Anthropic chip. Why?

Everyone has to have their own chips. The latest company to get on the bandwagon is Anthropic — Anthropic is hiring an AI chip design team. This announcement comes a few days after OpenAI shared with the world its own AI Chip made in collaboration with Broadcom.

Every company wants to be in control of its own AI compute destiny. Google has one. So do Amazon and Apple. Why? They don’t want to pay and be dependent on Nvidia or AMD or Intel.

The question is: Does this provide a competitive advantage?

The narrative to support this theory that everyone needs their own AI chips is that if you build your own, you can make it far better suitable for your own particular needs.

In the case of OpenAI and their collaboration with Broadcom, the chip is using much simpler architecture compared to the ones coming out from Nvidia. It means that it is built for a very narrow set of tasks which the chip can perform.

One can expect that Anthropic will take the same path. I am not questioning the ability for Anthropic to hire very smart people who can design chips like that.

However, it took OpenAI and Broadcom about 18 months before the announcement and the expected full scale, mass deployment is anticipated in the first half of 2028. That’s about 4 years.

What else do you have to do in order to realize that competitive advantage I mentioned earlier?

You have to design and build the servers which will be able to utilize this new chip. And you have to create the software which allows your developers to take advantage of this new hardware. And you have to put in place where these new chips can be used. We call it the data center.

All of the above takes time, people, and money.

The money part is especially interesting. Do you know who the entities are that are funding Anthropic? Among them are Google, Nvidia, and Amazon. The same companies which already have custom chips, while investing in Anthropic and then taking the money back for hosting.

Where else is Anthropic looking for money?

This latest announcement provides a clue — Anthropic, Macquarie and GIC Form Venture for AI Data Centers. We learn that ‘Anthropic committed to covering any consumer electricity price hikes from the facilities’. The rest will be funded by the asset management companies. If the timing is right, by the time the new AI Chip is ready, there might be a place to put it.

(Side note: I remember in the dot.com bubble era, to name the company, people would come with names originating in Sanskrit. Now, it must be from Greek mythology. The Anthropic project is called Theseus Infrastructure. Theseus was a divine hero. His claim to fame was killing the Minotaur. Or perhaps it is tied to the Ship Of Theseus, a paradox regarding identity and change across time.)

What else would have to be true for this new AI Chip to become a strategic advantage for Anthropic?

For one — Nvidia, AMD, Intel, Google, Amazon (and others) would have to stop any development on new chips.

Let’s take Nvidia. It employs about 43,000 people and of that, 37% are in engineering. That’s close to 16,000 people. Anthropic? Depending on the source, somewhere between 2,500 and 5,000. So, that’s already a huge advantage for Nvidia, which already, like Intel and AMD with their chips, knows how to design new chips and I am sure have something similar in the pipeline.

Google and Amazon (and Apple) decided to make their own chips to remove the dependency — at least partially — on Nvidia and AMD. But Nvidia, Google and Amazon are the companies which Anthropic depends on for money and compute infrastructure.

The money part Anthropic wants to solve by going public. The just-announced IPO would value the company at $2 trillion.

I commend Anthropic for mapping out a long term strategy. Similar to OpenAi, Anthropic is still a startup finding use cases for its technology. And to consider — technology still too unstable for true enterprise deployment. It is built by a team which is not even capable of testing for rogue behavior and whose benefactors are questioning if the closed model is better than an open one.

To answer the question — Will having its own AI Chip provide a competitive advantage? There are so many things which will have to be true for this strategy to deliver. We will know the answer in about 5 years.

I’ve found the current events around AI to be an incredible opportunity to study and learn about strategy. You can read books, go through case studies, but here in real life you can observe what companies are doing. You compare the theory with the situation on the ground. The AI industry compressed the timeline from years to months or weeks. I would like to extend my Thank You to all the people and companies funding it.

The recurrent pattern? There is always more to learn.