The AI Daily Brief: Artificial Intelligence News and Analysis - Who Will Create and Capture the Most Value in AI?

Episode Date: September 3, 2023

An exploration of defensibility, value capture and economic structure in artificial intelligence. With excerpts from: https://www.wsj.com/articles/ai-has-finally-become-transformative-humans-scale-language-model-6a67e641 https://a16z.com/2023/08/03/the-economic-case-for-generative-ai-and-foundation-models/ https://a16z.com/2023/01/19/who-owns-the-generative-ai-platform/ https://a16z.com/2023/08/11/cloud-lessons-for-the-ai-era/ https://a16z.com/2023/08/30/financial-opportunity-of-ai/ ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're asking which companies will create and capture the most value in generative AI. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network for more information about our newsletter, our Discord, and our YouTube channel. Welcome back to the AI breakdown. For this long holiday weekend, we are doing another version of a long read today, although this one is actually a combination of not one, not two, not three, not four, but five different pieces all from Andresen Horowitz. Now, there has been a theme running throughout a lot of this venture firm's explorations that has to do with who is going to win the value of artificial intelligence, at least

when it comes to an economic or market perspective. I think it's a really interesting question, and it's certainly one that shapes the way the industry will evolve. What I mean by that is that if everyone thinks that applications are going to capture all the value, well, then you're going to see a lot of funding heading towards applications. If, on the other hand, there is a large sense, that it's big enterprises where the value is going to accrue? Well, that has entirely other economic implications. But to understand and get into this topic, let's start a little bit higher level. In the Wall Street Journal at the beginning of August, Martin Casado from A16Z wrote a piece called AI has finally become transformative. After a decade's worth of innovation, new models can

change the world the way the internet did. Now, as you can tell from that title and the subheader, the stakes are being laid out pretty clearly. This is not just another hype cycle. This is the real deal, and the comparisons to the internet are deserved. Now, the piece points out initially that the vast majority of the gains in machine learning or artificial intelligence so far have gone to the tech giants, including meta and Google. That's, of course, because most of the economic value of those efforts have come through better ad targeting, better recommendations for social. Indeed, they say, despite AI's enormous capabilities, the economic realities for using it haven't been great for startups. Often the amount of value a company gets from AI diminishes quickly over time and

therefore requires significant continuing investment. As a result, they write, AI's primary value has been to improve existing operations for incumbents who have the resources to invest at the required levels. However, they say this is changing. Fortunately, this doesn't seem to be the case with the current wave of generative AI applications, such as ChatGPT. While still very early, we're already seeing use cases in large existing markets with orders of magnitude improvement in time, cost, and performance. This has led to some of the fastest growing technology and product adoption in the history of the software industry. We may be experiencing what is likely the start of a new super cycle on par with the advent of the microchip or the internet.

They point to a couple reasons why things are different right now. One, they point out, despite all of the hype around hallucination and things like that, as they point out, quote, accuracy isn't that important for many applications. The example they give is text to image generation where what matters is entertaining the user, not being perfect. When it comes to things like helping developers write code, the user is, as they put it, the human in the loop. Another reason this time is different, they say, is that in many ways, this new set of Gen AI startups are dealing with experiences that are totally new,

interactions with computers that are completely novel. That means they say, we don't really have a good understanding of what the behaviors will lead to, nor what the best products are to fulfill them. Amazingly, while the use cases for these new behaviors are still emerging, millions of users have already shown a willingness to pay. Net net, this all means opportunity for the new class of generative AI startups to evolve along with users, while incumbents focus on applying the technology to their existing cash cow business lines. Okay, so two big things to take away from this. One, AI is a very big deal and generative AI represents a fundamental shift in even that space that leads to something that looks more like a secular shift in the actual computing era.

And two, startups are actually fairly well poised to create value in that because of their inherent ability to go explore use cases that haven't been imagined yet. Now, A16Z expanded upon this thesis in a larger piece that they published on their blog called the Economic Case for Gen. generative AI and foundation models. Since I think it's very salient to the point that we're discussing in this particular show, let's read their expanded section on this idea that startups can go explore new behaviors. They write, this is a very yet underappreciated point. Likely as a result of AI largely being a complement to existing products from incumbents, it has not introduced many new use cases that have translated into new behaviors across the broader consumer population.

Remember here, they are talking about machine learning type products pre-chatcheeBT, not this new generation of generative AI startups. Continuing, they write, new user behaviors tend to underlie massive market shifts because they often start as fringe secular movements the incumbents don't understand or don't care about. This is fertile ground for startups to cater to emergent consumer needs without having to compete against entrenched incumbents in their core areas of focus. Now, compare this now to what has happened with generative AI. They write, the new user behaviors that have emerged with the generative AI wave are as startling as the economics have been. LLMs have been pulled into service as software development partners,

brainstorming companions, educators, life coaches, friends, and yes, even lovers. Large image models have become central to new communities built entirely around the creation of fanciful new content or the development of AI art therapy to treat use cases such as mental health issues. There are functions that computers have not to date been able to fulfill so we don't really have a good understanding of what the behavior will lead to, nor what are the best products to fulfill them. This all means opportunity for the new class of private genitive AI companies that are emerging. They also reinforce that users have already shown a willingness to pay for these tools, making the economics of them look even better.

So they write, are we just fueling another hype bubble that fails to deliver? We don't think so. Just like the microchip brought the marginal cost of compute to zero, and the internet brought the marginal cost of distribution to zero. Generative AI promises to bring the marginal cost of creation to zero. So again, the takeaway for this, when it comes to the question that we are exploring of where value will accrue in and around artificial intelligence, one big takeaway is the idea that generative AI is unlocking a fundamentally new set of, A, relationships between people and computers, and B, human behaviors that result, that tends to be really fertile ground for new companies, i.e. startups, as opposed to incumbents. So if the question is where value accrues

in the space, put at least one notch in the startup column. Now, another piece from A16Z's blog was called Who Owns the Generative AI Platform. They write, there is enough early data to suggest massive transformation is taking place. What we don't know and what has now become the critical question is, where in this market will value a crew? Here's their sum up of what they think. We've observed that infrastructure vendors are likely the biggest winners in this market so far, capturing the majority of dollars flowing through the stack. Application companies are growing top-line revenues very quickly, but often struggle with retention, product differentiation, and gross margins. And most model providers, though responsible for the very existence of this market,

haven't yet achieved large commercial scale. In other words, the companies creating the most value, i.e. training generative AI models and applying them in new apps haven't captured most of it. Predicting what will happen next is much harder, but we think the key thing to understand is which parts of the stack are truly differentiated and defensible. This will have a major impact on market structure and the drivers of long-term value. So far, we've had a hard time finding structural defensibility anywhere in the stack outside of traditional modes for incumbents. Now, one of the things that I think that is easiest to disagree with here, but to be fair, might be a little bit of hindsight bias, is their assertion that, quote, model providers invented

genitive AI but haven't reached large commercial scale. They write, what we now call generative AI wouldn't exist without the brilliant research and engineering work done at places like Google OpenAI and stability. Yet the revenue associated with these companies is still relatively small compared to the usage and buzz. In image generation, stable diffusion has seen explosive community growth supported by an ecosystem of user interfaces, hosted offerings and fine-tuning methods, but stability gives their major checkpoints away for free as a core tenet of their business. In natural language, models OpenAI dominates with GPT3 and 3.3 and 3.5 and chat GPT, but relatively few killer apps built on OpenAI exist so far and prices have already dropped once. This may be just a temporary

phenomenon. Open AI, for example, they write, has the potential to become a massive business earning a significant portion of all NLP category revenues as more killer apps are built, especially if their integration into Microsoft's product portfolio goes smoothly. Well, obviously, if you are listening to the AI breakdown with any frequent, with any regularity, you will know that as of the last week or so, OpenAI is generating a reported $80 million a month and is on track to generate a billion dollars over the next 12 months, and that's before revenue from the just-released ChatGPT Enterprise features are included and incorporated into that. Meanwhile, when it comes to that OpenAI tie-up with Microsoft, it seems less clear cut than it ever

was before. A couple weeks ago, Microsoft announced plans to integrate with Databricks, which is a service that effectively allows enterprises to spin up their own versions of ChatGPT and GPT4. customizing existing open source models for something controlled by the enterprise end-to-end. Now, these two things alone show just how much the space is evolving and how quickly. OpenAI is actually capturing a significant amount of value pretty quickly out of the gate, especially before they've launched their major money-making feature in chat GPT enterprise. Secondarily, however, their biggest partner in Microsoft isn't convinced that a third-party solution like ChatGPT Enterprise offers is what enterprises will ultimately want and demand.

Instead, it seems they're taking a route that's more diversified and more akin to something like Amazon Web Services Bedrock, which is giving enterprises more choice to customize existing open source models to something that they control and where their data doesn't leave and flow into a third-party cloud service. Still, even with those caveats aside, I think it's worth coming back to the conclusion of this piece from A16Z, in a section they call so dot-da-dot where will value a crew. They say, of course, we don't know yet, but based on the early data we have, our intuition is the following. There don't appear today to be any systemic moats in generative AI. As a first-order

approximation, applications lack strong product differentiation because they use similar models. Models face unclear long-term differentiation because they are trained on similar data sets with similar architectures. Cloud providers lack deep technical differentiation because they run the same GPUs, and even the hardware companies manufacture their chips at the same fabs. There are, of course, the standard modes, scale modes, I have or can raise more money than you. Supply chain modes, I have the GPUs you don't. Ecosystem moats, everyone uses. my software already, algorithmic modes were more clever than you, distribution modes, I already have a sales team and more customers than you, and data pipeline modes I've crawled more of the

internet than you, but none of these modes tend to be durable over the long term, and it's too early to tell if strong direct network effects are taking hold in any layer of the stack. Based on the available data, they say, it's just not clear if there will be a long-term winner-take-all dynamic in generative AI. Now, this they say is a good thing, and I tend to agree, and what's notable is that even though their assessment of the ability, of OpenAI in particular to capture value might look a little off just a couple months after this was published. I think that broadly speaking, their assessment of the lack of moats, or at least the long-term weakness of these moats, is still pretty true. Now, another question or another way to look at

this question might be to look at precedence. Luckily, in another essay, that's exactly what A16 did. That piece is called the race to capture value cloud lessons for the AI era. Their question simply when a platform shift reshapes the technology landscape, who wins? The first thing that they note is, quote, despite popular David versus Goliath's zero-sum disruption narratives, platform shifts are usually positive some games, as technology advances the size of the pie grows for both startups and incumbents. The example they give, since the NASDAQ bottomed out in 2003,

aggregate revenue across public B2B software companies has grown from $99 billion to $587 billion, a 5.9x increase in new annual revenue generation produced by both incumbents startups. Software incumbents have grown from generating $99 billion to $323 billion, maintaining 55% of market share. Or put another way, incumbents have added revenue but lost 45% of the market to new entrance. And two decades into SaaS, there's still plenty of value left to capture. Morgan Stanley recently estimated cloud workloads to be only 29% penetrated in the enterprise. Now going back to the previous piece, where they divided things up between the application layer, the infrastructure layer, and the foundation layer or model layer, while they're a little bit wary to draw direct comparisons,

do say, while we can't derive a natural law of software physics from a single platform shift, in the SaaS era, startups one more market share at the application than infrastructure layer. Now, this kind of makes sense. If we go back to the idea that we shared before, that one of the things that startups are going to do well is follow their noses to new types of behavior, to new types of human behaviors and experiences, that stuff tends to happen at the infrastructure, at the application layer, rather than at the infrastructure layer. Infrastructure requires more capital, bigger scale, and potentially more trust with enterprise customers than startups or want to have. Now, another thing they note, however, that makes the SaaS comparison potentially

less relevant is the sense that incumbents are adapting faster in the AI era. This time around, incumbents have ostensibly been faster to adapt and launch applications than they were in the SaaS era. ChatGPT launched in November 2022, and the vast majority of big SaaS incumbents have since shipped their own implementations of generative AI products. They also point out that just generally speaking, the adoption of AI is happening much more quickly than the adoption of SaaS. And what about this infrastructure side of things? Well, They write, when infrastructure innovation does occur, such as with the rise of big cloud platforms, it creates new layers of the software stack and unlocks tremendous value.

Amazon Web Services launched in 2006, Microsoft Azure 2010, and Google Cloud Platform 2013, now account for 173 billion of run rate revenue, 30% of all public software revenue. With increasing economies of scale, the cloud computing platforms have come to dominate in the order in which they came to market. AWS at 33%, Azure at 22%, and Google Cloud at 10%. In the AI era, they say, the same market forces are already emerging at the model layer, with OpenAI leading the pack in a competitive setup similar to the early years of AWS. This time around, AWS and the other big cloud players are the incumbents.

They see AI as an extension of their legacy cloud compute business. Given the high compute demands of running AI models, the current chip shortages and the capabilities these cloud providers have in providing their own chips or building out data center capacity on third-party chips, most have rolled out their own foundation models to compete at both the compute and model layers. Now, in this piece, they expand what they see as defensibility in the AI era. They call out commercial lock-in, i.e. commercial agreements that guarantee exclusive data access, data life cycle control, data scale effect, regulated data, i.e., for example, regulatory frameworks and government procurement processes that limit access to sensitive data. In the category of product

and technology, defensibility in the AI era includes product network effects, which is one we're familiar with, internally developed foundation models, i.e. having an actual model differentiation. and then something that they call legacy product workflow depth. The last category of defensibility they call go-to-market, which includes things like legacy customer bases, i.e. GitHub, and or a security posture and privacy concerns. Now, I think that last one is hugely important. One of my base cases is that incumbents have a better chance of winning at least enterprise

business when it comes to AI, because since questions of data privacy and security are so important, it feels likely that companies might be more willing to trust companies that they already trust with their data with this different use of their data, rather than having to add a whole new trusted partner in the form of someone like an open AI. The last piece in this theme, however, they called barbarians at the gate, the financial opportunity of AI. And again, the thing that they reinforce is how hard it is to differentiate between the conversation of where value will accrue with use cases that we understand now versus use cases that seem possible versus use cases we can't even imagine yet. They call these known-nones, known unknowns, known unknowns, and unknown unknowns. In the known-nones, they include sell software to incumbents, compete with, incumbents with generative AI at the core or buy incumbents and remake them with AI what a

generative AI KKR would be doing. Known unknowns are areas where it's not clear if there's going to be product market fit for certain categories of AI products that seem obvious or that lots of people are working on. Those include AI personal assistants, AI sales coaches, counterfeit monitoring, and graphics generation. In this, they show how the reduction of cost in a particular area opens up massive demand that didn't exist before. They write, Upwork and Fiverr both have extensive marketplaces for customers.

images and artwork, but it seems Mid Journey has demonstrably more revenue than both of their graphics categories combined. Why? Because $20 a month unlocks a tremendous amount of demand that simply did not exist at $500 an image. It's not always about cost, it's also about speed. A mid-jorney image takes less than 30 seconds to create, unlocking demand that was literally impossible with a human artist as a bottleneck no matter what the cost. Finally, they talk about unknown unknowns. Something by definition, they say, is unmodalable. There's no mental model for finding them except once it happens, it's clear. They conclude, nobody doubts that AI will change the world. At the limit, artificial intelligence may exceed the importance of the wheel, fire, and

electricity for the human species. But for the economic impact, the changes will likely be shepherded by incumbents and a new era of corporate raider for known-knowns and an exciting number of yet-to-be-imagined startups for known unknowns and unknown unknowns. The barbarians are at the gates. And I think this is the big theme that you're left with, looking at the corpus of all of these pieces, is almost that the question of where value will accrue has to be the big. divided to in the short term and ultimately. It's easier to map out where it might flow in the short term,

just based on the nature of who's buying what and what they need AI for right now. But there is clearly a strong sense, and perhaps this is to be expected from a venture capital firm like A16Z, that the real big opportunities are in the things that are far flung, that are unknown now and in fact unknowable because they have to do with fundamental shifts in human behavior that will come up from the interaction with this new technology. Of course, if you think that's where the big value is going to be. Being in the VC business makes a lot of sense. Anyways, guys, I think that this economic lens is an interesting angle on which to understand the artificial intelligence space. And even if you don't particularly care, it's worth

understanding about because it will shape how the field evolves. Though in some pure world, we might wish it not to be the case, startups follow capital. And where venture and other types of capital are flowing is going to shape what comes out of this entire AI space. Let me know what you think. Come join us on the AI breakdown Discord. This is a great conversation for that space. You can find a link at bit.ly slash AI breakdown. I hope you are having a great long holiday weekend. And until next time, peace.