Inside Microsoft's Trillion Dollar Playbook with Chief Strategy Officer Bobby Yerramilli-Rao

Bobby Yerramilli-Rao: You know, in tech there's no such thing as franchise value. You're only as good as your last decisions. Technology is is in the nature of disrupting everything, including itself. And so you can't just rely on saying, okay, I did well yesterday, so I'm gonna do okay tomorrow. And when you look at productivity benefits that could potentially be delivered through ai, our sizing is that it.

Could be as much as $20 trillion of additional GDP in today's dollars achieved over the next 20 years.

‍Sriram Viswanathan: Do you worry about what has happened in the semiconductor space from a manufacturing standpoint? You know, it portends what might happen in the AI landscape where us loses advantage 

Bobby Yerramilli-Rao: in some way. We have to look out for zero trillion dollar markets because of the size of the company that we're in and the tech industry actually.

In some ways at births these sorts of opportunities at some regularity, sort of once every five to 10 years, something comes along, which is a big enough paradigm shift or a totally new way of addressing customer needs.

Sriram Viswanathan: Hi everyone. This is the Tech Surge Deep Tech Podcast presented by Celesta Capital. Each episode, we spotlight issues and voices at the intersection of emerging technologies, company building and venture investment. I am Sri Ramathan, founding managing partner at Celesta Capital. If you enjoy tech Search, now is a perfect time to hit the like and subscribe button.

And while you're at it, you can leave us a review on your favorite podcast platform. If you're just discovering us, visit tech search podcast.com to sign up for our newsletter and check out the archive of some very interesting past episodes. We're pleased to welcome Bobby Ya Lee Raw to the podcast Today.

Bobby is the Chief Strategy Officer at Microsoft where he helps the company identify and execute some of their most important strategic priorities for growth and he has served Microsoft in a very interesting period. Starting from 2020 where Microsoft has made some of the most profound commitments in the AI space, some of which, uh, Bobby has been involved in.

Bobby has been, uh, really instrumental in shaping the company's strategy. In his previous roles, he was also the managing partner at a growth equity firm called Fusion Global Capital. Bobby has spent years at McKinsey, uh, and also was the Chief Strategy Officer and head of m and a at Vodafone. So clearly a very rich background.

I am super excited to talk to Bobby today on a variety of topics. Great to be here. Israel, I. So we have, we have lots and lots of topics to talk about, but I wanna really start from your current role and your in Microsoft. Talk to us about what your, you know, day job really is, and how do you really go about thinking about the strategic priorities for a company that has been so wildly successful.

It's almost like the heartbeat of the computer industry for many decades. How do you come up with a strategic priority list? That you can go and talk to your board and executive management about what, what, what is that like? I think

Bobby Yerramilli-Rao: the, uh, the thing with Microsoft is that strategy at Microsoft is very different to how you'd think about it in other companies just because of the breadth and the size of the company.

And at its very core, it is about identifying the priorities for the business and the opportunities that we have. We do that in combination with, uh, with Satya and the whole, uh, of the executive team. But really the role that we play is around making sure there's enough foresight and rigor that we know the choices we're making are the right ones.

And, um, to borrow a phrase from Vinod where Vinod Kler, Vinod Kler, who says, uh, his job is to invest in companies that are creating zero billion dollar markets. Uh, in some way we have to look out for zero trillion dollar markets. Mm-hmm. Because of the size of the company that we're in, and the tech industry actually in some ways births these sorts of opportunities at some regularity.

Mm-hmm. Sort of once every five to 10 years something comes along, which is a big enough paradigm shift or a totally new way of addressing customer needs that. It's possible to continue to identify these things. And a big part of what, uh, I feel I have to do in the company is to make sure that we anticipate those sorts of big shifts and are well positioned to take advantage of them, uh, when they

Sriram Viswanathan: occur.

You know, before we get into some specifics of some of those important strategic, uh, priorities in, in the last four or five years that you've been at, uh, Microsoft, if you were to sort of look at. Things that you've thought are high priorities for the company. On a scale of zero to 10, how many of them have actually gotten incorporated or become priorities for the board and for the leadership?

Bobby Yerramilli-Rao: Well, first of all, it's not just me. So there's a whole executive team. Uh, in addition to my team, I'd say we've done pretty well in identifying the sorts of things that could happen and making sure that we are at least well positioned on most of the big things. I don't think there's anything big that we've.

Completely missed or totally. We're sidelined in. Mm. So I think that we're relatively, we, we can give ourselves some decent grades on that. Mm. Um, and I think that we're relatively well positioned in a lot of categories, but in tech there's no such thing as franchise value. You're only as good as your last decisions Technology has is, is in the nature of disrupting.

Everything including itself. Mm. And so you can't just rely on saying, okay, I did well yesterday, so I'm gonna do okay tomorrow. Mm. So this is a game of continuous, continuous vigilance and making sure that we're making the adjustments and the bets that we need to. Yeah.

Sriram Viswanathan: So that's a, that's a good place for us to sort of put a pause on where things are today.

And before we talk about the future, I reflect on this one conversation you and I had several years ago where we were talking about, you know, the importance of strategic. Inflection points, and we were discussing it in the context of Andy Grove and, uh, innovator's Dilemma. And we're talking about how companies often make mistakes because the places where they think they have the strategic threat and focus on invariably doesn't happen to be the strategic threat.

And something else comes about and. Fundamentally undermines the company's future potential and all of that. And we were discussing it in the context of low power and Intel's inability to really deliver a whole bunch of things. And this was many years ago. If you were to reflect on the major strategic inflection points that Microsoft.

Was facing over the, let's say the last 10, 20 years. What are the ones that sort of stand out for you? That Microsoft really got it. And what are the ones that may have been missed? How, how would you, how would you characterize that? Well, I think

Bobby Yerramilli-Rao: you can go all the way back to the inception of the company.

Uh, of course, uh, well before my time, uh, in the company, uh, the company's now 50 years old, so there's a lot of history to draw on, and I think you can say that the founders at that time. Obviously had an amazing degree of foresight in realizing that the intel architecture ultimately would cost reduce, uh, and create, uh, a, a major demand for a PC on every desktop, a server in every, uh, enterprise.

And then the associated productivity software that would be going alongside those. And that is really the founding Yeah. Uh, ideas of the company and the, the foresight and the drive to position there is what made Microsoft in the first place right into, at that time, you know, one of the most valuable companies, if not the most valuable company, over periods of time.

Mm-hmm. So that was incredible. Those three decisions around, yeah, around those three inflection points was clearly, uh, well taken thereafter. There was a period of time where. A number of things occurred where the company, for whatever reason, was not well positioned in those, uh, including, for example, search.

So while browser, I think the company did, uh, actually anticipate it, the real game was search. Mm-hmm. And that is a different story, uh, mobile where again, you know, that has become, uh, an area where the, the company didn't. Ultimately end up with a winning position. Mm. Uh, e-commerce, for whatever reason is chosen to not be an area in the company.

Social networking, of course we have LinkedIn, but it's not that, uh, social networking became a core of, of the company. And some of these were, uh, deliberate, uh, decisions and some of them were decisions I think were, uh, perhaps it wasn't, uh, well anticipated enough or well executed enough. Uh, I wasn't there.

More recently, I think you can say that, uh, certain decisions were called correctly. So cloud software as a service, and we'll see where we are with, uh, with ai. Hmm. But, uh, in each era, I think we can say that there are, there are these sorts of opportunities over a 50 year period. I've rattled off maybe. 10.

Yeah. So roughly every five years one of these comes up, and if you don't call it right, you could be in the penalty box for quite a while. You call it. Right. Then you're, you're good until the next one comes along. Well,

Sriram Viswanathan: I mean, it's, it's really profound in this period, in the last 20 years, 25 years, because 90 99 Cisco, Intel, Microsoft, were the three companies that crossed $500 billion in market cap.

Mm-hmm. And 25 years later, Cisco is at. Two $50 billion and Intel is sub hundred billion dollars and Microsoft is above $3 trillion. So clearly the, you know, the hit rate that you had or the significance of the decision that you as a company Microsoft made, made the difference between one being so wildly successful versus the other two Not as much.

Is it the significance of these decisions aligning with the industry trends that made the difference? Or is it the timing of it? Or is it the corporate commitment to go behind these decisions and you know, sort of bet the company and put all of your muscle behind it that made the difference in your opinion?

Bobby Yerramilli-Rao: You know, it's a bit of everything. Yeah. But I'd say the most important thing to start with is to make sure that you are catching the, uh, upswing at the time when it's going up Right. And so you can be right. Five years early, in which case you may as well be wrong. And being right five years late means you are wrong.

Yeah. So you have to catch the upswing at the right time. Right. And that, I think is a big, big portion of the success. Right. Then beyond that is how do you make sure you actually catch it right?

Bobby Yerramilli-Rao: And there, um, it's in some ways a little easier for a small company. Because you don't have anything else, and this is what your bet is.

And if the upswing is as you expect it to be, then you're gonna do great. Otherwise, you, you are a business with a, with a big company. You've got all sorts of other factors that weigh in, including existing businesses that may in fact be getting disrupted by this new upswing that's happening. Uh, or this new inflection point.

You may not have the right. Skills. Mm. And people to be able to take advantage of this. Uh, there may be a new business model that's put in place Mm. That causes a disruption that you are not familiar with or, or comfortable with. And especially if you're a large public company, you have got a variety of different, uh, you know, stakeholders that you need to watch out for.

Yeah. So these are all factors that need to be navigated, and the art is to find a way to be able to take the strengths of the assets that you have. As a big and successful company, not completely abandon those, but not be held back by those in reaching for the new frontier. Yeah, and that's an art form, which is very difficult to get right even in a company like Microsoft, which is arguably successful now or un undeniably successful now undeniably, undeniably successful now, but arguably has been successful for 50 years.

Even Microsoft, you could say. Didn't completely get it right for some of those decisions. So this is, this is hard to do.

Sriram Viswanathan: Yeah. One of the strategic decisions that you did not mention is, uh, gaming and entertainment, uh, you neither called it, uh, as one of the things that you succeeded in, nor failed in is, is, you know, Microsoft Made, was one of the pioneers in, uh, Xbox and had a phenomenal success in a very brief period.

And then of course, she went ahead and made a very large acquisition. Activision, how would you rate Microsoft's play, you know, are not intended in the gaming, uh, opportunity? Well,

Bobby Yerramilli-Rao: I think in gaming, the, uh, the business model is shifting and I think this is one where the acquisition actually speaks directly.

I. To riding the inflection point, which is that what we are seeing is that the opportunity in gaming is, is to sort of emulate what Netflix does and create cloud gaming services, streaming gaming services that are device independent because there are 3 billion people on the planet who play. Games. Mm.

But, uh, a small fraction of those actually do it through a dedicated large console or pc. And so we are, uh, trying our best to make, uh, gaming more democratized. And that's kind of what the business model shift is. Mm. And if you want to play that game, then really what you need is you need a lot of content.

Yeah. And so you need studios. So this was a way for us to play that inflection. Yeah. Uh, in a way that was intentional and again, making sure we're not abandoning our heritage. 'cause Xbox is a extremely well loved brand, and there's a whole group of people in the world that relate to it extremely, uh, strongly Hmm.

At the same time as moving to a, a new way of experiencing the brand and, and the, uh, and, and the content from that.

Sriram Viswanathan: You speak of content and I, I spent my early days working very closely with, uh, with Microsoft and Microsoft and Intel had a presence in Hollywood and. You know, you remember that period when it was actually called Silly wood as in Silicon Valley and Hollywood coming together, and content was a very big piece.

Uh, and Microsoft was, you know, was pioneering in those days to even, you know, form M-S-N-B-C and content was a big thing. Does Microsoft continue to have a focus on content in that sense outside of, um, you know, whatever you're doing in, in gaming? I, I would say

Bobby Yerramilli-Rao: not, other than the fact that there are.

Potentially some, uh, logical brand extensions from some of our game game franchises. Right. Uh, some of our game franchises now spawning TV shows, and in fact some, some movies as well. Those are not things where we necessarily play a leading role. Mm. But, uh, it's great to see that there's potential for those brands to become strong enough, right?

For people to want to see, uh, more experiences in, in, in that way. Right. I imagine over time, as we think about. Different ways in which people will experience content that strong brands and strong stories will find new ways to be, uh, shared with the audience.

Sriram Viswanathan: Some of the major acquisitions that Microsoft has made have worked out phenomenally well know LinkedIn, GitHub, Activision, blizzard, but some things have not like, like Skype.

And without getting into Skype, particularly since that that happened before your time, is there any observations that you would make as to why some things work very well and why some things don't? Well, I think that. There is always

Bobby Yerramilli-Rao: the potential for a big company to mess up a small company. Mm. Uh, there's a variety of different reasons for that to happen and, you know, Microsoft is not immune to making those sorts of mistakes as well, uh, even today, so it's something we have to watch out for.

The pattern that occurred with LinkedIn and GitHub was a very interesting one where the companies were not heavily integrated into Microsoft. They were left a little bit to their own devices. Uh, not completely to their own devices, but, uh, not totally integrated in a way that would, uh, prevent them from having their own, uh, ability to, to execute in the way that they thought was successful in the past.

Mm-hmm. And that's actually worked out really well for, for both LinkedIn and GitHub. And with GitHub in particular, there was a great deal of sensitivity because among the open source community, if we started to. Make it just an Azure only kind of, uh, uh, activity that would've turned off a lot of people.

So that, I think was, both of those were handled very well in allowing the mission and the purity of what made those, uh, into great products to be able to continue to thrive without kind of squashing it. With a whole bunch of, uh, Microsoft stuff that could, uh, potentially have theoretically been synergistic, but actually in practice would've led to the companies being less successful.

Um, I can't comment whether that's what happened with Skype in particular. Yeah, you were not there. Yeah. Uh, I wasn't there. Um, but in the end with Skype, we were able to recover because once again, there was an inflection that we were able to call correctly at that time. So we were able to. Kind of, uh, have

Sriram Viswanathan: a, a lucky, uh, recovery there.

I can sort of see the point that you made about leaving some of these companies to do their stuff by themselves. When, when you, as a big company are, are acquiring them and specifically in the context of LinkedIn, if you are on LinkedIn, you would never know that it's actually a Microsoft product as a service.

You would, you just be in the LinkedIn umbrella. How does Microsoft leverage. That community because you clearly, it's, it's undeniable that Microsoft didn't really make a big play or maybe not succeed in making a big play in the general social networking space as we all, you know, come to know what social networking is.

But in LinkedIn it's a very targeted cohort of people that are very enterprise driven and they are, you know, they are coming in there, you know, not to discuss their, you know, family vacation. They're talking about their skills and. Career objectives and all of that. How does Microsoft really leverage that demographic of people in the Microsoft network, so to speak, and how do you monetize or how do you leverage that?

Bobby Yerramilli-Rao: Yeah, the synergies with each company. So first of all, we've been successful in, in the model of keeping companies relatively independent. Yeah. Uh, with some degree of synergy, but not totally, uh, integrated. And then we've also done well or tried to, uh, uh, integrate companies so that both models have, have worked.

And, uh, there's also some, uh, cases you pointed out where the integrated model hasn't worked in, in terms of, uh, LinkedIn. And GitHub, we've done our best to try to get synergies without disrupting the core of the product and the value proposition. So when you look at LinkedIn, you might not see that it's Microsoft, but actually there's a lot of stuff behind the scenes, which is more and more becoming, leveraging other stuff that we're doing in terms of models and in terms of, uh, you know, the way that we think about introducing ai.

Equally, we learn a lot from how the community is managed. Uh, how you think about making sure you are, you are really following the community, but also, uh, gently steering as well where you can. Hmm. Uh, those sorts of, uh, things are, are lessons that we, we learn, which actually come back to the core of what we do in Microsoft in general.

Hmm.

Sriram Viswanathan: Each company has a different strategy on make versus buy versus partner. In fact, hp along with, I forget if it was Stanford or Harvard, did a case study. Comparing Microsoft, Cisco, Intel, and HP on how they partner and how they acquire and how they invest in companies, uh, external to, to your business.

Can you shed some light on how does Microsoft look at what's the best opportunity and what. Time does it make sense to buy versus partner versus versus invest in, and those are tough decisions. And I, you know, I understand that each situation is different, but broadly, can you just shed some light on that?

Bobby Yerramilli-Rao: Well, as a big company, the first question we have to ask ourselves is, why can't we build this ourselves? Uh, it, it should be the thing we are able to do, but for many reasons, soften is, is not possible to build it ourselves. And you know, we obviously have a lot of things that we're doing ourselves as well, so you have to kind of take.

Resource from somewhere else if you're gonna do that. And so, so time to market would be one. And so the, one of the things that we, we are always thinking about is, look, we could, in software, you can build anything, right? Or at least you can convince yourself. You can build everything. But actually the, the thing which actually causes you to decide that it's worth, uh, looking at.

Uh, either partnering or or buying a a company is because they've got something which is difficult to catch up within a reasonable amount of time. Yeah. Or they've got some access to a network effect or a community that is very difficult to replicate even if you showed up.

Yeah.

Bobby Yerramilli-Rao: So those are the things which really would, you know, primarily cause us to say, I.

That, you know, that's something that we should probably think about, uh, not trying to do ourselves. Then the question is, do we partner or do we buy now? Part of it comes down to also what that company wants. Mm. Because there are companies where you say, I'd love to acquire this company, but that's not in their game plan themselves.

They want to continue doing what they're doing, and frankly, it might be the best way to do it because. Uh, all of the companies are based on the talent that they have. Mm. And if the talent they have decides that being part of a big company isn't part of their game plan, then you lose them and then the whole thing becomes a lot less valuable straight away.

Mm. So that's part of it. Also, part of it is frankly, that I, in many cases, we are able to accomplish a lot through partnerships without having to acquire. And that's just a lighter. Approach from every aspect. So, you know, the first thing is, you know, can we build it if we can't? And, you know, should we partner?

If we, if we, if we can't partner or partnering isn't the right approach, then uh, it, we, we'll look at buying. And of course, there, it's just a, a very normal process of understanding what the, uh, what the synergies are and what we can accomplish by, by having it as part of.

Sriram Viswanathan: I think I, I wanna go back to the, the point you made about the key strategic opportunities for companies to make some decisions on and sounds like cloud, which was largely, you know, under Satya's leadership, the company made a phenomenal shift from packaged software and embracing the cloud and office entirely, you know, moving over to the cloud.

That was a phenomenal transition. Would you put a company's focus in embracing of AI in that same class? Is it bigger in terms of strategic impact, smaller, comparable? What? What, how would you think of that? Yeah. I think that

Bobby Yerramilli-Rao: AI is bigger. Yeah. Uh, I think it may be the biggest of all. Most likely should be the biggest of all.

Frankly, the way we look at AI is on what's the impact going to be on GDP Mm. The North Star for us ultimately is, is making sure that our customers. Are getting value from our products. And so we have to find the, a way to be able to, uh, demonstrate that. And, uh, for the enterprise side, the way to demonstrate it and to understand it is through looking at productivity benefits.

And when you look at productivity benefits that could potentially be delivered through ai, our sizing is that it could be as much as $20 trillion of additional GDP in today's dollars. Achieved over the next 20 years,

Sriram Viswanathan: $20 trillion for, for the world's GDP, for world's GDP to be

Bobby Yerramilli-Rao: increased. Yeah. Yeah. So World's GDP right now is something like $110 trillion.

Yeah. So an addit $20 trillion is like landing another USA or another China onto the planet. Now it's gonna take 20 years because adoption takes that long. But if that were to occur over the next 20 years, what you're looking at is adding a whole point of growth to global GDP. Each year for the next 20 years, which is a phenomenal kind of, uh, uh, you know, uh, shot in the arm for right, for global productivity and for everyone's prosperity.

So we are very focused on making sure that we are able to allow that to happen for society because when society benefits and customers that we are selling to are, are the ones who will also be able to benefit. And then that's how we grow our own business. Right. And that is a much, much bigger figure. Mm.

Than is associated with almost anything else that has occurred in the past. Even big, big shifts like cloud or mobile Mm uh, are smaller than that. Mm. Uh, and that, by the way, is based on the AI that we can project now, but of course, it's moving so rapidly. You know, you're getting new advances almost, uh, in, you know, every couple of months.

Mm. So it could be that as we look at this, we end up revising these figures upwards, to be honest. Sure. Uh, the, the timeframe is a little difficult to predict. Yeah. Because adoption cycles take longer. The benefit of the technology that we have right now has, is, is hardly adopted or, or seen in adoption yet.

Hmm. So there's still a long way to go in terms of getting the benefit. Across the whole of society, of even the, the, the software that we have now. But if you think about what's gonna be there in five years from now, 10 years from now, let alone 15 years from now, the benefits are just gonna be enormous.

Sriram Viswanathan: So, and I get that, I get the overall macro level, uh, uptake in GDP, uh, you know, whether it is. 20 points or 15 points. I think it is not debatable that there's phenomenal opportunity for value creation at a macro level. Now, if you were to sort of hone in on a, you know, individual company or individual business line, you know, adding co-pilot to Microsoft office, which is largely driven by.

AI in the infrastructure in Azure, uh, how does it translate to a revenue stream for the enterprise business within Microsoft? Because, you know, if I'm used to using Microsoft Office as a cloud application, and just because you added a co-pilot for which you actually spend billions of dollars in infrastructure.

On ai, but if you're not able to collect incremental revenue dollar, uh, how, how would you justify that? Uh, we do charge for it. Well, you, you charge for it. Uh, but does it justify the tens of billions of dollars that you have to invest to just enable that capability?

Bobby Yerramilli-Rao: Yeah, so, you know, we are in the early stages of this, uh, business growing.

It's the fastest growing business that we've ever had inside of Microsoft. So it's actually performing really quite well. But it's still early innings in actually how we measure the performance of the business itself. And the AI business is one where you have to invest a lot upfront in training, right?

Uh, when you think about things like cloud or inferencing, you can roll out the spend in some way. That's. Gauged or, or gated by the expected demand. Mm-hmm. So you can look out 18 months, 24 months and say, this is what we expect demand to be. And then you can build out, uh, capacity ahead of time. So that's something you can at least connect to demand.

Yeah. With training, you have to spend all the money upfront. Right. Build the model. Put that into products, and then the products are then, uh, uh, rolled out and adopted. So there is a lag. Uh, and there's a bit of an upfront, uh, there. The good news is for us, frankly, is that of the, and we've been public, we're spending $80 billion a year on, on building out, uh, of the $80 billion.

It's actually a, the minority of that is actually spent on training. The majority of that is spent on classical cloud, uh, compute and, uh, classical, uh, and, and inferencing, both of which are informed by demand. But there is a, a big amount, it's a minority of 80 billion, but the minority of 80 billion is still a lot of money.

Yeah. Uh, that we have to spend upfront. And so when you say does it justify it? Well, we obviously think it does because we wouldn't otherwise be investing in it. But it is really, this is a startup at massive scale.

Sriram Viswanathan: You know, I want to draw the parallel to, uh. Traditional semiconductor, uh, fab investment that a lot of the foundries make.

And yeah, you know, tens of billions of dollars on CapEx and r and d for process technology for next generation process. And the software ecosystem would pride its, you know, business model by saying, you know, the software business doesn't require huge CapEx that seems to have completely reversed or at least become equal to the kind of investment a hardware.

Vendor would have to make, and I don't see multiple semiconductor manufacturing foundries, you know, save TSMC, perhaps making $80 billion of investments per year in semiconductor manufacturing. But you have multiple hyperscalers committing, you know, tens of billions of dollars. So from your vantage point, do you think that the argument that software historically is not a CapEx intensive business, do you think that that has changed?

Uh, thanks to AI where everybody's really, the software companies, you know, are investing so much in the hardware infrastructure, in building the cloud versus the amount of money that's going into fabricating the underlying chips that have to power these infrastructures.

Bobby Yerramilli-Rao: I'd say software is not becoming a.

CapEx intensive business, but some people like us are operating an integrated model. What cloud is becoming cloud has always been a CapEx intensive business. Right? And so again, the majority of what we're spending is on traditional cloud compute and inferencing, which is yet another form of cloud compute, right?

It's the training is the part which is. The new bit. Right. Right. And so, uh, you know, there are software companies that, uh, like in the semiconductor industry are fabulous, right? Sure. They, they write on other people's clouds. Sure. And there are those which are like in the semiconductor industry, are IDMs because they have both like us.

Sure. Uh, like Google. And so, uh, we are, we are spending on, on the, on the, uh, compute fabric at the same time as. Uh, launching software.

Yeah.

Bobby Yerramilli-Rao: But that doesn't mean software itself is a CapEx intensive business. Right. 'cause you can run it without it. Sure. Now the new bit is the training. Right. And the question that, uh, you know, you are focusing in on is, is the training worth it?

Yes. And I think that a couple of things are, are interesting in terms of training. So the, the change recently. To the ability to continuously improve, uh, the, uh, the models. Yeah. Uh, and be able to continue to, to get value outta those is, is a big shift from the, you know, let's just train the largest thing we can and then put that out there and then train the next largest thing we can.

Mm-hmm.

Bobby Yerramilli-Rao: So that actually has changed a little bit, the, the whole, uh, economics of, of the training. Mm-hmm. Uh, in addition to which there's post-training, uh, activities as well, which mean that you can leverage. A base model in multiple different ways. Hmm. So are those of changing the economics of, of, uh, of training in a way that I think is very beneficial?

Quite apart from, frankly, the fact that, uh, there are multiple different vectors in which you can cost reduce training, uh, as we've seen in a deep seek, et cetera. Right. Which is a good point. So, so what's your point of view on deep seek? How you guys think about it? It, it wasn't a surprise to us. We was a company that we've been tracking for a long time and there's a couple of others, uh, like that around the world as well, including in China.

I think they've done a remarkably good job of putting together a number of techniques, many of which were known, but they've done a remarkably good job of putting them all together in a coherent way with a specific mission of being able to develop the most cost reduced, uh, model, uh, that's out there.

And I think that that was a, a fantastic way to demonstrate that these techniques can be put together in that way.

Sriram Viswanathan: But is it, is it largely. In your mind because of the model distillation that they got that kind of a performance improvement at that low cost of spend on the training? Or what would you attribute that to?

Bobby Yerramilli-Rao: No, there were a number of different techniques that they used that were, uh, about how the model was actually trained and how it's actually, uh, operated that caused 'em to be able to do that, so. Right. You know, I think it's, it's, uh, it's not fair to say that it was a distal model and that's all it is. I think there was some very, very innovative.

Sriram Viswanathan: And, uh, important engineering that was done there. Was there anything surprising about the fact that it came out of China versus Mountain View? You know, some of the

Bobby Yerramilli-Rao: best AI researchers in the world, including in Microsoft, are Chinese. Yeah. So,

Sriram Viswanathan: no surprise. Yeah, that's great. That's, uh, that's very interesting.

Okay. Another, another area that I wanna sort of. Uh, dig into a little bit is, uh, the whole quantum computing space, and love to get your thoughts on where do you think we are in the evolution of that technology itself. You know, are we in the, as in your parlance, are we in the first inning or, you know, are we like, you know, have not even started the game yet?

Where are we in quantum computing?

Bobby Yerramilli-Rao: I think that quantum computing is one of these areas that has been overhyped for a long time, but I think that what we've got right now is a real path towards one, and I think it's important to acknowledge both the overhype that exists and the part that actually, I think we have a, a real, uh, shot this time.

There's multiple different approaches to this, as you may know. And what we've been following is, is an approach called topological. Qubits, which, uh, it would be great if you could invent one. It was theoretically proven in the 1930s or so, and no one had ever actually created one. If you could create one, you could create many.

That was the, uh, that was the reason why people wanted to pursue it, because as you know, to solve any, I. Useful problem in quantum, you need many, many hundreds, if not uh, uh, thousands of logical qubits. Other people have followed approaches, which are easier to stand up a few qubits, but much more difficult to scale.

So what has happened, I suspect, in the industry, is a lot of people have extrapolated from early progress and said, well, you know. The, uh, finish line is near. Mm-hmm. Which I think is actually proving to be damaging for an industry, which is, you know, actually quite scientifically challenging. In our case, it was a long period to create even one of them, which I think we've now, uh, been able to do successfully and actually reduce it to a chip that should

Sriram Viswanathan: allow us to scale.

Yeah. Well, I mean, look at, I think some in the industry will push back, you know, much to the point that you make, which is that there's a lot of hype about it. Uh, and would. Probably point a finger at Microsoft also, and I'm not saying they're right, I'm just saying that there are people that don't really think that it accurately reflects what Microsoft announced to what can actually be done with it.

I mean, you know, clearly you've seen the articles about, uh, Microsoft's, uh, claim that you've created a new state of matter is a bit overblown and a bit hyped. Would you, would you, how would you react to that? I

Bobby Yerramilli-Rao: think it's. Healthy that in a field like this, which is about scientific discovery, that there's challenge and peer review and pushback because that's the only way you can really be sure that you have a scientific discovery.

So I think it's an an entirely normal and healthy process that rather than people just accepting what everyone else says, right, that people are going to ask questions. And I think that's a good thing. Uh, I'm hoping that as they ask the questions, they'll get satisfied that actually what we've said is correct.

The important thing is that I think we've moved. From an era where it might be possible in the some distant future to an era where I think it's going to happen in timeframe that's tractable. Mm. And that will allow us to address a range of problems. So there's two areas of, of hypo confusion in quantum.

One is when or if it'll ever happen. Mm. Which we've spoken about. The other is what it will be able to do. Mm-hmm. And the, the. Primary thing that quantum does is simulate, uh, chemistry. Hmm. Now you could say, well, that's not much. It's chemistry. But except everything is chemistry because everything that is made is, is chemical matter.

Hmm.

Bobby Yerramilli-Rao: So if you think about the ability to develop new, uh, materials, uh, new drugs. Energy stores. There's a, a whole panoply of things that will be opened up as a consequence of this over time, becoming a real, uh, adjunct. And, and by the way, it's quantum plus AI plus, uh, high performance computing. It's not just quantum on its own.

It's all three Yeah. Are gonna be, uh, mixed up together in ways that will allow us to invent useful stuff. So,

Sriram Viswanathan: you know, you and I are old enough to know that. AI had a lot of hype, uh, for decades, and then there were a bunch of great innovations that happened including, you know, Jeff Hinton's work on back propagation and deep learning and machine learning and all of that.

Is there a comparable thing in Quantum that needs to happen before it becomes truly mainstream or it'll continue to be in, sort of in the science project realm for decades to come? I would say that there's a couple

Bobby Yerramilli-Rao: of different. Technologies now that have kind of had that moment, I would say in Quantum, yes.

And I would say Meyer Honor is one. It's not the only one. Yeah. There are a couple of others as well. Right. Uh, which are looking very, very good in terms of where they can get to. We still believe that the, um, the topological approach is going to be ultimately the, the most. Scalable and the best, uh, and the fastest, uh, uh, within quantum computing, uh, fastest operation.

Sriram Viswanathan: Uh, so it's not just, you know, millions of qubits, it's the approach of topological qubits that you would compare. To, you know, something like what happened in AI with back propagation. I know that these are different, but I'm just trying to understand, you know, the significance of this announcement. Yeah, it's a, it's, it's reasonable

Bobby Yerramilli-Rao: analogy, Ian, because the point was that those mathematical innovations that took place coupled with large scale compute and a lot of engineering and some very, very intelligent computer science yeah.

Led to where we are now. But you had to have those breakthroughs at the time and I think that we could. In some way draw an analogy and say, these are similar to those sorts of breakthroughs. There's still some very significant engineering to be done, but it's, the path is a lot more clear than it's ever been.

Sriram Viswanathan: That's interesting. So you would see a future where many of these AI models, uh, at least in the Microsoft environment running off, uh, one of these mirna. Uh, chips or a systems No, no, no. You wouldn't do that.

Bobby Yerramilli-Rao: AI models don't run on quantum computers. They would run on, you know, traditional classical computers.

Yeah. And GPUs and all the rest of the stuff. So what you have to think about with Quantum is that it's, it's useful for solving certain problems. Right? Or would they be simulating quantum processes? So it's things at the atomic level, at the molecular level, yeah. About how interactions occur. Yeah. Uh, between atoms and electron clouds.

Right. And being able to solve those allows you to then be able to invent. New materials, new atoms, new molecules, new drugs. Yeah. Uh, new, new stores of energy, you know, these sorts of things. Yeah. And so, but what you would want to do is to be able to use it in combination with high performance computing and an ai, because any given problem of that nature would need some element of quantum to be able to solve some elements of it, and the other parts of it can be solved.

More efficiently and more effectively, more quickly using AI and HPC. Mm. So typically it's gonna be a combination of all three, but you wouldn't be running AI models and algos on, on a quantum computer that would run, that would run different stuff.

Sriram Viswanathan: Got it. So as, as a chief strategy office, uh, you are obviously focusing on what's coming down the pike in some of these new technological breakthroughs.

What would you say are some of the big ideas that you would be focusing on over the next, let's say 10 years? Over the next 10

Bobby Yerramilli-Rao: years, it's still going to be largely speaking around AI and the development of ai. It's, there's a very rich seam to be, uh, to be mined. There. We're still, only if you can believe it, about two and a half years after the Chad GBT and G PT four moment.

So it's really. Been, uh, not a very long time, although it seems like a whole lot of things have happened in that period of time. It's really kind of short. Mm. So for me right now, I think there's the combination of what we can do with ai, introducing it into products, different form factors, to be able to make sure that it becomes more pervasive and more.

You have a more immersive kind of continuous experience that's gonna be really the focus for a long time, augmented by Quantum as it arrives as well. I think those two things are gonna see us through for a long period of

Sriram Viswanathan: time. Let's switch to, uh, another topic, which is in, uh, in some areas that you are involved in outside of Microsoft.

Uh, specifically you're on the board of Global Foundries for the past several years, and you obviously have a front row. Seat on semiconductor manufacturing industry at large. And foundries like, you know, gf, uh, operate in a very competitive landscape, and they have huge scale and competitive pressures of, uh, of the, the leading, uh, foundry, TSMC.

Can you talk about the, just the state of the Foundry business? From your vantage point specifically in the US context? I think that

Bobby Yerramilli-Rao: nobody is confused about the importance of manufacturing semiconductors in the us. Uh, I think that it's, you know, the industry better than I do, uh, siram in terms of all of the decisions that were taken that caused manufacturing of semiconductors to not be primarily, uh, based here when this was the country that actually started the whole thing in the first place.

Mm-hmm. But, uh, from the GF standpoint, you know, we are, we're very proud to be actually leading. This here and is a whole host of things that I think that are valuable quite apart from creating manufacturing jobs and skills. It also generates incentive for people to, to learn semiconductor engineering and manufacturing engineering, which is a, a declining enrollment in universities now.

And that's a shame because it's the sort of skills that are valuable, especially in the era of AI when you need more compute Mm. Is the sort of thing you really need. In addition to the geopolitical aspects of it, there's very good economic arguments to be, uh, in the US as well. So I think people are, are realizing the importance of it.

And you know, it takes a long time, as you know Yeah. To set up a foundry, to set up a fab, to be able to operate it in a way that's efficient and gets the right yield, the right cost points. These are not overnight things. They take ages and so I think that global foundries in some way is a sort of a.

Underappreciated crown jewel of the, of the us mm-hmm. Because all the effort that went into creating it and making it as effective as it is, is something that, uh, the, the country should be proud of. I know that everyone who works there is very proud of it. I'm certainly very proud to be on the board of it.

And, uh, over time I think people are coming to realize the importance

Sriram Viswanathan: of this. It's interesting you should say all of this because, uh, you know, my co-founding partner. Uh, has just been announced as the CEO of Intel. Absolutely. So as you can imagine, we pay particular attention to this space because a lot of what we do is in deep tech investing and lots of semiconductors and all of that.

But I want to draw a parallel to the ai. Landscape and the semiconductor manufacturing landscape. If you were to look at the semiconductor manufacturing landscape, US was clearly the leader for many, many years with Intel and a MD before they actually shed their manufacturing. That's right. Perform global foundries, but.

Only to find, uh, a decade later that the center of gravity of manufacturing and competency and expertise move to this little island in Taiwan is ai, which clearly one could argue that US is, has been way ahead relative to many, many countries in all aspects of the AI technology. Only to find that somebody in China is building out a model which is supremely more efficient.

From a effectiveness from cost and all of that. Do you worry about what has happened in the semiconductor space from a manufacturing standpoint? Just is a portends what might happen in the AI landscape where us loses its advantage? Look, it's a

Bobby Yerramilli-Rao: ary lesson. I. And I think people should learn from that.

Right now, the evidence is that nobody has created the leading frontier model outside of the us. Mm. So that hasn't happened. A lot of fast followers have taken, approached those levels of performance. Uh, certainly cost reduced, created different variants. All of that stuff has happened, but nobody has yet created a leading frontier model outside of the us but isn't the giant equalizer.

Sriram Viswanathan: But, but we, but we can't take it for granted. True. But isn't the giant equalizer just the availability of GPUs from Nvidia? And if that was, you know, let's assume that that is gonna continue and everybody was able to have access to it. Does it reduce the time it might take for someone to catch up talent?

Bobby Yerramilli-Rao: Exists all over the world. The US does a great job of attracting talent from all over the world, which many countries are not as good as as the US in doing that, uh, let's hope that's able to continue because there is going to be a talent premium on top of the GPUs.

Sriram Viswanathan: Well, some of the other areas that you've been quite active in.

You know, some of your personal investments and interests seem to be in the genomics and biology or life sciences space. Can you, can you talk about what your interest is in that area and what do you see as some of the big opportunities? Yeah. I'm sort

Bobby Yerramilli-Rao: of an accidental co-founder of a, a couple of companies in that space because I have a friend who's a scientist who's invented a bunch of things, including next generation sequencing.

So we started a couple of companies together and I really enjoy the, um, uh, the fact that it's actually quite different from. What I do as a day job. Mm. I'm on the board now of both the companies. This is, uh, genome Therapeutics. Yeah. Genome Therapeutics and bimodal. And what I like about both of them is that with bimodal, what we've got is a set of tools that, uh, measure, uh, aspects of the genome, uh, specifically the epigenome.

And it's just novel data that is, uh, you can't simulate, you have to be able to collect it de novo. It's dynamic data. It changes all the time because the epigenome is what regulates each one of your cells being different from. You know, skin cell from being different from, uh, brain cell, from being different from liver, it changes all the time based on your environment is very useful in diagnosing diseases, particularly cancer.

So there's a lot to be learned there in terms of understanding the value of data sets and how you can use that to, uh, get to great products, but also the value of tools, which effectively in, in our parts would be sensors. Mm-hmm. But these are tools that actually measure aspects of life sciences and, and genome therapeutics.

This is about. Developing drugs for cancer, completely different to anything that I've ever been involved in. And you just understand how you need to be very humble in the face of Mother Nature because you know, nature doesn't care about your AI and doesn't care about your technology. It just does it its own way and you just have to kind of, I.

Uh, submit to experimentation sometimes to really get good results.

Sriram Viswanathan: This is an area that we are at Celesta, are very actively involved in. You know, we've got lots of companies in using AI for cancer detection and, and radiology and mammograms and all of that using AI techniques and. And finding some very creative ways of diagnosing diseases and, and antibody development.

So this is an area that's very, very interesting for us as well. Is this an area that you think that Microsoft might get involved in at some point? We

Bobby Yerramilli-Rao: want to be partnered with companies that are using AI for, uh, life sciences and sciences in general. Uh, we are not gonna get into that space ourselves as a business, you mean?

No, I mean, we don't, we don't want to get into those, we're not competing with our customers on that.

Sriram Viswanathan: But one of the AR areas that you have gotten involved in, I think through one of the companies that you are personally involved in is a company called Wave. It's the autonomous driving startup and Microsoft invested in it.

Do you, do you see you guys, uh, doing more in that space? Yeah, I mean,

Bobby Yerramilli-Rao: I've known Alex Kendall, uh, since he was a PhD student because his PhD advisor was a, is a good friend of mine. So I've known him, uh, much before I've, uh, I was at Microsoft. And he's just a, a generational founder, a remarkable, uh, individual who's created a end-to-end machine learning system for self-driving cars that's doing remarkably well.

So we are, uh, an investor, but more importantly we're, we're the partner around cloud computing and, uh, we're just delighted with the progress that

Sriram Viswanathan: he's making with this is not something that, uh, we would see a Microsoft branded Robotaxis at any point in future. I don't think that's on the cards. No. Well, this has been a, you know, a fascinating conversation, Bobby.

I think I've, uh, I've always enjoyed the breadth of things that you're involved in and your interest from genomics to, uh, to autonomous driving to ai, to cloud, to gaming, to a bunch of things. And you probably have one of the most, uh, most fun jobs, uh, working for a company as successful as Microsoft is.

So, thank you really for, uh, taking the time to talk to me on this, and I can keep going on for. Another hour drilling into all of these topics, but really appreciate your, your time with us. Always a pleasure to spend time with you, Graham. Thank you.

Thank you for tuning in to the Tech Search podcast from Celesta Capital. If you enjoy this episode, feel free to share it, subscribe or leave us. A review on your favorite podcast platform. We'll be back every two weeks with more insights and discussions of all things deep tech. Bye for now.