June 20, 2024

Zoom hits an all-time low. Here's what AI startups can learn from the WFH hype cycle.

Zoom was THE work-from-home stock. It's down 90% from peak. The same hype that fuelled work-from-home stocks is now fuelling AI. What happened to Zoom will happen to many AI companies. AI startups today need to carefully position against foundational models and incumbents. Here's how to think through it. Why you should listen: Learn why what happened to Zoom will happen to many AI companiesHow to position yourself against the two AI giants: Foundational models & incumbents.Ho...

Why you should listen:

Learn why what happened to Zoom will happen to many AI companies
How to position yourself against the two AI giants: Foundational models & incumbents.
How to think through the product vs distribution race all AI startups face.

Keywords
Zoom, AI, positioning, foundational models, incumbents, use cases

Send me a message to let me know what you think!

00:00 - Lessons From Zoom and AI Hype

02:24 - Positioning in AI

08:09 - Product vs. Distribution

WEBVTT

00:00:00.179 --> 00:00:07.841
So I got this alert today on my Yahoo stock app, which is the only thing that Yahoo is used for these days.

00:00:07.841 --> 00:00:17.928
Anyways, I get this alert that Zoom hit an all time low, which is pretty wild because Zoom is like the I remember, actually this February 2020.

00:00:17.928 --> 00:00:23.227
The pandemic February or March 2020, right, like the pandemic was already clearly happening.

00:00:23.227 --> 00:00:44.167
It was either I think it hadn't yet made its way really in a big way to like Canada and the US, north America, still mainly in Asia, and I remember a friend calling me and talking to me about you know, this pandemic thing is, you know, probably going to come, it's probably going to happen here, and his question was how to play it?

00:00:44.167 --> 00:00:49.604
What stocks could you buy to be well positioned from what was about to come?

00:00:49.604 --> 00:00:53.454
We talked back and forth and the obvious contender was Zoom.

00:00:53.454 --> 00:01:01.121
Zoom was this pure play on the pandemic, because it was the product that everybody was going to have to use.

00:01:01.121 --> 00:01:06.361
Now, at the time, zoom's shares had already gone up from $70 to about $90 a share.

00:01:06.683 --> 00:01:09.891
Frankly, I didn't buy Zoom because I thought, well, it's probably already priced in.

00:01:09.891 --> 00:01:12.990
Now it turns out I was not just wrong, but extremely wrong.

00:01:12.990 --> 00:01:20.611
That thing went up from $90, let's call it $100 a share to like $550 a share at peak six months later.

00:01:20.611 --> 00:01:27.923
So I mean in six months you could have five extra money.

00:01:27.923 --> 00:01:28.725
We actually had the discussion.

00:01:28.725 --> 00:01:29.468
All I had to do was press buy.

00:01:29.468 --> 00:01:31.013
That's the difference between talking about it and actually doing something.

00:01:31.013 --> 00:01:32.037
But my point is it's come all the way back down.

00:01:32.037 --> 00:01:45.980
It's crashed 90% and now at an all time low the lowest it's ever been and so this really marks, in a way, the end of all of the accelerated stop buying and demand that was generated as a result of the pandemic.

00:01:46.540 --> 00:01:52.563
As everybody knows, a lot of tech products saw accelerated revenue growths because of the pandemic.

00:01:52.563 --> 00:01:59.548
A lot of that pulled back and I just thought this really marked kind of the beginning and the end, because Zoom was such was like the pandemic stock.

00:01:59.548 --> 00:02:03.325
But it also made me think about what's happening now with AI.

00:02:03.325 --> 00:02:10.265
With AI we're seeing a pretty similar hype cycle in the sense that there's a lot of accelerated spending today on AI.

00:02:10.265 --> 00:02:12.427
Nvidia's revenue is going through the roof.

00:02:12.427 --> 00:02:15.770
It's hitting you know the stock is hitting like a $3 trillion market cap.

00:02:15.770 --> 00:02:17.366
It's now one of the biggest companies in the world.

00:02:17.366 --> 00:02:23.591
Chad GBT just last week announced $3.5 billion in ARR up from effectively zero two years before.

00:02:23.591 --> 00:02:31.564
So this is not just eyeballs, this is real revenue, but it's happening at an extremely fast pace.

00:02:31.585 --> 00:02:38.811
But what I wanted to think about was what can you learn from what happened with Zoom and some of those pandemic stocks to what might happen now with AI, especially from the perspective of an early stage founder?

00:02:38.811 --> 00:02:50.479
And the thing that I go back to is positioning and this is something that we're thinking about as investors into AI companies is how any given opportunity is positioned relative to what we think of as the two monsters.

00:02:50.479 --> 00:03:00.009
The two monsters in the space are, on the one hand, foundational models, so open AI, anthropic cohere, the open source models like Lama, mistral and so on.

00:03:00.009 --> 00:03:01.013
That's one monster.

00:03:01.013 --> 00:03:09.712
The other monster in space, like in any space, really are the incumbents, and startups have to find a way to position against both of those.

00:03:09.712 --> 00:03:31.444
In the pandemic world, you just had the one monster, which is the classic monster, which is the incumbents and, by the way, that's really what happened to Zoom Partially is that demand in general for a lot of these video telecommunication apps decreased from the absolute peak, the absolute height, but demand overall definitely lifted, and if you look at your average life, you do way more video calls now than you used to pre-pandemic.

00:03:31.806 --> 00:03:35.313
The problem is Google's there, microsoft is there.

00:03:35.313 --> 00:03:57.503
If you're a Microsoft enterprise, you're for sure using Teams, and so Zoom might be a better product, but because it's a paid product and because it's mainly B2B buyers, the fact that it's the best doesn't necessarily mean it's going to get the most market share, because sometimes what matters more than product quality is distribution, and in that both Microsoft and Google are leaps and bounds ahead.

00:03:57.503 --> 00:03:59.550
So how does that translate over to AI?

00:03:59.550 --> 00:04:04.788
Well, if you think about it, I think there's two things you have to think about as a founder founder, if you're building in AI today.

00:04:04.788 --> 00:04:06.912
The first one, like I said, is the foundational models.

00:04:06.931 --> 00:04:07.554
I'll tell you a story.

00:04:07.554 --> 00:04:10.594
I was speaking with actually a handful of companies, but I'll tell you about one.

00:04:10.594 --> 00:04:18.987
I was speaking with this company and this is before ChatGPT 4.0, when ChatGPT's voice was not all that great.

00:04:18.987 --> 00:04:25.473
You'd speak to it and it would take 30 seconds or so to kind of get back to you, and they were building AI for call centers.

00:04:25.473 --> 00:04:53.819
They wanted to help call centers use AI to automatically answer level one calls, so the simplest call, so they would answer the phone and they would talk to the person on the other end, and only if the call was complicated enough then they might pass it on to a human agent, and so the problem was, at the time, none of the foundational models and even in Levin-Lauz there's a little bit adjacent to some of these foundational models was not good enough in terms of latency, which was one of the biggest problems.

00:04:53.819 --> 00:04:59.439
So this company was spending a lot of time fixing latency, and that was, frankly, the most compelling part of the demo.

00:04:59.439 --> 00:05:11.540
You could call this number and you could speak to their AI, and I noticed firsthand how much better it was at voice than what I was used to in my ChatGPT app.

00:05:11.750 --> 00:05:26.939
The problem is a handful of weeks later, two, three, four weeks later, chatgpt 4.0 comes out and all of a sudden, 100% of what this startup had focused on all of their IP, all of their work on latency was redundant.

00:05:26.939 --> 00:05:36.437
The value of that work plummeted to zero, and so that's lesson number one, which is you've got foundational models that are only going to keep getting better.

00:05:36.437 --> 00:05:39.098
They've got the best researchers in the world.

00:05:39.098 --> 00:05:46.264
They've got hundreds of millions, billions, in some cases tens of billions of dollars developing their foundational model.

00:05:46.264 --> 00:05:48.730
They're working across all modals.

00:05:48.730 --> 00:05:54.437
They're working in a multimodal world Text, voice, images, video.

00:05:54.437 --> 00:06:12.019
All of that stuff is going to get better, and so you have to ask yourself are you working on pieces of the puzzle that open AI could be coherent, could be anthropic, could be minstrel, doesn't matter but are you working on pieces of the puzzle that these foundational models are at some point going to tackle?

00:06:12.019 --> 00:06:13.382
The answer is yes.

00:06:13.382 --> 00:06:14.634
You're wasting your time.

00:06:14.694 --> 00:06:48.817
The historical analogy for what it's worth is if you think about microchips back in the day, when Intel was kind of putting out these chips that kept getting better and better and Moore's law was in full effect, you could either develop an app for today's chips and spend a lot of time optimizing and making sure all your features work on today's chips, or you could just assume tomorrow's chips would have more memory, more compute and just get progressively better and build for those, and of course, the ones that build for the next gen ultimately won because they didn't have to waste time optimizing for something that was about to become obsolete.

00:06:48.817 --> 00:07:05.932
The places where you really want to develop our workflows, our integrations, are parts of the feature set that are very use case, specific Things that obviously the foundational models aren't going to spend time doing and I'm not talking about being a GPT wrapper, right, originally it was the GPT would come out and they didn't have, for example, the ability to import PDF.

00:07:05.932 --> 00:07:08.276
So people build these PDF summarizer tools.

00:07:08.276 --> 00:07:22.052
Well, obviously, the foundational models, especially because of ChatGPT specifically, which is direct to consumer, we're going to add that sort of horizontal functionality, the sort of functionality that's going to be useful for so many different use cases.

00:07:22.052 --> 00:07:27.673
So you have to think about the sort of things that are too specific, too vertically deep.

00:07:27.673 --> 00:07:36.252
Again, workflows and integrations are a great thing to look at, because those are typically places that these foundational models aren't going to spend their time on.

00:07:36.612 --> 00:07:43.660
And I think the key question you have to ask yourself as a founder is am I excited or am I scared for the next GPT update?

00:07:43.660 --> 00:08:00.684
So when Sam Altman goes on stage and he's announcing GPT-5, are you excited because you can't wait for the features and how they're going to make your product better, or are you scared because you're not sure if something that he's going to announce is going to make a bunch of your IP completely useless?

00:08:00.684 --> 00:08:01.769
So that's the first thing.

00:08:01.769 --> 00:08:04.317
I think the second thing is versus incumbents.

00:08:04.317 --> 00:08:10.930
Right, that's the other monster in the room, and I think that one's a little bit more common, in the sense that that's always been the case.

00:08:10.930 --> 00:08:15.079
Startups always have to think about where are incumbents playing.

00:08:15.079 --> 00:08:26.262
I think that with AI, because it's such a new technology, incumbents in general take longer to put these features out, and a lot of startups I find are thinking hard enough about where incumbents are going to go.

00:08:26.689 --> 00:08:28.634
Perplexity, I think, is a great case in point.

00:08:28.634 --> 00:08:41.356
I love the product, I use it all the time, but the reality is they're going directly after search and the original premise was well, google is never going to do it because they make all their money on ads and they're not going to disrupt themselves.

00:08:41.356 --> 00:08:50.383
The problem is we're talking about companies read and understood the innovators dilemma, and so they know what disruption is and they're not going to let startups have their cake.

00:08:50.383 --> 00:08:55.245
They're going to have their cake, even if it means disrupting themselves, and that's why Google ultimately put out a product.

00:08:55.245 --> 00:08:57.248
That is exactly what Perplexity does.

00:08:57.248 --> 00:09:07.720
And, yeah, right now it's really funny to laugh on Twitter about how bad it is, about how it tells you to put glue on pizza and eat rocks, but the reality is, the question is it's a product versus distribution race.

00:09:07.720 --> 00:09:08.902
So who's going to win?

00:09:08.902 --> 00:09:15.841
Perplexity today has 22 million visits a month and Google has 86 billion visits a month.

00:09:15.841 --> 00:09:24.648
So is Google going to fix its product faster than Perplexity can add another 85.9 billion monthly visits, probably?

00:09:24.648 --> 00:09:26.053
Is Perplexity going to go to zero?

00:09:26.053 --> 00:09:26.977
No, I'm not saying that.

00:09:26.977 --> 00:09:31.923
They've built a billion dollar company and they'll probably get acquired, but it's not a strong position to be in.

00:09:32.307 --> 00:09:37.159
So the first thing you need to think about is building something that's not a number one priority for the incumbent in your space.

00:09:37.159 --> 00:09:45.794
So, as a startup, you need to think about this kind of product versus distribution race and whether you have a good chance of realistically winning that.

00:09:45.794 --> 00:09:54.361
The incumbents always have better distribution than you and you usually have a better product, usually ahead on product, at least on the AI side of the product.

00:09:54.361 --> 00:09:59.270
The question is are you going to be able to get distribution faster than they can catch up on product?

00:09:59.270 --> 00:10:03.261
That's the question that you need to think about and it's very case specific.

00:10:03.261 --> 00:10:05.434
But you've got to look at who are my incumbents first of all.

00:10:05.434 --> 00:10:10.336
Is it Google, is it Amazon, or is it some legacy player that doesn't know how to spell the word tech?

00:10:10.336 --> 00:10:11.961
That's one of the first questions.

00:10:11.961 --> 00:10:21.888
The other question you got to look at is is there one huge incumbent with incredible market share, or is it spread, in which case maybe you could partner with a bunch of different ones and gain power that way?

00:10:21.888 --> 00:10:33.619
The last thing you need to think about is am I going after one of the most important use cases for my incumbents and where they're likely to spend all their time, or am I going after things that they're more likely to ignore?

00:10:33.619 --> 00:10:36.264
So those are the two monsters and you just you can't ignore them.

00:10:36.264 --> 00:10:41.860
You've got the foundational models, you've got the incumbents, and you've got to find your wedge between the two.

00:10:43.471 --> 00:10:45.416
I just gave you content that you liked so much.

00:10:45.416 --> 00:10:47.822
You actually listened to the end and guess what?

00:10:47.822 --> 00:10:50.053
You didn't pay a single dollar.

00:10:50.053 --> 00:10:53.279
Not only that, I didn't even put any ads in your face.

00:10:53.279 --> 00:10:59.520
So you just got a bunch of content for free, and now that I've delivered that value, I'm asking for something in return.

00:10:59.520 --> 00:11:06.216
Open your app, open Apple Podcasts, open Spotify, open whatever app you use to listen to this and hit that follow button.

00:11:06.216 --> 00:11:11.917
It's actually going to help you, because it's going to help you make sure you don't miss out on the next episode, which you like so much that you listen to the whole thing.