Google - The world's first "Full Stack AI Sovereign"

When talking about AI models, people often think of the three core layers to it:

1. The Chip (GPUs) - the hardware layer that allows parallel processing, allowing AI models to perform thousands of mathematical calculations and complex reasoning simultaneously.
2. The Cloud - the infrastructure layer that allows developers to train and deploy the AI model at scale.
3. The AI model itself.

When a company tries to create its own AI model, for example OpenAI with ChatGPT, they will need to buy tons of GPUs from Nvidia at a massive premium for the initial development. Then, they spend another hundred of billions of dollars to train and run their AI model on Oracle, Azure and Broadcom cloud services.

At this moment, anyone can see that building an AI model requires an enormous capital investment to run and maintain, especially for many AI start-ups. In the above example, even a company with rich capital like OpenAI, they are still experiencing heavy losses YoY while the CapEx is increasing unpredictably. They are not expected to be profitable until 2029 and I wonder how long until they ran out of money with this spending rate or how many more circular financing we will get to see. This combines with the uncertainty of whether AI can actually make money or just a bubble, adding more risks for investors when considering these AI plays.

On the other hand, we have Google - a full stack company in the AI race. You might be wondering, what makes Google so special? Let's me explain.

First, they designs their own custom AI chip called Tensor Processing Units (TPUs). This prevents them from being dependent on expensive third-party hardware such as Nvidia. It also allows them to specifically optimize the chips for Gemini (their AI model).

Secondly, they have their own infrastructure layer which are Google Cloud & Vertex AI. This includes the software and hardware systems that allows developers to train and deploy AI at scale. Additionally, because it is in-house, they can tune the software to talk directly with their custom chips with ease.

Lastly, their AI mode - Gemini. As they make their OWN chips and run on their OWN infrastructure, they can co-design the models to run specifically on their hardware, leading to better efficiency and speed.

But there's more ...

Before Gemini, Google is already one of the top 7 most valuable companies in the entire world and profitable for many years. Their revenue is growing at high teens (~18%) for the past 10 years with profit margin hanging around 20%-30%. While investors are worrying about the financial health of their AI company when it is having another massive CapEx, Google still has $60-$80 billion to burn and this is AFTER their CapEx.

Your investment in Google is down-side protected. Even in the worst case scenario when AI doesn't play out well, Google still has a strong advertising business with Google Search and Youtube that drives a big portion of the current revenue. They sill have a fast growing, high-margin cloud services as well as a bet on autonomous driving vehicle (Waymo).

Google's share prices have been doubled in just one year, reflecting the intrinsic value of the company. To be conservative, I believe that Google is reaching its fair value but I'd still hold it for long-term given its moat in the AI era. I will closely monitor their cloud growth/margin as well as their future CapEx to further inform my investment decisions.