AI Infrastructure is the first upgrade to basic technology since the iPhone. Right now, it’s highly proprietary. History says it won’t stay that way.
Clouds were built with open source because low costs were the priority. But there’s more to open source than free code. Open source puts eyes on the code, and this lets many companies compete for business.
Today’s upgrades, however, are mainly being built with Nvidia chips and (more important) Cuda software. It’s proprietary stuff. There are open source alternatives but they have yet to gain traction, just as Linux took years to gain traction against Microsoft.
One result of making GenAI proprietary, as I’ve written before, is that GenAI services can’t be free. Microsoft Co-Pilot costs $30/ month per seat. Notice they don’t say $360/year per user (which it is). The price of Co-Pilot is a lot higher than the Office software it’s built on.
The cost of Nvidia is putting a strain on budgets, even for Cloud Czars. Cloud capital spending came to $127 billion in 2023, and it’s expected to jump another 17% this year. Alphabet alone will spend $50 billion.
So far, the Cloud Czars are all spending with cash flow. But debt loads are also increasing, not least because the data center REITs which connect these clouds, like Equinix, must by law pass earnings on to shareholders. Debt is a sterner taskmaster than cash flow since it must always be repaid if you’re to stay in business.
Zuck a Hero?
Were it not for Meta Platforms, which made its AI models open source, over half the world would be closed from any GenAI benefits right now. But just as Mark Zuckerberg cut cloud costs with his Open Compute Project, he’s relying on open source again with a Large Language Model (LLM) called (appropriately) LLaMa.
As the Android is to the iPhone, Meta hopes LLaMa will be to OpenAI. I think history is going to repeat itself.
Proprietary AI models will open the market. Some will have their business models proven. Others will not. Those that do see their business models proven will quickly gain open source competition, assuming Meta doesn’t decide to unilaterally close the source (as it could, as OpenAI did).
Investors pouring money into closed source AI don’t want to hear this, but it’s true. It’s hard to compete with open source. Apple, for all its wealth, only managed to get half the market against open source Android. The same will be true with AI.
This is the genius of open source. It doesn’t innovate. Instead, it lets the world catch up with innovation. It makes innovation solid, builds standards, and creates competitive markets around it.
The people who now dominate the GenAI market, like OpenAI’s Sam Altman, know I’m right. That’s why he keeps trying to scare people about the threat of “Applied General Intelligence,” which doesn’t exist, hoping to keep its entrance portals closed. He wants the source closed so only he (and his friends) benefit.
But that’s not how the world works. Open source will win in AI as everywhere else. It will just take time.
