Year 3 of AI: What’s Really Happening Underneath

Happy Friday!

Last week was all about credit conditions and how quickly did market move from that right?

This week, I want to focus on the optimization happening across AI as highlighted by companies like Alibaba, Airbnb, and Supermicro. Like Alibaba showing that their new pooling system cut GPU usage by 82%.

We are AI believers. But we also try to stay disciplined. Looking to the horizon while paying close attention to the here and now. The scale of this buildout is reshaping every layer of the stack and will influence everything from hardware demand to how AI products reach users.

Let’s get into the data!

Here is the summary if you want just that:

• Anthropic 1M TPUs

• Optimization across AI board

• 90% efficiency

• 1,000 apps in AI or more.

• Alibaba market leader

• Usability is next frontier

Let’s get this going.

So, yesterday SuperMicro reported and I couldn’t help but add it here. The cleanest takeaway to me as an investor looking out years and not quarters.

SuperMicro’s new MicroBlade systems claim 95% cable reduction, 70% space savings, and 30% energy savings over traditional servers.

If these numbers hold in real-world environments, we are seeing one of the biggest step changes in physical data center efficiency in years. Optimization is no longer then theoretical. It is showing up in metal, silicon, and power.

Keep reading

This comes just 5 days after we got this.

Alibaba saying that they optimized chip usage with a GPU pooling system that cut GPU usage by 82%. That is an enormous reduction in cost and energy for large-scale model deployment. Hmmm.

Remember that Alibaba is the leader in China. Growing fast and optimizing at the same time.

Growing so fast that the model they launched is being used everywhere now. Airbnb has started using Alibaba’s Qwen model to power its chatbot.

We then got word this week too that Anthropic plans to scale beyond one million TPUs through Google Cloud by 2026.

Side note (A TPU (Tensor Processing Unit) is a specialized chip designed by Google to accelerate machine learning tasks. Unlike general-purpose CPUs or GPUs, TPUs are optimized specifically for tensor operations the math behind neural networks.)

Hope that helps little bit but moral of story is a TPU doesn’t come from AMD or NVIDIA and 1 million is a lot.

Now this is a random image but it shows the hundreds if not thousands of AI tools people are creating.

99% of these will be zeros.

People are demoing, trying, training.

Basically lots of wasted spend on unoptimized infrastructure.

This is changing.

Then plenty of surveys are done. Here’s just another. 90% of businesses say AI has not yet boosted their revenue.

Only 9% have seen measurable lift, and most of that came from increased traffic, not productivity or profitability. This shows how much of the current AI wave still lives in experimentation rather than execution.

I bring this up not to say AI is dead. But beware of the optimization that is taking place across the stack. We are basically in year 3 of this AI journey. AI will be very large market, but as an investment firm and in times like these, when momentum feels unstoppable, disciplined investing becomes the real edge.

---

About Avory & Co.

Investing where the world is headed. 

Avory specializes in high-conviction equity strategies, emphasizing Secular Growth and Transformation Stories driven by exceptional teams. Data guides decisions. We cater to high net worth investors, family offices, and institutional investors. Note: This information doesn't constitute a recommendation to buy or sell any mentioned securities. Avory is based in Miami, Florida with clients all across the globe.

Speak to us: Schedule a Brief Zoom Meeting

Send us an email: Team@avoryco.com

Want to invest? We are on most platforms. 

Want More

🎥 Avory YouTube Channel

🎙️ Avory Podcast 

Disclaimer: Not a recommendation to purchase or sell any securities mentioned. This is for educational purposes only.