00:00You've watched this India startup ecosystem for years.
00:04How has the landscape changed in the past few years?
00:07And what have you seen in terms of investment opportunities,
00:10particularly in artificial intelligence?
00:15My thanks to be here, Shari, and thanks for the time.
00:19Yeah, I think about maybe three to four years ago
00:22when AI companies began to come to us and show up in our deal flow,
00:26maybe about 20% of the companies were AI, 80% was non-AI.
00:31And within the AI category, I would say maybe 80%
00:34were more of the infrastructuring businesses.
00:37We made an investment.
00:39We were the first investor in Servum.
00:41We made an investment in a company called Composio.
00:44We made an investment in a company called Portkey,
00:47which recently got acquired by Palo Alto in the U.S.
00:50So these were more infrastructuring companies.
00:53They were kind of closer to the metal.
00:55In the last year or two, I think the ratio has flipped.
00:59A, from 20% of our deal flow in AI three, four years back,
01:03now about 80, 90% of our deal flow is AI.
01:06So majority of the founders that come to us and knock on our doors
01:10are building something in or around AI.
01:13And of that 80%, I would say...
01:17Go ahead.
01:21And of the 80%, I would say maybe the majority of the companies
01:24are now building at the application layer,
01:26which is where a lot of the momentum, a lot of the excitement,
01:29a lot of the value has also shifted up.
01:35So let's talk a little bit about the AI sector
01:37and the narratives around it,
01:39because they seem to be flipping fast.
01:42I mean, remember when software was all the rage
01:45and so much money went into the sector in the past few years,
01:48all of a sudden, generative AI potentially making this obsolete
01:52and you saw the big sell-off.
01:54What's your criteria in a sector that continually changes so fast?
02:02I think you cannot invest in a category
02:06that you don't understand as an investor.
02:08So all of us at Lightspeed are very big AI native investors.
02:13We are using AI tools every single day in our own business.
02:17I still write a lot of code.
02:20We stay ahead of the research segment.
02:23We speak to a lot of researchers all the time.
02:26I used to be at AMD.
02:27We actually designed a lot of the chips, the GPUs and the CPUs
02:30that are actually powering a lot of this.
02:32So, you know, we are fairly comfortable as a firm
02:35all the way to the bottom of the layers
02:36where power, energy, data centers, chips come
02:40and all the way to the top of the layer
02:41where applications and services sit.
02:43So as investors, you've got to be really cutting edge
02:45on your own and stay ahead of the curve
02:48to be able to understand what's really happening
02:49at the frontier and invest in the frontier.
02:54As a firm, Lightspeed has also invested in Anthropic,
02:57not only Sarvam, right?
02:58So you actually have seen both sides
03:00of the very different ecosystems.
03:03And you, as you mentioned,
03:05you've also worked at these different Western companies.
03:08What is the differentiator there?
03:11And have you seen the AI landscape in India
03:14develop in any different way from the West?
03:20Yeah, so India has some very unique challenges.
03:22So you're right.
03:23We are pretty much everywhere in the world
03:25where there's entrepreneurship.
03:25We are investors in Mistral in Europe,
03:27which is also building a sovereign playbook for Europe.
03:30We're investors in reflection AI in the US,
03:33which is building a sovereign playbook for the US.
03:36Ananthropic, which is a lot more commercial
03:38and global in nature.
03:39And we are investors in Sarvam,
03:41which is building the sovereign playbook for India,
03:43but also going to go more broadly about that.
03:47I think India has some very unique challenges.
03:50Number one, India is a 2,500 to 3,000 per capita GDP country.
03:54And it is really hard to service this country at scale
03:58by building software and solutions
04:00in a 40,000, 50,000 per capita GDP country.
04:04So you have to build something locally
04:06to be able to serve something locally.
04:08And it's been, you know,
04:09I think that's been the bottleneck in India.
04:11Second, India is a very, you know,
04:13service-focused country.
04:16People are not as sophisticated to use software.
04:18So you need to almost walk them along.
04:21You need a lot of hand-holding.
04:23It actually is really good at that.
04:24So it's an opportunity,
04:25but it does require you to build custom solutions
04:28quite a lot to really push that
04:29into the Indian market.
04:34Third, price points have to be lower in India,
04:36given the GDP per capita is low.
04:38So as you know, OpenAI's, you know,
04:41lower tier plan is much cheaper in India
04:46than in the US.
04:47And the same is true for Cursor.
04:48Cursor just launched a $7 per month plan in India,
04:51which I think is the lowest in the world,
04:53even though India has the highest number of power users
04:56anywhere in the world.
04:57So India has to be serviced at scale
04:59at a lower price point.
05:00And that requires some very fundamental
05:02bottom-up thinking,
05:05which companies have to do.
05:06As an example, Sarvam is building
05:09a very voice-first AI playbook for India
05:13because India speaks dozens of languages.
05:16In fact, there are hundreds of languages and dialects,
05:18but there are 25 languages that are very, very primary.
05:21And they are building a separate tokenizer
05:23specifically on that.
05:24And those kind of things need to be thought
05:26bottom-up to service India,
05:28and they cannot be bolted on
05:29on top of an existing solution.
05:32Super quickly, Sarvam AI, very unique.
05:35You don't see many unicorns out there in India.
05:37What does it take to scale to a global tech giant?
05:44Yeah, so we've been very lucky
05:45to partner with them from day one.
05:47We have actually two unicorns.
05:48One is Emergent, and one is Sarvam.
05:50Sarvam was a three-year-old investment,
05:53and Emergent was a one-year-old investment.
05:55But the good way to separate the two is
05:58Sarvam is much more infrastructure
05:59and foundation model focused
06:01and goes up from there.
06:02Emergent sits at the application layer,
06:04which is what I was saying earlier.
06:05We are seeing a lot more application layer companies
06:07coming up in our deep flow now.
06:09I think, look, Sarvam's first and foremost,
06:12our primary goal is to service India at scale
06:14for India-specific problems at India scale,
06:18at population scale.
06:19So we're going to tackle both the sovereign side,
06:22where we have a strategy from central government top-down
06:28and state government bottom-up.
06:29So we are going to capture that entire ecosystem
06:33for civil and sovereign use cases, including defense.
06:36On the enterprise side,
06:38which is where a majority of our revenue is right now,
06:40we've got to compete with the global best.
06:42We've got to build the best models
06:44in performance, cost, and usability,
06:47and we are working on that.
06:48We're going to launch a 1.0 parameter model
06:51in the next six months,
06:53and then that will be a very competitive model
06:55anywhere globally,
06:56and then we're going to go above from there.
06:58And if the models are really competitive,
06:59they can be serviced out of India
07:00for any part of the world,
07:02like China is doing really well these days.
Comments