Is Generative AI the Next Bubble?

Generative AI has been transformational; there has never been a period in human history where a person's efficiency exponentially increased this much through adopting new technology. The printing press allowed mass production of books; the industrial revolution gave businesses the ability to quickly and cheaply scale production; the internet allowed us to get information faster and simplify our decision-making process; the iPhone put a mini computer in everyone’s hands to produce anywhere, anytime. Yet, none of those gave you the capacity to do 100X, instantly, with no extra effort.

Let me give you a few examples that are applicable to people, and not just businesses:

• With Midjourney or Stable Diffusion, marketeers can generate images for marketing content in a matter of seconds at a fraction of the cost. 
• Claude+ or GPT 4 enable you to draft emails or organise a trip using prompts, whilst constantly iterating to reach a perfected version.
• ElevenLabs or Google TTS can voice emails, news, books or any content you want to hear aloud in minutes.

These are just some examples of this tech being ‘productised’ and there are hundreds of other applications disrupting how we learn, work and play.  Character AI or OtherHalf are ‘redefining’ companionship, Gendo AI changing how architects/interior designers work and Synthesia democratising video creation, to name a few. 

Are all of these perfect? No, but we do not need them to be in order to see a boost in workforce productivity. Do businesses require a top-notch output? It depends on the use case, although the benefit of using GenAI's often imperfect solutions already outpaces the traditional approach in many cases.

Is there a bubble? No, and yes; the key lies in expectation vs reality and the answer is a bit trickier here but before getting into it, let’s compare some previous FOMO-fueled (and arguably popped) bubbles:

What bubbles have we seen before?

• Crypto: The crypto boom took most of us by surprise but in short, there was a general understanding that it would/could replace the centralised banking system and make everything cheaper and faster. In practice, however, the majority of people were mostly doing [multiple] one-off transactions to speculate; not to buy or sell actual products or services. Startups did not have any sort of recurring revenue but, rather, spiky inflows. Maybe this one takes longer to play out though.
  
  
• Metaverse: The promise was fairly simple here; a large portion of our lives will happen in a virtual reality setting because we could be whoever we wanted, anytime. Facebook even changed its name to Meta. Fast forward, there are neither one-off transactions nor subscriptions. Metaverse startups have had to transition towards corporate metaverses that are closed ecosystems. These startups may never recover investments because the wider population has not really bought into the idea. Will it change with Apple’s Vision Pro? Apple plays a different game.
  
  
• Quick delivery: 10 min deliveries, anytime, were supposed to improve our lives. In exchange, we had to sacrifice product variety and pay for a delivery fee. It was a simple promise believed by millions of people around the world. The tricky part, however, was that people used it as long as subsidies were in place. How many times do you eat ice cream at 11pm every week? Startups had zero recurring revenue, technically-impossible margins and rampant churn.
  
  
• Biotech: The Covid pandemic brought us the idea that new breakthrough treatments could be developed quickly for any disease and we would substantially improve our quality of life and life expectancy. Yet, everyone forgot that biotech requires a large upfront investment, has no recurring revenues, public health systems do not have the financial strength for expensive treatments and the vast majority of therapies don’t pass phase 2 of clinical trials.
  
  
• Artificial intelligence (Gen 1): Back in 2016-2022 there was a boom in AI startups, of which I was also part of, more focused on Enterprise solutions that leveraged what we called HiL (human-in-the-loop) to provide the type of accuracy clients expected. At that point, neither investors nor clients fully understood the extent of HiL and the difficulty of getting fully automated outputs. What’s more, Gen 1 AI companies had to do customised solutions to be able to close contracts. As such, unit economics were bad and those startups are stuck in an endless Enterprise customisation cycle.

The biggest problem with these examples is that often, near-term tangible value vs. future promised value was too misaligned. Everyone expected those technologies to be able to transform society while making billions and good margins. Once investors understood that they were longer shots or unfeasible/unsustainable models, they dried up funding and moved on to the next thing. In my opinion, however, those investors that stick to their original thesis and help their portfolio evolve their business models will see the benefits.

How does it compare to Generative AI (Gen 2)

Generative AI has also generated very high expectations but, differently from other bubbles, they are outperforming even the highest early expectations. GenAI startups have been expensive to build due to the scarcity of talent and the high cost of using top-of-the-range GPUs. Once those models are built and deployed, GenAI companies bring the best of all possible worlds:

1. 100X productivity increase
2. Recurring revenue (the famous MRR or ARR that for other bubbles was overlooked)
3. Unit economics (aka profits) are unmatched
4. PLG (Product-led Growth) and PLS (Product-led Sales) allow anyone (people & businesses) to extract value without customisations
5. Tiny teams to scale globally
6. Product cycles that are truly short and iterative
7. New industries and use cases that were not possible before, are now enabled

Generative AI is not perfect and has a substantial amount of challenges ahead. From ethics & regulation or model accuracy to helping society understand that we will always need people. Yet, despite the challenges, businesses have adopted AI in record times.

Is there a bubble in Generative AI? Yes, and no. According to Pitchbook, $4.5bn was invested in GenAI in 2022 and a further $1.6bn in Q1 2023. Q2 will break all records after Inflection AI, ElevenLabs and Mistral AI, amongst others, announced their rounds. There is technically a bubble if you think about the pace at which startups are pivoting to GenAI or raising large rounds. There is also a bubble in startups doing foundational LLMs when the foundations have already been built. Yet, there is not a bubble in phase 2 of Generative AI: vertically-integrated applications.

Vertically-integrated GenAI companies are just getting started. They are the next natural layer in Generative AI because their applications harness the strength of foundational LLMs with proprietary datasets to solve niche problems in each industry. These applications will take productivity to a whole new level for everyone, across all industries. These companies are being built today.

Is there an actual bubble in Generative AI? No. The industry is just getting started and profits (yes, that positive thing at the end of the P&L) are real.

‍

Carles is a Venture Partner with Concept Ventures and VP of Revenue at ElevenLabs.

To hear more from Carles, follow him on LinkedIn and Twitter.