- 10 hours ago
Investment in artificial intelligence is accelerating rapidly, with billions being poured into data centres and advanced AI models. But as spending rises, one question is gaining attention: how soon will investors begin to see meaningful returns?
Category
🗞
NewsTranscript
00:00Thank you very much with this very fantastic panel on a subject that is very, very important.
00:06Hundreds of billions of dollars are being invested around the world on AI. India too,
00:13just 24 hours ago, one of India's largest companies, the Arani Group has said $100 billion
00:18in data centers. Previously, $75 billion, a large amount of that going to the state of
00:25Andhra Pradesh, again in this space. Individual startup companies, I don't know how many of you
00:30managed to get inside the AI summit, I went there, young founders, this company that many of us
00:36hadn't even heard about, straight away, billion dollars being raised, billion dollars plus, all of
00:41this pretty much in the last 24 hours. But the critical question, ROI, return on investment.
00:49For a student, ROI is the numbers that he will get from an exam. For a business, it is capital,
00:56deployed, and the returns that are paid. Set this ROI question in context, Mr. Rangaswamy,
01:02why is it fundamental to this debate that centers around AI and the investments that we are seeing
01:08in it?
01:09Thank you, Siddharth. Again, the key question here is, like you mentioned, hundreds of billions
01:14of dollars are going into data centers, into investing in LLMs, and the reward has been promised
01:21to come in the future. The question is, how long can investors wait for the rewards to
01:28be recognized? A trillion dollars is going to be invested in data centers. That means
01:33the revenue of those companies have to be larger than a trillion dollars. And if you look
01:39at the entire revenue generated by hyperscalers and the startups, it's less than, I don't know,
01:4625, 50 billion maybe, being generous. So when is it going to reach that trillion dollar threshold
01:51to pay back the investors? And that is the key question. And I think there's so much euphoria
01:58here. I think there's definitely a bubble here. But good things can happen. The question is,
02:04how patient will investors be? How patients will the stock market be? And that's why ROI
02:10is so significant, because people are going to use AI if they see a return on investment.
02:16And that's what I hope we can have a good discussion here on.
02:18That's a very, very important point that you made. And as I come to you, Jonathan, explain
02:26to our audience here, your perspective on ROI. Is there so much capital awash in this world,
02:34liquidity, that it's just going for the next thing without looking at what the return would
02:41look like?
02:42Thank you, Siddharth. So I have an interesting vantage point, because my company works with
02:48all the frontier AI labs, like OpenAI, Anthropic, DeepMindMeta. So we see the investment up front.
02:54And I would say this year has been an important year where AI has shifted from passing tests
03:06to doing real work. I think the ROI question maybe misses a bigger point, which is, if you
03:14look at digital knowledge work today, like across every industry, financial services, life
03:19sciences, retail, healthcare, every function in that industry, like software engineering, sales,
03:23marketing, every role in the org chart, every workflow that a human goes through, that's $30
03:29trillion of digital knowledge work. That's the work that a human does in front of a computer
03:33and uses a keyboard and a mouse to get tasks done.
03:36And if you think about robotics, you can scale that up to $110 trillion. So today, with artificial
03:46superintelligence, we are reaching a point where these computer use agents can now do tasks
03:52that can span days. Pretty soon, it'll be tasks that can span weeks. So it's going to impact
03:59every single job on the planet. And almost all the problems that humanity faces are intelligence
04:06constrained. And what we have here is intelligence as an API. If you had abundant intelligence
04:12as an API, you could cure diseases, you could go to Mars, you could extend lifespans. So there's
04:18so much potential here. And unlike with self-driving cars, which was sort of a hard failure problem
04:26for AI, where you have to get 100% accuracy or 99.9999 accuracy to deliver value. Here,
04:33these are soft failure problems. So even with today's models, there is so much value that
04:38you can unlock. Because many of these systems are human-in-the-loop systems.
04:43Right.
04:43And when we work with enterprises, we see that the work of a human has changed from doing
04:50the work to verifying the work of agents that are doing the work. I think we're about to
04:55enter a golden age of productivity. It's going to be incredible to be a human knowledge worker.
05:00Okay. Golden age of productivity. Anish, as I come to you, I got the sense from his answer
05:06that should we be even asking this question, right? If you're in the phase of discovery,
05:11when the wheel was invented, nobody asked the ROI question. Look where it has brought the
05:15world. So is that something that you would want to answer?
05:20Yeah. Let me just, a little context. I'm going to speak to this answer as a former policymaker
05:26in the US. And I want to just highlight the dynamics. Three layers to this question. One,
05:31we had a chapter of this movie earlier when we did the digitization revolution. And we spent
05:38considerable sums of dollars to put dark fiber throughout our respective economies. And at the
05:44time, the ROI question was very risky because the utilization rate of the fiber when it was initially
05:49deployed was much lower. In today's infrastructure investment analogy, I think every GPU is accounted
05:56for. So the utilization rate of the infrastructure already exceeds what we saw when we had this prior
06:01discussion about bubble or not bubble at the infrastructure layer. Number two, when we launched
06:06Startup America in the United States, one of the important policy objectives was to say all the job
06:11creation in the US going back a decade came from companies less than five years old. MR has been an
06:17entrepreneur his whole life. Entrepreneurship has been driving economic growth in the US story.
06:21But what often happens is the rules of the road of a particular sector change. And it's the
06:28entrepreneur that races to the future. That's an open question in ROI. As we get to the total
06:33addressable market of labor in various sectors of the economy, we don't know if we have the rules
06:44that. But do you even need those rules? If you were again, you were the first CTO of the US
06:53government
06:53during Barack Obama's presidency. But if you were again a CTO of the US government, would you even
06:59advocate the need for rules? MR. Let me explain. These are different rules. What I'm describing here are market enablement.
07:05So this Trump administration in the United States context,
07:08has just created new rules to create essentially a more AI powered healthcare delivery system. So by changing the way
07:16you reimburse, instead of it being a human's documented time spent treating the patient to something that's more about outcomes,
07:22you can unleash this AI revolution where you've got an army of AI agents supporting a doctor. You sometimes need
07:29enabling policies to unlock ROI. And my last comment, and I think this is another area of ROI, this is
07:35an area, we're coming into this age of abundance. So we have to start asking
07:38more of a societal and multi-stakeholder question of ROI. There will be capital deployed and earned. But there will
07:44also be questions, what is the return on getting everybody educated at a gold standard? What is the help of
07:48accessing universal high-quality healthcare?
07:51And these are the three frameworks that I believe will make the ROI question central for the next decade.
07:55Mr. Rangaswami, hundreds of billions of dollars were blown up in the dot-com bubble. The financial meltdown, probably trillion
08:04and more. If AI were to solve the problem,
08:08if AI calls long-pending medical issues or new drug discovery 10 months instead of 10 years, then why should
08:16we even be bothered about asking that ROI question? This is so fundamental to humanity.
08:22I agree, but it's very altruistic, right? At the end of the day, an investor wants to see returns, whether
08:28Jonathan or Anish say it's long-term going to be gainful for the economies, right?
08:34So there's short-term, long-term. So somewhere there has to be some kind of convergence where short-term investors
08:41go, we'll be medium-term investors. And the long-term people like Jonathan and Anish go, no, we'll come to
08:47the medium-term.
08:48So I think we need to move the horizon a little closer. That's what I'm advocating for is refined objectives,
08:56refined ROI. That kind of addresses both the long-term needs of society. And I agree with you. There's 30,
09:0250, 100 trillion dollars of market space and TAM, as you call it.
09:07But how quickly can we achieve it? And can we pay back these investors? Or like the dot-com bubble?
09:13I'm not saying the dot-com bubble is bad. Because out of the dot-com bubble, there was 100 billion
09:18spent by private capital. And out of that came companies like Google and all these companies that are now worth
09:24trillions of dollars.
09:25The new empires of today were built out of the rubble of that era.
09:28Exactly. And so the same thing can happen again. I'm just saying, let's bring some coalescence around...
09:35So let me interject here, because we are here on the sidelines of the India AI Impact Summit, and the
09:42speaker before said it's the first one in the Global South.
09:46What is the path for India? Because this is a debate within government circles. How much money should the government
09:53commit?
09:53We have committed some money, but if you look at the size and scale of the Indian central government budget,
09:58it is not too much.
09:59What is the right path for a country like India, to you first and then to both of you?
10:03So I would say, let's look at the dot-com bubble again. Most of the money that went into the
10:10dot-com bubble came from private investments.
10:12There were no big write-offs by government when the fiber and all this thing happened and they weren't utilized.
10:18I think the similar thing is happening here as well. I don't see governments putting that much money.
10:24It's data centers being built by sovereign wealth funds. It's being built by investors.
10:29So if those data centers go dark, the guys are going to take the loss. It's not a government.
10:34It's going to be mainly private capital. So that's my take on it.
10:38I think governments should invest some money, but they're not going to be the primary investor in the AI boom.
10:44Okay. So I have three comments to make.
10:49Firstly, I think India should invest in sovereign AI.
10:53And second, there are big implications for jobs and education.
10:57So I'll come through all that.
10:59So when I say sovereign AI, I mean, I think it's important for India to build its own foundation model
11:07that's fine-tuned by Indian citizens, Indian nationals, at least in post-training, where the...
11:15Does Sarvam fit the bill? Because it's being launched.
11:19I mean, their latest work is being launched as we speak.
11:22It could. I haven't dug into them.
11:25Okay.
11:25But I think the future will be some type of a hybrid architecture
11:29where a government like India will have her own sovereign AI model.
11:34That model could be fine-tuned on proprietary data that the Indian government has.
11:40You want proprietary intelligence that's in the minds of the humans
11:44that work in government in India to be distilled into the models.
11:47You want to teach the model how to do proprietary tool calls,
11:51but still give the model the ability to talk to other frontier models,
11:55perhaps a GPT-5 or a GROC-4 or a Gemini 3,
11:58so the model can orchestrate and collaborate with other models where needed.
12:02I think this will be important, like, to make the government more efficient,
12:08to better serve the citizens of India.
12:11It's important for national security.
12:13I think it's important for India to also be at the frontier of superintelligence.
12:19You have to be in control of your destiny,
12:21and a big part of it is being in control of the data that goes into the models.
12:25The algorithms are relatively diffused across the industry.
12:30Compute, again, is relatively diffused.
12:32You can invest.
12:33I think data is where you'd differentiate,
12:37so you'd want to make sure you invest there.
12:39Second, I worry deeply about job displacement
12:42because there is a tsunami coming with superintelligence.
12:46Like, every type of digital knowledge work
12:50that a human does in front of a computer is going to change.
12:53Not necessarily replaced, but it's going to change.
12:55Humans are going to switch from doing work to verifying the work of agents,
12:59so we have to train humans to help them adapt to this new way of working.
13:05And third, that comes to education.
13:08I think I worry even more about education
13:11because, like, there's, like, 12 years of humans sort of in the pipeline
13:14before they get into the job market,
13:16and these models are getting smart, like, quarter over quarter.
13:20I think we have to fundamentally rethink education
13:22for a post-superintelligence world.
13:24Okay, before I come to you, Anish,
13:26we had invited a lot of students.
13:28All I want from you here is,
13:30for those who think that AI is a worry to your future job prospects
13:35and that you need to retool,
13:37please raise your hands.
13:41Okay, so I hope you, they heard my question right,
13:45but not many seem to be,
13:46at least not the majority of this room seem to be worried,
13:49but your take on this.
13:50Let me just say at the outset,
13:52capital is flying where it's economic highest and best use,
13:56so I don't know if a government contribution in addition is necessary.
13:59I'm with MR on that.
14:00But let me just make a couple of observations.
14:02Almost 15 years ago,
14:04President Obama came to India,
14:07and we did an expo on what was at the time
14:10the beginning of the Adhar Revolution and DPI.
14:14And when President Obama observed all of this,
14:16he had one word to say to India, leapfrog.
14:21There was an opportunity to leapfrog into the next generation
14:24what this infrastructure would mean.
14:26And if you remember, in the COVID pandemic,
14:28in the U.S., we were still handing out paper cards
14:30to give people the vaccination records.
14:32In India, that was all digital, leapfrog.
14:35So I say what motivated me to the AI Impact Summit
14:38was the impact message,
14:40was that we're going to unlock the use cases
14:42that will drive the greater good.
14:44The question you posed was a worrying one.
14:47Are you worried?
14:48What if the entrepreneurial spirit of this country
14:50created the conditions to train people
14:52and how to be better at prompting AI tools
14:55to do their job even better?
14:56And this became the center of excellence
14:59for how to cross this chasm
15:01so that we get to a jobs abundance
15:03even if we're thinking about the tsunami on the other side.
15:06So impact in healthcare, impact in skills development,
15:09this is an opportunity.
15:10And I'm a disciple of C.K. Prahalad.
15:12His son was my roommate in school.
15:15And so I believe in fortune at the bottom of the pyramid.
15:17We can invent innovative models
15:19in the use of these new technologies
15:20in key sectors of the economy
15:22and then scale them globally.
15:24And the beneficiaries will be the world.
15:26Quick follow-up.
15:27Where do you think AI in the context of India
15:30will have the greatest impact?
15:32For example, healthcare sector that you are involved in.
15:35Health.
15:36Unquestionably, in one of the...
15:37In fact, the session we're having later this afternoon
15:39is on this exact issue.
15:41In order to understand the power of AI in health,
15:44one has to ask the question,
15:45how easy is it for someone to get access
15:47to a high-quality doctor today?
15:50And that access is difficult.
15:51So this is the ideal leapfrog moment.
15:54You need data access,
15:55you need compute,
15:57and you need models that are tuned
15:59to make sure that we're doing
16:00in the best interest of the patient.
16:01These are all opportunities.
16:03Jump ball.
16:03We can actually see that collaboration work.
16:05And I'm hopeful with ABDM and the initiatives here,
16:08combined with what we're doing in the U.S.
16:10with health tech ecosystem,
16:11there's a chance to collaborate in these areas
16:13as we did with the U.S.-India Open Government Dialogue.
16:16Mr. Rangoswami,
16:17and Jordan,
16:18before you come in,
16:19another show of hands,
16:21how many of you have used
16:22any form of AI in the past week?
16:26That's a cheat question.
16:27Of course the answer is...
16:28Okay, go ahead.
16:29Go ahead.
16:30See, that's the group.
16:31And the earlier speaker alluded to it, right?
16:34They're all using AI
16:35and generating a lot of data.
16:37And so there's some questions
16:38to be answered on ROI.
16:39Who owns that data?
16:41I don't know.
16:42Right now it looks like the models own the data, right?
16:46So I think there might be something
16:49regulations-wise and others
16:50to kind of provide consumers
16:53some benefit of using these systems.
16:55So there could be some ROI
16:57that's more than just making you more productive,
17:00giving you more abundance.
17:01Maybe there's also money to be made
17:04in using AI
17:05and training the data models to be better too.
17:09Yeah, go ahead.
17:10So a great comment to Ammar
17:12on who owns the data.
17:13Like, when we work with enterprises and governments,
17:16we see a strong desire for sovereign AI
17:19where I do think governments
17:21and enterprises should own their data.
17:23And there are ways to architect these systems
17:25within the compliance boundaries of an enterprise
17:28so the data doesn't leave the company.
17:31I do think a company's data is its moat.
17:35I don't think algorithms and compute will be.
17:37I think it will be data.
17:38And ideally a data-driven feedback loop
17:40where as humans use that company's services,
17:44the models keep getting better,
17:45the agents keep getting better,
17:47and it becomes more useful.
17:48And I really liked Anish's framing on leapfrogging, right?
17:52Like, I love that optimism.
17:54And in case the tsunami sounded a bit scary,
17:56you could surf the tsunami, right?
17:58And maybe all of us,
17:59I mean, you're all already using AI pretty heavily.
18:02I think we have an exciting path ahead.
18:05To your point, in addition to healthcare,
18:07for India, I'm extremely optimistic about education.
18:10I think there has been no better time to be a student, right?
18:15Like, so if you're studying something, right?
18:17Like, you have an infinitely patient tutor
18:20that can talk to you at your level of difficulty.
18:23In the same classroom, in the same classroom.
18:26This is a brilliant point.
18:27Imagine 50 of you students in a class,
18:30one teacher delivering the same message in the same tone
18:33and with, you know, in his own style.
18:36But imagine each of you getting a dedicated tutor,
18:39even if you are in a physical classroom.
18:41That's the world I believe we'll perhaps move to.
18:43That's exactly right.
18:44And I benefited greatly from my teachers in Chennai,
18:47where I grew up, and at Stanford.
18:49I still remember my awesome teachers.
18:50And now we can scale those teachers to,
18:53I mean, to the rest of the world.
18:56So we've built this thing called EDU Arena.
18:59So if any of you are interested in sort of playing with,
19:02like, an education-focused LLM product,
19:05you should check it out.
19:06But I think there's never been a better time
19:08to be a student or a teacher.
19:09A teacher can now create curriculums
19:12at, like, infinite dimensionality, right?
19:14It's such a wonderful time to learn
19:16and be a lifelong learner.
19:17Absolutely.
19:18You know, India, Spora, and you work,
19:21must have worked with tens or hundreds of Indian founders.
19:25I wanted to ask you this question.
19:28Frugal Indian engineering,
19:30Jugaad as it is called, you know,
19:31is there, you know,
19:33when we talk about huge amounts of capital needed,
19:35do you see somewhere this hope of Indian Jugaad
19:38doing the same thing at 110th, 120th, 150th of the cost
19:43as we did in our space program, for example?
19:45I agree.
19:46I think Anish alluded to this thing as well with India, right?
19:51Leapfrogging.
19:51So there's an ability to leapfrog,
19:54but also leapfrog with less money, right?
19:56I don't think India can afford
19:58to invest a trillion dollars in any of these things,
20:01but maybe using frugal innovation
20:03and leapfrogging as a concept,
20:05let's get from LLMs to whatever is next.
20:08You know, maybe it's its own proprietary data,
20:11maybe new models that are developed within India,
20:14taking advantage of all the Indic languages,
20:17taking advantage of the speech.
20:19So there are many, many ways I think
20:21India can leapfrog at lower cost
20:24and knowing our mindset and our DNA in India
20:27and with the diaspora, we'll come up with that.
20:30We're just at the beginning.
20:31We are at the beginning.
20:32Closing comments from both of you.
20:34You want to go first, Anish?
20:35No, no.
20:35I'm bullish on collaboration
20:37and I hope that this spirit of solve it here
20:40and scale it everywhere is going to be the mindset
20:42so we can see some real impact
20:44on societal issues like health and education.
20:47I would say for all the students in the audience,
20:50now is the time to work on super intelligence
20:52to drive real economic progress.
20:55Last year was AI doing tests,
20:57getting the gold medal in math.
20:59Now it's AI that can actually do real tasks.
21:02So I would encourage everybody
21:03to study computer science,
21:05learn how to build applications on top of LLMs
21:07and the future is bright.
21:09Okay, the future is bright.
21:11So if you thought that it takes tens of billions of dollars
21:14as it does for a large language model,
21:17the future can always allow the language of Jugaad,
21:21JLM, Jugaad language model to come forth from India.
21:24I thank the entire panel for their time with us
21:28for sharing those thoughts
21:29on that very, very fundamental question
21:30and you the audience.
21:32Thanks very much.
Comments