Companies are Investing $1.5 Trillion in IT/AI Infra in the Next 3-4 Years
The preconditions for some sort of economic boost are already in place, but the markets seem to have picked up on that right now.
On Episode 705 of The Core Report, financial journalist Govindraj Ethiraj talks to Mario Morales, GVP & General Manager, Semiconductors & Enabling Technologies at IDC.
SHOW NOTES
(00:00) Stories of the Day
(00:41) The Markets Jump
(02:44) Tech Results Are In
(04:34) Semiconductors and AI
NOTE: This transcript contains the host's monologue and includes interview transcripts by a machine. Human eyes have gone through the script but there might still be errors in some of the text, so please refer to the audio in case you need to clarify any part. If you want to get in touch regarding any feedback, you can drop us a message on [email protected].
—
Good morning, it's Friday, the 17th of October, and this is Govindraj Ethiraj, broadcasting and streaming weekdays, usually from Mumbai, India's financial capital, but presently still in transit. So, we have a short capsule and a longish conversation on semiconductors and artificial intelligence, which I found most educational and insightful, and I think you will too, so stay tuned right till the end. And let me jump right in.
The Markets Jump
The preconditions for some sort of economic boost are already in place, but the markets seem to have picked up on that right now. Bear in mind that nothing has changed on the US-India trade tariff war.
If anything, matters are as tense as they were earlier, with India just rebutting a claim by US President Donald Trump that Prime Minister Narendra Modi had promised he would stop buying Russian oil, one reason why we have been slapped with an additional 25% tariff on top of the 25% tariff for exports into the United States. Now, we of course don't know much about what discussions are going on at the back, but this is the public face of it, and Trump tends to go all out in public. On the other hand, foreign institutional investor buying has stepped up.
India's benchmark indices have risen to more than three-month highs, thanks to stocks like Nestle and Axis Bank, and strong earnings and earnings outlook, including from financials. Domestic inflows also continue to be strong, though gold and silver are continuing to attract attention. On Thursday, the Sensex and Nifty were both up, gaining for the second consecutive session, and at close, the Sensex was up 862 points to 83,467, and the Nifty 50 was up 261 points to 25,585.
The top stocks that gained were M&M, Kotak Mahindra Bank, Axis Bank, as we mentioned, Titan, this is on the Bombay Stock Exchange. In the broader markets, the Nifty mid-cap 100 and small-cap 100 were up 0.46 and 0.24%, respectively, according to Business Standard. The rupee touched a one-month high on Thursday.
There is some link between the rupee's trend and the market's current bullishness. One reason the rupee touched a high or has been on a stronger wicket is because of Reserve Bank of India intervention, particularly in the day before session. The rupee rose to a peak of Rs 87.70 in early trading before closing at Rs 87.82, up 0.3% for the day, according to Reuters.
And finally, as expected, gold prices hit a record high as investors continue to move towards safe-haven assets. It's still above $4,200 per ounce and now gold is up 61% year-to-date.
Tech Results Are In
Infosys reported a 13% increase in consolidated net profits, hitting about 7,300 crores for the second quarter of the financial year, compared to about 6,500 crores last quarter. The company also declared an interim dividend of Rs 23 per share. During the quarter, under review, revenue from operations increased by 8.6% to Rs 44,490, according to an exchange filing.
Attrition rates were up for July-September to about 14.3%, from about 12.9% year-on-year. Infosys management said that their proactive investments over the last three years in embracing an AI-first culture within Infosys have ensured that their people are re-skilled to thrive in a human-plus-AI workplace. And Wipro also reported stronger-than-expected second-quarter revenue on Thursday, driven by strong growth in Asia and America's communication businesses, and has also forecast revenue for the current quarter to be in line with some analysts' estimates, according to Reuters, which added that India's fourth-largest IT services company, Wipro, saw a 1.8% rise year-on-year in consolidated revenue to about Rs 2,269 crores.
Wipro also reported stronger-than-expected second-quarter revenue, driven by strong growth in Asia and America's communication businesses, and forecast revenue for the current quarter to be in line with some analysts' estimates. Remember, forecast and outlook is what most analysts are looking for when it comes to IT services companies. Reuters also added that India's fourth-largest IT services company posted a 1.8% rise year-on-year in consolidated revenue to about Rs 22,000 crores, or about Rs 2.5 billion, for the July-September quarter.
And Wipro expects a 0.5% decline to 1.5% revenue growth for the third quarter.
Semiconductors and AI
“Of course, we are in an AI bubble” Former Intel CEO Pat Gelsinger said on CNBC earlier this week, an article in the New York Times by Jared Bernstein, chair of President Joe Biden's Council of Economic Advisors from 2023 to 2025, and Ryan Cummings, who served the Council as an economist from 2021 to 2023, said that it was time to call the third bubble of our century the AI bubble.
They all point to how investment in artificial intelligence has been so huge, with venture capitalists investing nearly $200 billion in this sector this year alone. Additionally, data centre investment has tripled since 2022, and together these investments are driving growth across the entire U.S. economy, pumping up the stock market, and generating increasingly eye-popping valuations of the technology firms driving the AI revolution, according to that New York Times article. On the other hand, Mario Morales of Technology Research from IDC, who you will hear more from in a minute, argues that enterprises globally are expected to spend $1.5 trillion in IT over the next three to four years as they race to build their own AI infrastructure, all of which will give the semiconductor industry a big push, which I felt was a good moment to go deeper into semiconductors themselves, the heart of the AI revolution, and understand how they stack up from the most ordinary ones to the sophisticated ones, and they're all traded almost like arms and ammunition now.
So who is making them, where, and what does supply and demand look like globally in the near future? I spoke to Mario Morales, GVP and General Manager, Semiconductors and Enabling Technologies at IDC, and I asked him first about where he saw the tech revolution powered by semiconductors heading. This is a fairly long conversation with Mario, who joined me from California's Bay Area. Do stay tuned.
INTERVIEW TRANSCRIPT
Mario Morales: First of all, just to give you some background, when you look at the semiconductor space, it's been very essential to the growth that we're seeing today in AI infrastructure. So the spending has been there now for a couple of years. We've seen it accelerate across the cloud service providers.
And now we're starting to see the investment by enterprises and then ultimately sovereign investments as well. So every major nation around the world looking to not be left behind are now beginning to plan to invest in the technology, in the infrastructure to build this out. And you have a company like OpenAI who is a fairly new company, but they've of course have grabbed a lot of the mindshare.
They're one of the first companies to really start beginning to look at the potential of gen AI. And they've brought in a very large base of consumers of their platform. And so they are now sort of at the front side of all of this infrastructure, but they can't do it alone.
As much valuation as they've gotten as a private company, they need partners to continue to build out what's happening. And this is what you're seeing today. It's been an important driving force for the semiconductor industry as a whole.
Govindraj Ethiraj: Right, so if you were to take a step back in a manner of speaking, how would you define or draw the landscape in terms of demand and supply of chips? How many chips does the market really need? How much is being produced?
Where is it being produced? What are the kinds of chips that are most useful and important in building the kind of AI infrastructure that we're talking about?
Mario Morales: So when you look at the semiconductor space, last year in 2024, the total industry was about $680 billion in size. So it's a pretty sizable market. This year, we're expecting the market to grow about 17.6%. So we're looking at an industry now that 's gonna be about 800 billion. And over the next couple of years, we're likely gonna see a trillion in terms of size of the overall semiconductor space, which is about two years earlier than consensus, right? A lot of that growth that you're seeing in this industry is really coming down to what we're seeing today in the data centre. So the data centre just three or four years ago represented about 15% of all of the demand for silicon.
But as we start closing at the end of the decade, you're gonna see over 40% of the semiconductor industry driven by the data centre and the silicon that you're seeing deployed in infrastructure. So it's a very significant movement that we're seeing and the data centre silicon is becoming a very critical part. So that includes things like GPUs from NVIDIA and AMD, but also ASICs from Broadcom and accelerators just across the board.
And so far, what we've seen is a lot of the demand has really been centred on training these very large frontier language models. And over time, this same training and reasoning and inferencing is gonna move from the data centre to the edge. And I think that that will bring in even more opportunity because despite what people think about the fact that you're gonna have to get more power efficient, that's one facet of what we're seeing as a trend.
But we also believe that the more you're doing inferencing at the end points, the more computing you're gonna need locally on these devices, whether it's a smartphone or a PC or a wearable. And so we're seeing all of this equipment, all of this hardware being deployed and that's driving tremendous growth and recovery for the semiconductor industry. Right, so you said 40% is data centres today.
Now, 40% will be the data centre market by 2030. Four years ago, it was about 15%. Today, I would say it's about 24, 25%.
So we're seeing some very rapid growth and data centres are an important driver for the growth in our industry.
Govindraj Ethiraj: And data centres would typically need or use the highest end of graphic processing units and so on?
Mario Morales: Yes, it uses the most dense hardware and many GPUs and CPUs and also memory are very key. And more so now as you're building larger and larger systems or GPU clusters, you're also gonna need more advanced communications and connectivity. So you're also seeing an explosion of demand for networking topologies and technology to support these very large AI servers and data centres.
Okay, where's the other 75% going today? Well, a lot of the other markets that support the semiconductor industry continue to be traditional spaces like PCs and smartphones. You look at the automotive industry, the industrial industry and also consumer electronics.
So think of the smartphone, think of white goods appliances, all of those devices are consuming a lot more silicon over time. These industries will continue to grow over time because we're seeing silicon in general become more pervasive. And in areas like automotive and industrial, what we're seeing is a movement to more electrification and less motors and so forth and analogue.
And I think that's what's driving some of the growth in those industries as well.
Govindraj Ethiraj: Right, you said 680 billion in the beginning. So if you were to now split the value between the high performance chips that would be used in data centres or for the kind of AI applications that we've just talked about versus the more conventional, which is let's say in appliances or cars and so on, what would that be?
Mario Morales: Well, if you look at the most advanced technology that's being used in this space, a lot of it is going to the data centre. So that would account for one fourth of the market today. The other three fourths are gonna be made up of the markets that I just mentioned earlier.
If you're looking at it from a manufacturing standpoint and you look at all the capacity in the industry as a whole, only 20% of the capacity is allocated to advanced technology nodes. So think of TSMC and Samsung and Intel offering technologies that are seven nanometres and below. And today we're just beginning to see three nanometres in mainstream use moving to two nanometres later this year and going into next year.
All of that advanced technology accounts for about 20% of all the capacity. The other 80% is still being allocated primarily to mainstream and mature process technologies that today you can find in any device.
Govindraj Ethiraj: Right. And I'm gonna come back to applications in a moment. Walk us through where these chips are now being manufactured predominantly and how do they now flow?
As in where do they go? What are the borders they cross and so on?
Mario Morales: So traditionally the semiconductor industry has been broken up between integrated device manufacturers, companies like Intel and Samsung. And then you have this collection of leading companies that are doing, we call fabless companies. So they design chips, but they use a manufacturer to build them.
And then there's the foundry industry as a whole. These are manufacturing service companies like TSMC, UMC, Global Foundries, SMIC in China, Huaguang Group in China. These are companies basically selling services, manufacturing services, and they've become a bigger part of the industry as a whole.
So when you look at the most advanced technologies, those are primarily coming from Taiwan and to a certain degree, Korea. Later this year, Intel will be rolling out their 18A process in fab 22 in Arizona. So we will start seeing some advanced technology begin to be produced in America.
But today over 90% of the most advanced technology is coming from Southeast Asia and Korea and primarily from TSMC. So TSMC today controls about 70% of the total foundry industry, and they're driving and growing twice as fast as all the rest of their competition. Now, over time, what I see happening is that you're seeing both TSMC and Samsung continue to invest not only in their home country, but at the same time, they're diversifying into Japan, into Germany, and also into the U.S. And last year, TSMC began to produce four nanometres and five nanometres in the U.S. in Arizona. And you can expect that over the coming years, they're gonna have a lot more concentration of capacity in North America. And so that's just the advanced technologies. When you look at some of the mainstream and mature technologies, a lot of those technologies today are coming either from China or the U.S. in Europe, but mostly in Asia Pacific. So companies like UMC and Global Foundries and SMIC, they are the biggest players in those spaces, and they serve industries like automotive and industrial and consumer electronics. And over time, even TSMC is also a big player. They control about two thirds of the mature and mainstream process, even though they're the leader on the leading edge.
But it's an industry that's growing about 15 to 17%, and the foundry industry is what supports all the major fabulous companies that we talked about. So these announcements that NVIDIA and AMD have made and Broadcom is making, these are companies that are large, large customers of TSMC. And TSMC is an important ingredient company that is really enabling these companies to be able to address the growth that we're seeing in AI infrastructure.
Govindraj Ethiraj: Right, and in terms of both the end use and the destination for all these chips that are being manufactured, how would you split that up geographically today?
Mario Morales: Well, the demand side is a lot more evenly split, right? Today, China is still the largest market as a whole, and they control about a fourth of all the demand for semiconductors. It used to be closer to 30%, but because a lot of the growth is now coming from AI, we've seen a little bit of a down number for China, but over time, China will continue to lead the industry as being the largest semiconductor industry.
Second would be North America. So in the Americas, of course, PC demand has been pretty healthy, smartphone demand, but the build out of AI infrastructure is what we're seeing today in North America. And it's been basically a land grab.
You know, companies like OpenAI and Tropic and others are working very closely with the semiconductor companies to build out the hardware that over time is going to drive a lot of the services and drive the training and inferencing that we want to see across enterprises. If you look at other sectors like Japan and Southeast Asia and Europe, these industries are also heavily driven by automotive, industrial areas, and also consumer electronics.
Govindraj Ethiraj: Right, and if I were to come to the enterprise side of the market now, companies are also heavily investing in AI solutions, agentic AI, we hear a lot of that even here. All of that is obviously calling for more computing capability and computing power. So where is that coming from today?
Where could it come from down the line?
Mario Morales: Well, when you look at the enterprises, I think they're just beginning now to experiment with AI. And what's key about this is that AI needs data to be able to train the models, to be able to drive inferencing. And so a lot of that data is still captured by most enterprises.
And so when you think about all the training data that CHAT-GPT has used, or GWEN in China, or all these different large language models, a lot of that data is coming from public domains and the internet. 80% of the data is still locked in with enterprises. So what we're gonna see is basically companies now are recognizing this, and they're trying to figure out how to monetise the value of the data that I've been capturing for my business and my operations.
And so you're starting to see enterprises begin to experiment with AI, looking at ways to improve how they connect closer with customers, how do they collaborate better, but also how to automate a lot more, and leveraging data all along those different paths. And so IDC believes that in the next three to four years, enterprises will spend about $1.5 trillion on IT, hardware, software, and services. But of that number, about $325 billion will be allocated towards building out AI platforms and building out hybrid clouds, because what they're recognising now is that they'll still continue to use cloud infrastructure from the Amazons and Microsofts and Googles of the world, but they also wanna own their own infrastructure.
And so this is the deployment and the growth that we're expecting to see in the coming years, and hopefully we'll continue to see the strength that we've seen in the market, and it'll have a positive impact on the semiconductor industry as a whole.
Govindraj Ethiraj: Right, so you're saying that, is this the first time that we're seeing a reversal of that trend in some ways? I mean, compute going from within the enterprise out and now coming back?
Mario Morales: I think you're gonna see a combination of two. I think we're gonna live in a more hybrid world where the cloud is gonna have its position, but so will the edge. And I think that when you think about a consumer, the consumer will want to use more AI, but on their devices.
They wanna make sure that they can trust their devices, that they have the proper security, and you're gonna see more inferencing and reasoning done. If you think about what's driving ChatGPT now and what's driving the new Gemini version from Google, a lot of it is consumers now beginning to implement that across their devices. And so I think that you're gonna see a hybrid world and computing will sit not just in the cloud, I think that's where you're gonna do the training for the model sets, not gonna move away from that, but more of the inferencing and the reasoning and the learning over time will be done with edge infrastructure and also with endpoint.
So our devices are gonna have a lot more compute, they're gonna have a lot more memory, and also newer technologies like newer processing engines in order to be able to do AI in a more performance efficient way. And I think that's where we're seeing the world move to.
Govindraj Ethiraj: If I can contextualise this through the sort of eyes of Wall Street, now almost 25% of the market is really the tech companies, which in turn is linked to AI, which in turn is linked to chips in many ways, though not always. How do you see this spanning on from here? Because on the stock side, they clearly, I mean, it is at record highs, whether it's a frenzy or not, I guess is subjective.
But how are you seeing it from your point of view? Because if things turn for whatever reason, then that will affect capital investment, that will affect outlook and really the way businesses are planning down the line.
Mario Morales: IDC does not do valuations of companies, but we do advise financial institutions more on the technology itself and the companies that we track very closely, the management team. So what I'm seeing still is a large amount of commitment to continue to invest until we get to that point where we get to general AI. I still think we're a couple of years away from that, but all of these companies are racing towards that.
So the only thing that will slow that kind of investment down will be the uncertainty around the policies, around trade between the US and China. And the more we see industries lock up and become more insular, I think those are the risks that I think investors need to be concerned about. If you look at the technology itself, it continues to progress.
And in fact, whether it's NVIDIA or AMD, these companies are now on an annual cadence in terms of the introduction of the technology that they serve to the market. And so this market will continue to evolve and it's moving very quickly. And I still feel that we're early in this level of investment.
We still have not yet figured out how to create a very sound TCO or justify the business value across this, but you need to build the infrastructure, just like we built railroad infrastructure in the early 1900s, you're gonna have to do the same for AI. And I think it's gonna be very transformative across many vertical industries. So I think we're at the beginning and I think investors are kind of seeing that, but I think there's risks, there's risks around the trade, there's risks around negotiations around trade and the administration in the US still has some uncertainty around the policies that they're trying to implement.
For me, I've always been a believer that the less governments are involved in the industries, the more the industries can continue to drive innovation on their own. You need the public sector more for R&D. And I think we're at a phase now where we're moving beyond R & R&D and into commercialisation.
So I'm more optimistic about it. Of course, you could have some pausing effect in the coming years, I think over the next year and a half or so, you might see a pause by some of the cloud service providers, but at that point in time is where I believe also the enterprises will begin taking off with their own commitments to invest in this space and also the government sector now. So you're seeing projects like Stargate, maybe not as sizable as that, but you'll see them in China, you'll see these kinds of projects in Japan, in Korea, and of course in Europe.
And so no one wants to be left behind in AI. And I think this is the land grab that everybody's looking at. And I think investors should be paying close attention to that.
Govindraj Ethiraj: Right. You talked about the role of Asia, including Southeast Asia, you talked about Samsung, TSMC, and China, of course. As you look ahead, if things do not improve in terms of trade relations, and countries like China are obviously incentivised to build their own chips because of trade barriers and so on on the US side, how do you see that situation evolving?
I mean, could we see new sources of semiconductors? Could innovation shift?
Mario Morales: Absolutely. That's inevitable. I think this industry has traditionally been very interconnected.
So, you know, you have suppliers working closely with the supply chain across Asia Pacific and vice versa. But I think if we continue to see this disconnect between the US and China, you're going to see new supply chains being formed. We're already seeing some manufacturing move to places like Malaysia and Vietnam and even Singapore, which is a high cost nation, but it's really in search of more talent and also potentially market second consume a lot of this technology.
So I think India will play an important role in the future. If India can continue to focus on building out the infrastructure and making that available. I think India has a very strong talent, especially over the years, the amount of work that they've done in the IT space.
I think some of that will pivot more towards trying to figure out how to impact the design of a lot of this AI hardware and also the applications that will sit on top of the hardware. I think in general, companies will look to find better ways to move up the value stack. And we're seeing it today with semiconductor companies.
And what's interesting to point out about that is the fact that for the first time in a very long time that I can remember, and I've been tracking this industry for over 30 years, the semiconductor companies are enjoying gross margins that are very similar to software companies. And if you look closely at semiconductor IP companies like Arm and Rambus and others, they're still generating gross margins of over 90%. And so this is sort of unheard of, but a lot of it is because we're moving to an era where it's not just about chips and packaging, but it's about designing systems within systems.
And so the level of knowledge and capability you need to have is gonna only escalate in terms of complexity. And so you're gonna need the talent pool to be able to take advantage of that. And so I see supply chains will evolve and you're gonna see nations that you didn't think could collaborate will begin to collaborate more and in some ways pull away from the U.S. because they see a bigger opportunity in close collaboration with China and with Southeast Asia. And so Japan will have to make those kinds of decisions, Korea will too, and even the European Union will have to do that as well. I do definitely see that evolution will happen and you're gonna see new players come to market because of it.
Govindraj Ethiraj: We saw the U.S. government taking a stake in Intel, something that one could have never imagined, I guess, and Intel itself facing a lot of challenges. That too was, I guess, not really anticipated a few years ago. What does that tell us?
I mean, in terms of what has happened and what could happen in the semiconductor industry, just to wrap up.
Mario Morales: It tells us quite a few things and it's a long story that's gonna be told for many years, but one of the things that's key is that, it's not new that governments take vested interest in certain industries. We've seen it in oil and gas, we've seen it in building out infrastructure. And even if you look at the early days of companies like TSMC or Samsung, there's been heavy government involvement from the beginning.
That's why they've subsidised and they've been able to take advantage of tax incentives, but there's a point in time where you have to kind of stop at a certain point and allow the company to fend for themselves, to survive on their own. And for me, that's the line that I think I'm watching very closely between the U.S. administration and Intel. I think Intel has a tremendous amount of IP and technology, but a lot of it has been missteps around execution and quite frankly, leadership.
A lack of leadership that made decisions many years ago that are affecting the company today. So I do believe that the industry as a whole is coming together and it does require more intersection with the public sector, because a lot of this buildup is around infrastructure that's gonna benefit the public services that you're gonna see. So you're gonna see that combination.
And I think that the government will benefit from keeping Intel around until they're able to kind of transform and get back on par with their competition, but it's just gonna take some time. And the unfortunate truth is that our industry, whether it's AI or the semiconductor industry, is changing so fast that some companies will not be able to survive. And in fact, one of the key things that could have predicted some of this is every time you see a very big shift in architecture is when you start seeing leaders be toppled and eventually you see others emerge.
And so that's happened with companies in the past like IBM, it happened to a certain degree with Apple and then some of these companies were able to bounce back. And I think Intel has a lot of interesting technology that they're gonna be able to leverage and they need focus and they need to make tough choices in order to survive and thrive in the future. But it's an interesting time and it tells us that as big or small a company might be, or even a government or country, you can't do this alone.
It requires a tremendous amount of investment. It requires a lot of technology and also leadership. And I think this is gonna be the most important thing across any industry is gonna be who are gonna be the leaders that are gonna take us forward with some of these interesting and incredible technologies that are in front of us today.
Govindraj Ethiraj: Mario, it's been a pleasure speaking with you. Thank you so much for joining us and sharing your thoughts.
Mario Morales: Thank you very much.
