2026-03-23 news

OpenAI scrapped a major Texas data centre expansion this week [1]. The stated reason: cost control ahead of a potential IPO.

This is the company that raised billions at an $80B+ valuation. The company that defined "AI at any cost."

I think the shift is telling: private investors ask "what's possible?" while public investors ask "what's profitable?" OpenAI is moving from one world to the other. Most enterprise AI strategies will eventually face the same transition.

Agentic AI cost risks are real. Gartner warned about what they called "Agentic Resource Exhaustion" where a single agent can burn thousands in compute costs in an afternoon through recursive reasoning loops [2]. OpenAI is likely experiencing this internally. Enterprises deploying agents should expect similar challenges.

Unit economics matter, even for AI leaders. If OpenAI, with ChatGPT's massive adoption, must demonstrate cost discipline before going public, internal AI projects elsewhere definitely need clearer economics. What does cost per inference look like? Cost per customer served with AI? Revenue impact versus cost to deliver?

OpenAI choosing current capacity over expansion suggests their existing infrastructure is sufficient for near-term product development. The implication: most organisations probably don't need to pre-build massive AI infrastructure "just in case." Build for known demand. Scale when demand materialises.

Google, Microsoft, and Meta are still projecting trillions in AI infrastructure investment over the coming decade [3]. Different business model, though. They sell infrastructure. OpenAI sells outcomes. Outcome businesses need unit economics. Infrastructure businesses can defer.

This is familiar territory for FinOps. The FinOps community spent years learning that cloud elasticity without governance creates runaway costs. AI agents are cloud elasticity with compounding loops.

Worth modelling AI unit economics if you haven't started already.

Sources and first draft AI, fine tuning Frank

[1] MLQ.ai, OpenAI Moderates Data Center Expansion Before IPO Launch
[2] Analytics Week, FinOps for Agentic AI
[3] Stocktwits, Google AI Infrastructure Spending