Stargate: Inside OpenAI's $500 Billion Bet on AI Data Centers

The AI race is often framed as a battle of models. Increasingly, it's a battle of concrete, power, and GPUs. Nothing captures that better than OpenAI's Stargate — a data-center buildout of almost incomprehensible scale.
The Scale
As of 2026, Stargate is a multi-continental initiative to build roughly 10 gigawatts of AI data-center capacity, underwritten by an unprecedented $500 billion investment. And Sam Altman has floated even larger numbers — commitments of about $1.4 trillion over the next 8 years.
For perspective: 10 GW is comparable to the output of several large nuclear plants, dedicated to running AI.
What's Actually Built
This isn't all on paper:
- The first large Stargate facility is fully operational in Abilene, Texas, running on Oracle Cloud Infrastructure.
- The campus spans ~4 million square feet with 1.2 GW of power capacity.
- OpenAI, Oracle, and SoftBank have announced five new Stargate sites, with additional energy financing (SB Energy secured $1B from OpenAI and SoftBank).
The Business Context
OpenAI can contemplate these numbers because its business is growing fast: revenue climbed from ~$12B ARR in mid-2025 to ~$25B ARR in early 2026, and an April 2026 round raised $122B, pushing valuation to $852 billion.
The Reality Check
Not everything is smooth. OpenAI paused its UK Stargate project, citing energy costs and regulatory concerns (affecting plans for ~31,000 GPUs). It's a reminder that the binding constraint on AI isn't just chips — it's power and permits.
The bet: whoever controls the most compute controls the frontier. The risk: spending $500B–$1.4T ahead of proven returns.
The Bottom Line
Stargate is the physical embodiment of the AI boom — and of its biggest question. If AI demand keeps compounding, it's visionary infrastructure. If it doesn't, it's the most expensive bet in tech history.
Brandomize helps brands understand the infrastructure realities behind the AI hype. The models run on power and silicon — we help you see the whole picture.
Related Thoughts
OpenAI's GPT-5.6 'Sol': The Model That Replaced GPT-5.2 in ChatGPT
OpenAI has moved fast in 2026 — from GPT-5.4 in March to the GPT-5.5 and GPT-5.6 'Sol' line now powering ChatGPT. Here's the state of OpenAI's frontier.
Gemini 4 Is Training: Google Confirms Its Most Ambitious Model Yet
Google confirmed on July 21, 2026 that pretraining has begun for Gemini 4 — a 'significantly larger' frontier model — while shipping Gemini 3.6 Flash in the meantime.