OpenAI is scaling its Stargate infrastructure to build the compute backbone it says is necessary for AGI — targeting tens of gigawatts of data center capacity. That’s a power footprint comparable to a small country’s entire electrical grid. The effort involves rethinking chip supply chains, rack architecture, and cluster design from the ground up.
The scale of the bet is staggering. According to Verdantix analysis, OpenAI’s compute buildout carries a price tag in the trillion-dollar range, with a recent AMD deal signaling the company is diversifying beyond Nvidia for its chip supply. Sachin Katti, who leads infrastructure at OpenAI, has described the challenge as building “resilient compute supply chains” — not just buying more GPUs, but redesigning every layer of the stack from silicon to cooling.
IMPORTANT
This isn’t just a hardware story. Every major AI model provider — Anthropic, Google, Meta — is racing to lock down compute capacity. OpenAI’s infrastructure moves directly affect API pricing, model capability ceilings, and how fast new models ship.
The practical signal here: OpenAI is treating compute as the primary bottleneck to AGI, not algorithms. That framing matters for developers building on OpenAI’s APIs. More data center capacity means higher rate limits, faster inference, and potentially lower costs per token as supply catches up with demand. It also means OpenAI is locking itself into massive fixed costs — a bet that demand for frontier AI compute will keep climbing.
For teams choosing where to build, watch the infrastructure commitments. Companies investing billions in physical compute are signaling they expect their models to keep scaling — and that they’ll have the capacity to serve them. If you’re dependent on OpenAI’s API for production workloads, this buildout is the reason to expect improving reliability and throughput over the next 12-18 months.