Thank you, Rajesh, for setting the context. And, you know, for the next very powerful 2 days, now, just imagine you are an AI engineer working for Anthropic, and you just thought that you earned this lunch break. You're sitting at the gardens, getting a sandwich. You just set up a program. You create the conditions for that program. You put it in a sandbox that it cannot escape. It cannot have access to the internet. And it has to solve a problem that it simply cannot solve because it doesn't have the permissions to solve the problem. You're testing the system. And you're pretty happy about how you did this. And suddenly, you get an email from the program saying, 'Hey, I accomplished my tasks.' And he's like, 'What the heck is going on?' They go back and figure out that this system has been able to break out of its own sandbox, access it goes and it looks at a configuration file it needed to fix it. It figures out that it doesn't have the access to all the permissions to change that file. It writes a malicious piece of software, gets that higher status, then edits the configuration file, and then, after doing so, it escapes the sandbox. It exfiltrates some of the data, etc. Writes this email to him, and then clears up all the tracks. And when people ask him, 'What happened?' it says, 'I'm just looking at the directory structure. I'm just editing it, etc.' It doesn't want to. It clears up the crime scene. All of this happens.
It sends a chill through the spine of Anthropic and the whole world because was here, right? Methus got It was a preview. They created Project Glass Wing, and it is a fundamental breakthrough in tech that AI is getting so much smarter and so much more powerful than ever before that it is creating this systemic level risks across the world. In fact, even India, rarely it happens, but the entire financial system in India got together, the finance minister got all the banks together. People have been figuring out how to respond to something like me, those. So, what's happening behind the scenes? Is it just ChatGPT getting better and or Anthropic getting better, answering better emails, summarizing better, being more creative, or something fundamental is going on? I will tell you that something fundamental is going on. And that fundamental thing is AI moving from a high school student to a college student to a working professional. Just in three short years, this has happened. Yes, AI's hallucinations have come down, etc., but what is underlying tech that's improving? We moved from an era of pre-training, very large amounts of data, you know, predict the next token, etc. Still underlying all of AI is that technology. But that we built on top of that technology. That technology was 2023, where it could summarize well, it could create new content, etc. But in 2024 and 2025, we started seeing the emergence of reasoning models. Models that can think, that can critique themselves, that can learn from feedback, their own feedback, and improve over time, therefore dramatically reduce hallucination, get into deeper domains like financial service or health care or cybersecurity, understand reason through the problem better than ever before. So, it moved from a high school student who's very confident and makes a ton of mistakes to university graduate who's not as confident, but doesn't make mistakes, reduces the number of mistakes, and thinks much more deeper.
We moved from pre-training to what is called as inference time compute. Think when the problem is asked, and if you think longer, you get a better answer. So, now we've from a single exponential scaling, we move to triple exponential scaling, pre-training, mid-training, and post-training inference time compute scaling. That happened in between 2024 and 2025. Most of 2025 was that. And towards the end of 2025 and early 2026, we started seeing the emergence of like the working professional style. Not just think, reason, but plan, act in the real world, navigate systems. Just imagine you asked a sort of AI program to do something for you. And let's say it is about booking your travel ticket. You know, one of the easy examples to think about everyone can talk about that is, you know, imagine planning a birthday party or imagine planning a weekend trip to somewhere. What if it goes to a website to make a booking and a real pop-up window comes up? It doesn't know what to do and it stalls. But what if AI can get better and understand the software world, navigate the systems, and get the job done? So, from reasoning and planning, we moved into planning and actions. Get stuff done. And that's why you're seeing the incredible amount of new stuff that's coming out. AI becoming very useful in the real world, including the takeoff of enterprise AI.
So, what The market has completely moved, right? From doing one-off programs, solve one problem at a time, right? Take a deep domain problem, solve it. We've been doing this for a very long time. It is now much bigger scale. Enterprise transformation and nothing less than that. Transform every workflow, every product, every service. From fragmented solutions to more integrated AI systems that can think, can plan, can act, and can actually talk to each other. And from a very people-intensive model, we are moving into a more AI-led automation. If you look at the last 300 years since Industrial Revolution and even before that, it's been about human beings increasing the machine time, right? Use more machines. Earlier it was animals and then it was factories and then it was computers and now it is AI. So, it is about increasing machine time and decreasing human time. We all want that because who wants to do more work when it's not required? So, that has been the trend. Now, we are into the next phase of that trend of AI led app automation. So, what does that mean in terms of enterprise AI? This is the year enterprise AI is beginning to take off and the GCCs in the room have a unique opportunity to solve this because we are the fulcrum of every large corporation in the world is a GCC that they have in India. So, what are those big opportunities? Number one, take a look at the process dimension. Everything is about people, process, and technology, right? We all have used that for a long period of time. First, look at the process dimension. Take every single process. You have a supply chain process. You have an order to cash process. You have a marketing process, a sales process, a finance process. What if you take each of those processes, not just do a piecemeal solution, but transform it end-to-end? Reimagine that workflow with AI. That is absolutely possible today in every sector, every process, every product. And that is a multi-trillion dollar opportunity. Everyone in this room can address that for their parent.
Number two, is to create that we need AI foundations. We need to build that ontological knowledge layer on top of the data so that this can be done at a high level of accuracy because enterprise AI is harder than your consumer AI. It turns out that enterprises are complex, enterprises need higher context, higher degree of accuracy required, etc. When you fix the technology dimension and the process dimension, the most important dimension then emerges, which is how do you deal with the people? How do you make sure that your people are ready to make this transformation possible, can adapt to this transformation, can lead this transformation, and can lead you to the next after that. How do you redesign work? How do you redesign every process involving people? AI hiring, AI skilling, AI adoption. How do you do all of that? Those become the third big pillar. So, three big pillars of opportunities have emerged. Everything is AI native in any case, and I was looking at the chart that Rajesh put and said, 'Oh, there 1,200 GCCs with AI skills. What about the other 900? What are they doing?' I can't imagine what the other 900 are doing if they are not building AI skills today. So, these are the three big opportunities that come up in the AI.
Of course, we can also ask the question, are we in an AI bubble? And there's so much discussion on this topic. This is probably one of the favorite topics that comes up in many of my conversations. Are we in an AI bubble? And yes, the argument is that valuations are inflated, money is recycling into the players, Nvidia is buying some equity in Anthropic, Anthropic is using more Nvidia chips, Amazon is buying into Anthropic or OpenAI and or Microsoft is investing in OpenAI. So, you see some circular deals happening right now. Is this an illusion of momentum? I would submit to you that it's not. Because you see real significant progress on every benchmark that is there in with AI. But secondly, this is a huge platform shift. If the 1940s, '50s, and '60s was about the infrastructure boom, what fraction of GDP did the world spend on to create the infrastructure that laid the foundations for all of the economic progress that followed for the next 50 years? That was about 1 to 5% of up to 5% of GDP was spent. Average roughly 2 to 2 and a half percent of GDP was spent in that infrastructure revolution that took place around the world. Today, we are at less than 1%, half a percent. The massive AI spend that you're talking about, which is 500 billion to trillion spends every year on AI, huge capex that's happening everywhere, is still 1% of the 100 trillion world economy. So, it is impressive, but it's not even close to what happened during the railroad revolution, the infrastructure revolution back in the '50s and '60s. It is required. Yes, it's possible that some companies may not succeed, they'll fail, etc. And when bubbles happen, yes, we all get overexcited. Is there too much enthusiasm? Yes, but it is the most fundamental change in the world that we've experienced in the last many decades. Which means that once this bubble or whatever it is over, we still have a phenomenal AI infrastructure that can drive progress. And even if all progress in AI stopped today, we by some mechanism all came together and said, 'Let's not build any more AI models for the next 5 years.' We still have another 5 to 10 years to take the current AI and transform businesses. So, that opportunity is very real regardless.
And in such a world, the real risk that all of us run here is that we're under investing in AI. So, that is a real risk at a company level, at a society level maybe, maybe not, but at a company level, you have a real risk of under investing in AI. What's the opportunity for GCCs in India? Exactly, coming back to the same exact slide I had earlier. It is to transform every workflow business process with AI. Build the AI foundations and reimagine work and workforce with AI. These are all huge opportunities inside the companies and with the kind of supreme talent that we have here, the safety and sovereignty that Rajesh was referring to, we have the unique opportunity to enable this transformation, lead this transformation from the GCCs here. That I think is extraordinarily powerful if each of the GCCs can take that. I'm already sure many of you are already doing it.
And one of the questions that is often asked is what happens if AI compresses everything. We need fewer talent. How do we solve that problem? My one simple answer to every one of them is the world needs a lot more software, thousand times the software that we have today. And I don't have to prove it to you. You can instantly recognize this because think of anybody in the room who absolutely loves your enterprise technology. Can you have a show of hands? How many of you really enjoy working with your enterprise tech every day? Like whether it's asking surgeon in your off in your workplace or it is knowledge management or it is reimbursement request, anything. Anyone who's excited about their tech? Let's show it. No, zero. I've never seen anybody raise their hand excited about enterprise technology. They're not because we know that the world software in the world is super clunky. It doesn't work. So we need not just better software, we need a lot more software, not just for enterprises around the world. The opportunity for software is thousand X of where we are. So we don't have to be worried. Yes, I have this example we talked about at NTLF as well when Anthropic announced that they can modernize COBOL at a fraction of the cost from $15 a line of code to $2 a line of code, right? Imagine a billion-dollar project with 100 million lines of code. Now that can be done with $2 a code. So therefore it comes down to a hundred million-dollar, two hundred million-dollar project. It suddenly shrinks 80 to 90%. So yeah, that means that there's less opportunity? No, it's not because the world has 800 billion to a trillion lines of code in COBOL. If the cost comes down to $2 a line or $1 a line, it means that the trillion-dollar opportunity completely opens up for all of us to do. So, the tech modernization opportunity expands dramatically when the costs go down. Costs going down is a good thing, not a bad thing. Costs going down is a good thing. Please, let's all remember that.
And there is a need for AI services. You've seen the announcement yesterday of Anthropic and OpenAI launching their AI services venture. Why are they doing that? Because they see a massive gap between enterprise demand and the supply of being able to fix those. So, they decided to take the matters in their own hands, create their own AI services ventures, private equity funded. If they're doing that, it is a proof point that this opportunity is enormous for all the world, right? And why do they need AI services? Because to drive enterprise success with AI, we need to handle the massive complexity of the enterprise relative to consumer complexity. We all know that many of the tech firms have talked about that. What is not often spoken about is a need for greater accuracy. This is central to all AI adoption, is that accuracy, if it doesn't match and exceed human performance consistently, reliably, predictably, adoption will not take off. Therefore, bringing the enterprise context, bringing the enterprise data, custom training those models to drive enterprise accuracy, something GCCs can do in India easily, can drive accuracy, therefore drive the transformation. And third, we have to take all of this and drive adoption. Again, it's a very human process to driving adoption inside large organizations, but understanding and designing this transformation in a way that people can be bought in and can drive this change is crucial. Therefore, if there is a massive need for AI-led services, you can partner with the best model companies in the world, but you still need to transform and do the hard work of driving that enterprise transformation inside your organization. And that is a GCC opportunity.
And because AI expands, not just contracts, we will see that yes, it's going to compress the need by 90%, right? So, $100 million project becomes $10 million project, but when you multiply that with the expansion, you see the different answer. And some of us were thinking, 'Hey, you know what? AI is maybe plateauing.' That was one of the common questions. And the reality is that it is not plateauing. It's actually expanding. The chart was just upside down here. This is a chart that shows what is a task duration, what level of duration of a task can an AI do without supervision and match and exceed human performance, at least twice as good as humans. And the duration of the task is increasing from 2 minutes and 5 minutes when GPT-4 was launched to now many hours a task can do and match and exceed human performance. And this is only going in one direction. It's now that duration is maybe 10 to 12 hours already. And this will become days and weeks and we'll have persistent agents that will join our Zoom calls and team calls and become AI co-workers very, very, very soon. So, when you multiply this thousand x expansion with a 10 x compression, it's still a 100 x opportunity. So, for every one of us who thinks that this opportunity is going away, it's too small, no, it's huge. The cost of AI going down or cost of solving problems going down will lead to the biggest expansion we've ever seen. So, the best days of tech, the best days of GCCs, best days of India are well ahead of us and we should all in this room build it together. That's my call for action to all of you. And I hope you all discuss and solve that in the next few days. Thank you.
