According to CB Insights, VCs invested over almost $22bn in Gen AI in 2023. Most of this money went to LLMs, over 70%, based on Dealroom’s analysis. These model makers like Open AI, Anthropic or Adept AI require large sums of funding in order to train and deploy these general models. I believe that we will not see many more funding rounds into model makers for various reasons (very capital intensive, regulation, how many are needed?) and it seems many investors and founders are focused on the infrastructure layer. While this appears a logical target in terms of how value accrues, I believe that the true potential resides in the application layer, particularly (but not solely) in the consumer space. Although GenAI is a revolutionary technology, incumbents in the infrastructure domain are aggressively investing. Consequently, I do not believe GenAI will disrupt the infrastructure market, according to Christensen’s theory of disruptive innovation. The greater opportunity lies within the application layer.
The Current State of AI Infrastructure
Big tech firms like Microsoft, Google, Meta, Nvidia, and Amazon invested a combined $374 billion dollars in R&D/Capex last year.
According to Tony Pasquariello, head of hedge fund coverage at Goldman Sachs:
“Another way to frame it: the Magnificent 7 reinvests 61% of their operating free cash flow back into capex + R&D … that’s tracking to be 3x the 493 of the S&P 500“.
Nasdaq
These companies are establishing the horizontal layer of AI technology, encompassing everything from cloud computing and AI chips to machine learning frameworks, LLMs and data storage solutions. Their scale and resources enable them to innovate rapidly, making it challenging for startups to compete directly in this space. If startups do decide to compete on this ground floor, then they need to raise billions of dollars, most of which seems more like a Capex expense.
Incremental Advancements and Platform Risk
Other startups focusing on the AI infrastructure layer often aim to optimize performance or introduce niche innovations. While these advancements can be valuable, they tend to be incremental rather than revolutionary. Additionally, startups in this space face significant platform risk. The big tech firms not only set the standards but can also quickly incorporate similar features into their existing platforms, potentially rendering a startup’s unique offering obsolete. The pace of commoditization is staggering. This risk obviously also exists on the consumer/application layer, more on that below.
Another question I have: how many infrastructure, tools, and shovel companies do we need that will not be built by the big tech firms? Given the comprehensive nature of the solutions provided by these giants, the answer is likely very few.
Challenges in the Application Layer
Before diving into the opportunities in the application layer, it’s essential to highlight a significant challenge. Many startups have created so-called “wrappers” around platforms like ChatGPT and other Gen AI technologies.
The problem with this approach is the high platform risk, as most of the value is derived from the underlying platform rather than the startup itself. This dependency can be precarious, and startups must carefully assess the value they add versus the value extracted from these foundational platforms. I suspect many of the first-wave of GenAI startups targeting the application layer will fall victim to this as the big platforms expand and partner to offer similar features.
Big tech firms are integrating GenAI into many existing tools, B2B and consumer facing, so this is something founders should watch out for
Advantages in the Application Layer
Despite the challenges, the application layer offers numerous opportunities, particularly on the consumer side. AI has the potential to offer entirely new user experiences, making them far more personalised and intuitive across various sectors such as finance, education, and gaming.
Here are some key advantages:
- Owning User Data and the customer – For startups in the application layer, owning user data is crucial. This ownership allows them to continuously add value and build a competitive moat. By leveraging user data, startups can refine their offerings, improve personalisation, and create unique insights that set them apart from competitors.
- Enhanced User Experiences: AI applications can deliver highly personalised experiences, tailoring interactions to individual user preferences and behaviours. For example voice interfaces, enabled by the latest model by chatgpt, 4o.
- Cost Structure Changes: AI can automate numerous processes, significantly reducing operational costs. You will not need the same amount of programmers, designers/creators, marketers etc. Native GenAi companies should leverage that cost advantage.
- Faster Go-to-Market: Using existing GenAi tools and infrastructure should allow startups to cut development time and test and iterate much faster.
- New Business Models: AI enables innovative business models, such as offering AI-driven services or selling the “work” performed by AI agents rather than traditional products or services, where the user still needs to operate some kind of dashboard/product.
If you want to learn more about these advantages, I recommend Harry Stabbings “20 Minute VC” podcast featuring Sarah Tavel, a partner at Benchmark. She dives deeper into the importance of owning the customer and the ability of application layer startups to nail the user experience and iterate quickly.
AI Agents and Vertical Opportunities
AI agents represent a prime example of application-layer innovation. Instead of competing with established players like Microsoft in the B2B space or Meta and Google in advertising, startups can explore other verticals where AI can add significant value. Potential areas include:
- Gaming – eg. AI-driven games and virtual worlds.
- Content – eg. Personalised content creation and curation
- Travel – eg. AI travel planning and exploration
- Entertainment – eg. Interactive AI storytelling and experiences
- Health – eg. AI-powered diagnostics and personalized care
- Trust and Security – eg. AI systems for identity, fraud detection
- Finance – eg. AI-enabled financial advisory and investment tools
- Legal Tech – eg. AI-assisted legal research and contract analysis
Conclusion
While it seems that a lot of value is currently captured by the infrastructure layer (both semis and hyperscalers) I believe that for startups the bigger opportunity will be in the application /consumer layer. Apoorv Agrawal shared a great article summarising the current economics of Gen AI. He demonstrates how mobile once went in a similar path, where value creation shifted from the semiconductors, to infrastructure players (telcos) and eventually to software application and services.
Big tech firms are already dominating the infrastructure space with their massive investments and rapid innovations. In contrast, incumbents operating on the application layer are slower to innovate and do not have the same resources and culture to lean-into Gen AI as fast. These vertices offer the highest potential to disrupt established markets.
If you are a founder looking to create a new user experience, especially on the consumer side, please come talk to us at Remagine Ventures. Make sure to think about the value you create versus the value you extract from the underlying Gen AI platforms, but utilize these to reduce your capital needs and accelerate your go-to-market roadmap.
- The Real Promise of Gen AI: It’s in the Apps, Not the Pipes - May 21, 2024
- Generative AI: Opportunity or the end of media as we know it? - April 15, 2024
- What can online communities learn from Eve Online? - April 28, 2023
