• Soda Road
  • Posts
  • 🥤 Everything You Don't Know About Generative AI

🥤 Everything You Don't Know About Generative AI

Generative AI for Dummies: All That Shines Ain't Gold!


Welcome to Soda Road, the newsletter for AI nerds, entrepreneurs, and all those who accidentally thought this was a soda-themed newsletter.


There's so much more to generative AI than prompt engineering and OpenAI. a16z released a fantastic (and lengthy 🥱) report on it. So here's the readable version of it!

After reading this, you can legally update your Twitter bio to "AI Expert" 🧠. And drop us a follow while you're at it.

The Tech Stack

Apps: End-user apps that don't own the AI they use. They could use OpenAI or Stable Diffusion.End-to-End Apps: End-user apps that own their AI.Open/Closed-Source Models (OpenAI, Stable Diffusion)AI that's either open-source or owned by another company and available to others via an API.Cloud Platforms (AWS, Azure): Cloud platforms that expose the hardware to train and use the AI.Compute Hardware (Nvidia, Intel): Companies manufacturing the physical hardware used to train and run AI models.

Winners and Losers

Application companies are seeing their revenues soar! 📈 But they're struggling to keep customers loyal and stand out from the competition because many apps use the same AI and APIs.

Model providers made this market possible, but they're still struggling to make money. OpenAI and Microsoft's partnership may prove to be a big brain move. But other model providers don't have a clear monetization path yet.

Despite this, three categories have hit $100 million in annualized revenue: image generation, copywriting, and code writing.

Now kids, remember these 2 life lessons:

  1. Growth isn't enough to build a successful company. Keeping customers loyal and creating high gross margins is crucial.

  2. If a stranger gives you drugs, say "Thank You" because drugs are expensive.

Big Tech's Big Bets Pay Off Big Time

Nvidia is the winner! They reported $3.8 billion in data center GPU revenue in Q3 of 2023, the majority came from generative AI uses. They're knee-deep in GPU architecture, software, and the academic community.

Google and TSMC are close competitors. Stable Diffusion and Google Cloud are leaning towards Google's hardware. It's believed TSMC manufactures many chips, even Nvidia GPUs.

What Does This All Mean?

  • Hosting is key to commercial success.

  • Proprietary APIs are in demand.

  • Hosting services for open-source models are being used to share and integrate models.

  • Fine-tuning and hosting agreements with enterprise customers may be a future monetization strategy.

  • Drugs are expensive.


That's everything. 

Follow us and send cool AI stuff on Twitter. Until next time ✌🏼