This week, the world thought the AI ​​bubble had burst. Until Nvidia's CEO released surreal results, exceeding 51% growth. And according to him, they're only just getting started.

What's even crazier is that few people know about the war between CPUs, GPUs, ASICs, FPGAs, and the new brain of modern AI.

Get ready.

Nvidia's growth is impressive. Jensen Huang (Nvidia's CEO) revealed that of the recent $57 billion in revenue, no less than $51.2 billion came from data centers.

He clearly stated: Nvidia has visibility into half a trillion dollars in orders for Blackwell and Rubin chips. HALF A TRILLION.

Today, each Nvidia rack with 72 Blackwell GPUs costs around US$3 million.

The company delivers 1,000 racks per week.

And OpenAI alone has purchased more than 4 million GPUs.

The GPU has become the new oil of the digital age. It processes thousands of simultaneous operations and has become perfect for training models like GPT, Claude, and Gemini.

But a silent revolution has begun.

The cloud giants (Google, Amazon, Meta, Microsoft) are creating their own chips to reduce costs and break their dependence on Nvidia.

ASICs are cheaper, more efficient, and up to 50% more energy-efficient. The downside is that they cannot be reprogrammed.

They are surgical and extremely fast, but specialized. Whereas GPUs are like Swiss Army knives, with multiple functions.

Google was the pioneer. It launched TPU in 2015 and is now in its seventh generation with the Ironwood chip.

Anthropic trains Claude on over ONE MILLION TPUs. Some experts claim that TPUs match or surpass Nvidia in specific workloads.

AWS (from Amazon) responded with Inferentia for inference and Trainium for training. In Anthropic's data centers, more than 500,000 Trainium2 chips are already operating without any Nvidia GPUs.

AWS estimates up to a 40% advantage in cost-performance.

Meta created its own chip in 2023. Microsoft is working on the Maia chip for Azure, but is facing delays. Tesla is developing ASICs for autonomous vehicles. Broadcom has become the world's largest manufacturer of custom chips.

In other words, the competition is fierce.

Behind all these chips lies a single invisible force: TSMC of Taiwan.

It manufactures Nvidia Blackwell, Google TPU, Amazon Trainium, Apple M-series, and Qualcomm Snapdragon.

Taiwan has become the epicenter of the 21st-century economy.

The battle is not just technological. It's geopolitical. With the risk of a Chinese invasion of Taiwan. The United States is funding new factories in Arizona. China is accelerating its own chip production, but is suffering from restrictions.

And the biggest limitation isn't silicon.

It's energy to power colossal data centers.