The Value of AI: Startups vs. Incumbents

The Value of AI: Startups vs. Incumbents

In the world of AI, the distribution of value between startups and incumbents has been rather lopsided. Despite the significant activity in the startup space, the majority of the value generated by AI has gone to incumbents. This is in stark contrast to previous waves of technological advancements, such as the internet and mobile, where startups were able to capture a larger share of the value.

During the first internet wave, companies like Google, Amazon, Paypal, and Facebook emerged as successful startups, while incumbents like Microsoft and Apple were able to extend their franchises onto the internet. The split between startups and incumbents in terms of value capture was roughly 60:40 or 70:30. Similarly, in the mobile space, incumbents like Apple and Google dominated, but startups like Whatsapp, Uber, and Instagram still managed to capture a significant portion of the value. The split in this case was around 20:80.

However, when it comes to crypto, startups have been the clear winners, with companies like Bitcoin, Ethereum, Coinbase, and Binance capturing nearly 100% of the value. Existing financial services and infrastructure companies have had little participation in value creation in this space.

To beat an incumbent as a startup in the AI space, you typically need to build something that is dramatically better and can overcome the distribution, capital, and pre-existing product moats of the incumbent. Alternatively, you can focus on a brand new customer segment or distribution moat that the incumbent cannot serve. In general, a 10X better product is required.

One possible reason why incumbents have been successful in capturing value is their data advantage. However, as companies use the broader internet as an initial training set and switch to models that work more robustly with smaller data sets, this advantage may be diminishing.

Many prior-wave AI companies either directly took on incumbents or worked in hard markets, such as education or healthcare, where technological innovation often faces challenges due to market structure, regulation, or a lack of focus on end-user needs. However, this time feels different. The speed of innovation across various areas is remarkable, and the technology itself seems dramatically stronger. This makes it easier to create 10X better products that can overcome incumbent advantages.

While GPT-3 has shown promise, it has not yet led to the emergence of many startups building big businesses on it. However, a model that is 5-10X better than GPT-3 could create a whole new startup ecosystem while also augmenting incumbent products.

Unlike previous waves of AI startups, there is now a clear set of infrastructure-centric companies with broad adoption and rapidly growing usage. These companies, such as OpenAI, Stability.AI, Hugging Face, and Weights and Biases, provide startups with access to the necessary technologies, creating more opportunities for them.

One area where AI can have a significant impact is in highly repetitive, highly paid tasks, such as coding, marketing copy, and website images. Workflow tools that incorporate AI features can become essential parts of broader workflows, making tasks more efficient and effective.

The key to success in this exciting wave of AI technology will be to identify actual end user needs and unserved product markets that can benefit from these advancements. It's crucial to focus on the needs of the end users and build products that address those needs.

In conclusion, after years of working on AI-related products and investing in them, it feels like startups are finally starting to get real value from AI. This current wave of AI technology is different, with stronger capabilities and a greater potential for startups to capture value. Exciting times lie ahead!

Actionable Advice:

1. Focus on building a product that is 10X better than what incumbents offer, or target a new customer segment or distribution moat that incumbents cannot serve.
2. Leverage the advancements in AI technology to identify actual end user needs and unserved product markets.
3. Build safety nets into all experiments and closely monitor metrics such as opt-out rates, uninstall rates, and actions per hundred pings to assess the effectiveness of your notification strategy. Also, personalize your notifications with relevant emoji to make them feel more personal and engaging.