OpenAI's $750B infrastructure spend is your token price
OpenAI raised planned AI infrastructure spending to $750B through 2030 and announced a $20B Georgia campus. Here's what that math does to what you pay per token.
Somebody has to pay for the buildings. OpenAI said today it will spend $750 billion on AI infrastructure through 2030 — roughly 25% more than the ~$600B it projected earlier this year — and announced its first campus where it's the principal builder rather than a tenant. If you're paying per token today, that number is the shape of your bill in 2028.
What actually happened
Per TechCrunch, the headline commitment is $750B through 2030. The first piece is Project Camellia, a $20B data center campus on 1,400 acres at the Savannah Gateway Industrial Hub in Effingham County, Georgia — four buildings, up to 1,000 permanent jobs, closed-loop cooling.
The power is the real story. Georgia Power will deliver at least 3.2 gigawatts in phases from 2028 through 2032. OpenAI says it will fully fund the infrastructure and electric-service costs so existing ratepayers don't subsidize it, and has pledged $80M toward local schools, healthcare, workforce training, and public safety. Effingham County approved a 50% property tax abatement for 15 years.
Note the dates. Power lands in 2028 at the earliest and fully arrives in 2032. This is not capacity that relieves anything you're running this quarter.
Why AI infrastructure spending matters for your business
Capex this size gets amortized into the price of a token. Not immediately — the current market is subsidized, and competition keeps the sticker low. But debt-financed concrete has to be paid back on a schedule, and the schedule runs through the exact years OpenAI is telling you it will be spending hardest. We've made this argument about prepaid compute and about the broader capex wave, and each new announcement makes it more concrete, literally.
The operator move isn't to predict the price. It's to make the price not matter much. Every automation you run should have a model boundary you can swap behind: a router, an env var, an interface — not openai.chat.completions.create scattered across forty files. When the invoice moves 3x, you want that to be an afternoon of config, not a rewrite.
Second move: know your cost per completed task, not your cost per million tokens. A workflow that costs $0.04 to run and replaces eleven minutes of somebody's day survives a price hike. A chatbot nobody uses does not.
Key takeaways
- OpenAI raised planned infrastructure spend to $750B through 2030, up ~25% from earlier 2026 estimates
- Project Camellia: $20B, 1,400 acres in Effingham County, GA, at least 3.2GW delivered 2028–2032
- Capacity that arrives in 2028 does nothing for your 2026 rate limits or your 2027 bill
- Keep one model boundary in your code and track cost per completed task, not per token
Hardcoded to one model provider? We build automations with a swappable model layer and per-task cost tracking, so a vendor's capex plan isn't your roadmap. See how we build it or send us your stack.
Sources: TechCrunch, The Current GA.
- #openai
- #ai-infrastructure
- #token-pricing
- #vendor-risk
- #data-centers
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Pilot Protocol's $4.5M: a directory isn't a standard
A seed-stage AI agent network wants to be where agents discover each other. Useful, but know the difference between an open spec and someone else's front door.
Read itNvidia backs Safe Superintelligence, discloses no terms
Nvidia's Safe Superintelligence partnership names no dollar figure and no term length. Reported at $5B. Here's how to read AI deals that omit the numbers.
Read it