The Morning Download: Oracle, OpenAI’s Extreme Economics

Good morning. To put Oracle’s $300 billion cloud deal with OpenAI in context, consider that the contract will require 4.5 gigawatts of power capacity, roughly comparable to the electricity produced by more than two Hoover Dams or four million homes.

The financial details are equally jarring. “OpenAI is a money-losing startup that disclosed in June it was generating roughly $10 billion in annual revenue—less than one-fifth of the $60 billion it will have to pay on average every year,” the WSJ reports. “Oracle is concentrating a large chunk of its future revenue on one customer—and will likely have to take on debt to buy the AI chips needed to power the data centers.”

It begs the question, “why?” Why would Oracle and OpenAI enter into such a deal, and why would anyone lend them capital to fund it? For one thing, they are focused on demand. A “near-constant computing shortage” is “hampering the rollout of OpenAI’s products and constraining progress building new AI models,” and that’s a broader issue for the AI industry, the WSJ says.

Historic bets on AI also assume that demand will skyrocket from here, on the belief that companies will ramp up use of agents that constantly run in the background for longer and longer periods of time. Such applications will require more compute power and energy than training models, big projects that occur in spurts. And they expect the price of both compute power and energy to fall as supply increases and their fundamental efficiency improves.

Large and risky investments on that scale only make sense when viewed alongside the construction of the electrical or telecommunications grids or the interstate highways. And they only make sense if you believe that AI itself is going to have a broad, societal impact on par with the lightbulb or the car. There are good indications that the bet will prove to be correct, but even so it will be a high-risk process with plenty of pain.

WSJLI: There's been a lot of talk that we're in an AI bubble, but maybe the tide is starting to shift. Oracle shares skyrocketed this week on news that it scored mega AI deals. What do you think?

Jang: You need bubbles to create intensity and focus of capital and to create, to attract a lot of talent, to be entrepreneurial and build new things that are very risky.

I'm very, very optimistic about this technology wave because it's sort of a mega-wave on top of every other technology wave that's happening: robotics, autonomous vehicles, software.

WSJLI: What aren't people paying enough attention to when it comes to AI?

Jang: In the future, everything will be inference based. This is the part that even the AI industry, the tech industry, but definitely the mainstream financial markets and traditional industries, underestimate.

Every action on the internet, every action on servers, every action in your mobile apps, even you walk home and you have home personal computing, security cameras, voice control ... it's all going to be using inference.

WSJLI: Where are you investing in AI right now, and why?

Jang: What used to be called "middleware" and tooling in the previous era was considered a thin layer and not as exciting, but actually is probably the most exciting area in AI because it makes the models useful. It enables the developers and the end users and the businesses to have access to GPUs and models.

This is a fat layer of value and excitement and sexiness for the industry because that's where the delta is. There's a lot of old-school VCs that don't like that space, but they're so wrong.