Inflated Expectations: Why the AI Hype Cycle is Probably Ephemeral

Artificial intelligence has possibly been the hottest topic for the last few months and the conversation has gone into multiple directions as always. The primary conversation goes like "AI is going to take all our jobs, which is relevant but might be overblown. The other side is about AI making our lives productive and what the utility actually is, which is surely relevant, but is definitely overblown once again. Looking at the investor's side, however, it's quite tough to really make sense of it. For some reference, $127B has been invested just through venture capital in private companies, across 14,500 deals since the beginning of 2022. I wouldn't want to include 2021 because over $107B was invested just in that one year, which is also the single biggest outlier for private market valuations we might have ever seen. AI has also been a part of our daily lives and we're arguably going through the best phase of democratization of technology, where everybody can access it on one's personal devices. Amongst all these opinions, and the whole world exploring the true power of technology, there is one central question in a lot of our minds, "Is it a hype cycle?". The reason I'm writing this is not because it's not right to ask this question, it's one of the most relevant things to question, but the opinions exist in extremes. Is there really something between AI destroying our jobs once we reach a stage of artificial general intelligence (AGI) and AI being nothing but hype. Some people's favorite thing to say is, lo and behold "The same thing happened in 2001 during the dot-com bubble". How surprising and inspiring to say something that we've never heard, without even realizing how nascent the private markets were back then and the S&P 500 itself was 2.5-2.7 times smaller than what it is today. If I had a dollar for every time someone had that analogy, I'd be as rich as an AI startup's founder. Anyway, here I am trying my best to state why every new industry that looks to disrupt the status quo goes through a hype cycle and why we're seeing the valuations normalizing and investors realizing where the true potential for AI might lie. Also, I'll primarily be talking about private companies, venture capital, and growth equity investments.

THE RISE OF INVESTMENTS IN AI/ML:

The trend and the investments in startups leveraging AI and ML and operating in this vertical primarily saw a rise around 2018. As you can see, in 2014 and 2015, less than $7B were invested in the vertical across the globe. 2021 saw $110B being invested and 2022, despite the dip saw $79B coming into the space. In 2015, the world of AI experienced a seismic shift from science fiction to everyday reality. It was the year Amazon introduced the Echo, a home assistant powered by AI, receiving rave reviews. Microsoft and Google showcased software that could outperform humans in image recognition tasks, and Tesla's cars seemingly woke up one day with self-driving capabilities. Baidu's AI software even spoke fluently in both English and Chinese. The key ingredient behind these remarkable advances was machine learning, a technique that allows AI systems to learn from vast datasets rather than relying on explicit programming, and although ML was popular in the "circuit", it's relevance soared like anything and the application became even more widespread.

Machine learning's success hinged on two critical factors: copious amounts of data and powerful processing capabilities, primarily provided by graphics processing units (GPUs). The explosion of data from platforms like Facebook, Flickr, and YouTube, coupled with GPUs' ability to handle complex mathematical computations efficiently, propelled AI forward. GPUs excelled in the kind of math crucial to AI, making them a staple in AI research. This combination led to significant advancements in computer vision, with AI's accuracy in recognizing everyday objects soaring from 72% to 96% in just five years. Furthermore, the availability of GPUs in the cloud and innovations like NVIDIA's "Pascal" GPU made AI more accessible and efficient, solidifying its role as a transformative force in 2015 and beyond.

2021: THE YEAR IT ALL MADE SENSE (OR DID NOT):

As mentioned before, 2021 was an outlier beyond an outlier, and not in a good way by any means. There were surely some investments that turned out to be truly disrupting like Waymo, Databricks, Relativity, Anduril and so on, but a majority of them were just overvalued investments gone apeshit. It's not just about startups based around AI/ML, but in general as well. About the valuations particularly, private valuations tend to follow public markets and there is usually a lag, probably a few months. For example, S&P 500 doubled between the crash due to COVID 19 and the beginning of Q4-2021, NASDAQ went up by 2.2 times during the same time frame. Private markets followed the same path and the valuations kept rising through 2021 and even until the first half of 2022, meanwhile the pubic markets valuations started coming down in the H1-2022, clearly establishing a lag. In 2021 itself, the median deal size across VC and growth reached $1.8M, compared to $0.9M in 2018 and the median post-money valuation rose from $8.7M in 2018 to $16.7M in 2021. Regarding the specific segments, nearly one-third investments are in enterprise software (business/productivity) quite obviously, followed by financial software (10%) and automotive (5%), varying quite a bit. Even on a stage basis, early stage VC saw a median deal size of $1.5M at a median post-money valuation of $10.2M in 2021, significantly higher than what the situation was a few years back. For later stage, the deal size reached $11.5M in 2021, with the post-money valuation rising to $68M. Even the number of mega deals in the US that 2021 saw was unprecedented, which was 156, more than the cumulative mega deals that happened from 2015-2020. 

The exorbitant valuations that we saw, was not only about following the public markets valuations and investing carelessly, but also about what a breakthrough year 2021 was for AI. Among the standout innovations was the introduction of DALL-E by OpenAI, an extraordinary multimodal AI system capable of generating intricate and highly detailed images from textual descriptions. This revolutionary development underscored AI's creative potential and its capacity to not only produce visually compelling content but also manipulate and reimagine visual concepts in an impressively sophisticated manner, opening new horizons for creative applications across industries. There were some other substantial strides as well, particularly in the domains of computer vision and natural language processing (NLP). The U.S. Department of Homeland Security's success in utilizing AI to identify masked faces highlighted the practical applications of AI in security and surveillance. Meanwhile, technology giants such as Uber, Salesforce, Microsoft, and Google pushed the boundaries of NLP models, with AI algorithms consistently surpassing human performance on various language benchmarks. These remarkable achievements in NLP demonstrated the growing maturity of AI and unlocked new possibilities in language understanding, sentiment analysis, and automated content generation. Moreover, AI made significant strides across various domains beyond language and vision. The European Commission's proposal for stringent AI regulations marked a pivotal moment in ensuring the responsible deployment of AI technologies. Facebook's AI model's capability to predict complex drug combinations showcased AI's potential in drug discovery and healthcare. Additionally, Cerebras' introduction of an energy-efficient AI supercomputing processor addressed a critical challenge in AI—power consumption and efficiency. All of it added to the enormity of a year that we had, which unfortunately didn't continue for too long.

THE END OF THE LINE?:

2022 and 2023 took a pretty big hit, not only because the valuations had to come down, but it just feels pale compared to the frenzy that continued through the first half of 2022. However, it is not that uncommon and this year does not seem like a bad one, especially from a post-money valuation perspective. There are terrible things to look at, such as massive down-rounds that a lot of companies have to resort to, limited exit opportunities across all the sectors, rising interest rates pushing companies to be torn between taking on debt and having those down-rounds, and I'm sure that many such things. A lot of these companies that are claiming such high valuations, are often companies that are working on something superficial and easily replicable. An example would be the surge of companies building NLP models and companies having absolutely nothing unique in their data models and input layers, but adding a thin application layer with limited functionality and seeking outlandish valuations. I have seen a couple of examples where there are a couple of founders who use Open AI's open-source model and use that as an input layer to add a layer, which doesn't add a single iota of difference to anything out there. Who wants another Chat GPT for phone calls or a Chat GPT for B2B marketing, if all it takes are two half-decent coders from a state university. Are valuations overblown, the answer is most definitely yes, and they have been on rise.

You know what's surprising, although the EV/Revenue multiples we see are already in mid to high twenties for the last couple of years, a lot of them don't show up because the companies are pre-revenue or even pre-product. Now, the valuations and the trend line that you see above are quite noisy, and the data is uneven, but it still shows that the vertical is quite overvalued. The silver lining is that investors might be taking note of it, or maybe not. The overall VC ecosystem is struggling to see enough deal flow, there are not enough buyouts and M&As happening, and IPOs' seem like a far fetched dream for most of the companies in this insipid environment. A trend that is proving to be a substantial one is the rise of early-stage valuations and investors being more interested in getting in earlier with the vertical, which is quite obvious, but it is also due to the caveat coming in from the falling valuations and deal sizes at the later-stage.

Investors are navigating the AI landscape with a sense of cautious optimism. While they acknowledge the significant potential for AI to revolutionize various sectors, they are also mindful of several factors. These include the need to assess how companies effectively incorporate AI into their operations and the potential risks associated with its misuse. Regulatory scrutiny, exemplified by the European Union's recent regulations, is another factor on their radar. Despite these considerations, AI's attractiveness to investors undoubtedly remains high, and the market is experiencing upward pressure due to AI-related investments. However, the second half of the year holds uncertainties, and investors are prepared to navigate this dynamic landscape with a balanced approach, weighing both risk and reward. It is definitely not the end just because the capital invested or the deal count has gone down compared to a monstrosity of a year.

FINAL REMARKS:

There is a not a lot to say, but surely a lot to fathom at this point in time. Every single sector goes through its hype cycle and no one is to really blame for this, as any technological development for potential disruption carries this inherent risk. I know I did mention that I don't want to speak about the dot-com bubble analogy, but see how it turned out in the longer term. Betting our expectations on a bunch of investments and extrapolating those to be synonymous with a whole sector doesn't seem entirely rational, but that's how stuff happens. Autonomous vehicles saw their own bubble, where everybody is imagining roads without a single driver and a world that Asimov warned us about, but we're not near to that sort of autonomy. Same with cloud computing, which has seen its own set of overhyped valuations and investments, but proved and still proves to be one of the most significant industries that we need now and in the future. The application and the potential offered by AI are unprecedented and unimaginable, and to shun it off is not the right answer. Everybody is over-excited about it and that's true and I have seen some potential investments like that as well, $400M Series B for a pre-product company making LLMs' for software development (no, thanks). However, since when did we start expecting a ton of idealism and rationalism in the private capital space, or any sort of investments for that matter. VC is most certainly about the massive potential of an idea and how big the addressable market is, or how many MOATS does a product have and all of that, but it's also about an investor's existing relationships, serial entrepreneurs or your previous work experience or how high can you ride on a fictitious tide. The plain thing is that it's foolish to expect all of us to behave rationally, at least for the time being and especially in a domain which is this nascent. With respect to VC investments in AI, it's about finding the next few winners out of a never ending pool of opportunities, and at the end, it might even just come down to who was there first. Maybe, it's that simple or maybe it's not.