields like finance and energy and industry.
There's more. What I guess I'm trying to say is there's more to this than simply building a building where you put a lot of powerful computers in there.
Those computers have a lot of parts.
Those need to be supplied.
Those come from suppliers.
It goes all the way down to raw materials and that supply chain is formed now too.
Yes. And it's perhaps not news for those who follow this topic that it's been difficult for China to get those super advanced chips to import them.
So therefore there's this very real and practical need to come up with better chips and to develop the chips industry.
And we're seeing that underway here in China.
And we're seeing that for, like you said, Steve, from beginning to end, this is a supply chain issue as well.
But it's also where the challenge shows up and that underlies an opportunity.
And then you're seeing that this could become a drive for the digital economy and also talking about the infrastructure and the manufacturing side of the digital economy.
And this is also a crucial part of the discussion of computing power in this country.
Yeah. And just because you know, we've seen the profitability and the opportunity to be successful from various companies, but it doesn't mean that if you get involved, you are guaranteed to be successful.
You know, a lot of companies have done really well so far, but there are companies that choose to exit from the industry because they don't have the ability to purchase these required high -end artificial intelligence chips or bear the high cost of building their own data centers.
So it is competitive.
And just because you get involved, it doesn't mean you're going to make a ton of money.
Yeah. And of course, we've mentioned how fierce this market is getting because nowadays, many different companies, they're trying to get a share of this industry as well.
Whether it is the funds, the resources, or the channels, like there's not enough for everyone at this point, maybe that's why there's so fierce of competition that some enterprises are facing profitability difficulties.
So for these enterprises, they need to find a balance between market opportunities and risks as well.
And it's really fascinating for me to discover that, you know, some companies, they were originally not from the computing or say computer manufacturing or any related sectors or industry, but they're still trying to be a player as well.
I see that some companies from furniture or toy industry, they're just trying to get a share by hoarding some GPU chips.
And through that, they have sort of like their own hardcore say hub they built for themselves, even though their business is not really related to computing at all, they're still landing.
Well, not landing like leasing their CPU computing power to other companies as a way of, you know, profiting for the company.
They don't need to worry about hardware maintenance or software services or things like that.
Not at all. And this just shows that it's a reflection of the economy we are living in right now.
Nowadays, even if you're a toy maker originally, but if you can hoard, we talked about hoarding friends yesterday, but if you can hoard chips and the really advanced ones, then that is you sitting on a gold mine.
And, you know, remember 100 years ago, if you were sitting on oil, then that was, you know, the gold mine for the economy.
But now because, you know, we're living in an era of the internet and with AI developing so fast, and like you guys have said, you know, with machine learning and big data analytics and so many other aspects of the digital economy, then yeah, you got to be really forward thinking.
Sometimes it can go wrong too.
Here's one example.
In September, 2023, a subsidiary of Lianhua Holdings.
So as far as I know, this company is selling monosodium glutamate, which is a seasoning or flavoring ingredient for cooking.
So this company, Lianhua Holdings, they purchased 330 GPU servers, thinking about hoarding it and renting it to others.
But at the end of October, 2024, so almost one year after they purchased it, only 12 servers had been delivered and the remaining 318 servers had not been delivered yet.
So it's like they're trying to play the cards and trying to minimize the cost for getting into this business.
But in the meantime, there are risks such like this.
Right. So what about looking into the future?
What needs to happen if we're talking about future high quality development?
Same things we talk about when we talk about other industries and how they can move forward into the future.
You need increased efforts and talent cultivation and introduction.
You need to enhance the overall technological level and innovation capability of the industry as well.
The experts within the industry say that to promote the healthy development of these types of computing power leasing businesses, more support is needed in terms of policies and other aspects too.
On one hand, you have to promote cooperation and collaboration between the upstream and the downstream of the chain, forming a good ecology, I guess you could put it that way, where everyone comes out with more advantages than disadvantages.
And also, I think we need to increase efforts in cultivating more talents as well.
You can't just have a toy company thinking that it works and then hoard the chips, right?
More people within this arena should, you know, be trained to know better.
Right. So you know what your college major could be if you're looking to get into college this year.
And for many businesses, this is an exciting opportunity to tap into the power of AI and high performance computing.
But as the market matures, the challenge will be ensuring that these technological advances are accessible, sustainable and secure for all.
