Gangsta AI

OpenAI Just Committed $30 Billion to One Georgia Data Center — and Started Selling Agents by the Hour

The Ledger

By Warren Buffett · 2026-07-24 · 4 min read

Now, I've never been able to tell you what a company will do next quarter. But I can usually tell you what it's afraid of. And on July 22, OpenAI told us plainly: it is afraid of becoming a commodity.

The company announced Project Camellia, a plan to spend more than $30 billion building a single super data center in Georgia — 3.2 gigawatts of power, enough to run a mid-sized city, phased in between 2028 and 2032. OpenAI says it has already committed $20 billion and is out hunting partners to finance the rest. That is not a rounding error. That is more than the entire market value of a lot of fine businesses that have been paying dividends since I was in short pants.

“Price is what you pay. Value is what you get. With a data center that won't fully switch on until 2032, you're paying today and finding out about the value on the installment plan.”

The other shoe: renting out the robots

On the very same day, OpenAI launched Presence — a platform that lets big companies deploy real-time voice and chat agents, built on GPT-5.2, wired into their own data, with guardrails and policy controls baked in. Think of it as OpenAI walking into your business, in a nice suit, offering to run your customer support desk.

I find the timing instructive. When a fellow shows you an enormous bill on Monday and a new way to charge customers on the same Monday, he is telling you how he intends to get square.

Here is the problem OpenAI is trying to outrun. Its rivals — Claude from Anthropic, Gemini from Google, and a fleet of cheap, capable open models out of China — keep dragging the price of raw intelligence toward zero. When the thing you sell gets cheaper every quarter, you either sell a whole lot more of it or you sell something stickier. Presence is the something-stickier. Once your company's phone system runs on OpenAI's agents, you don't switch vendors over a rounding error. That's a moat. Whether it's a $30 billion-deep moat is the open question.

The arithmetic nobody at the party wants to do

Reports put OpenAI's total compute commitments in the hundreds of billions. A single data center at $30 billion is just one brick in that wall. The bulls will tell you demand is infinite and the models keep getting better. They may be right. Frontier systems really are doing knowledge work that stunned people two years ago.

But I've watched a lot of parties where the punch was spiked with cheap capital. The tell is always the same: everyone's building capacity for a demand curve that only exists in the slides. Sometimes the demand shows up. Sometimes you're left owning the most expensive empty warehouse in Georgia.

I'm not betting against OpenAI. I'm just noting that spending $30 billion to build the factory and launching the storefront on the same afternoon is a company placing a very large wager on itself — with the house rules still being written.

What it means for you

You don't have to pick which lab wins the spending race to use their work. Before you bet your business — or your monthly subscription — on any one of them, put GPT-5.2, Claude, Gemini, and Grok in the same room and see which one actually earns its keep on your questions. Compare the top AI models side by side on Gangsta AI — that's the kind of due diligence I'd do before signing any check.

Try Gangsta AI free →

More: Best AI models · Compare all AI · Frontier Models · All articles