OpenAI is aggressively expanding its global data-center footprint and computing capacity, aiming for about 1.9 gigawatts of capacity in 2025 and eyeing a potential gigawatt of new infrastructure each week as part of a broader push to dramatically scale its AI infrastructure.

Sam Altman has framed a trillion-dollar-scale infrastructure vision for AI, aligning rapid expansion with supply-chain diversification to meet rising demand for AI compute.

Katti notes a cautious but steady investment stance in memory and wafer markets, as suppliers seek long-term demand assurances amid boom-bust cycles.

Katti’s background in industrial computing and his role at OpenAI tie directly to the ongoing infrastructure initiatives and the company’s hardware strategy.

Despite diversification efforts, Nvidia remains a core training-hardware supplier, with plans to deploy roughly 1 gigawatt of computing using Vera Rubin accelerators later this year.

OpenAI is building facilities across the United States, Europe, the Middle East, and South America, and has paused a Stargate data-center project in Abilene, Texas to pursue faster deployment paths using newer Nvidia chips.

The company is expanding internationally and exploring multiple projects beyond a single large site, aiming to accelerate capacity with newer Nvidia chips and other suppliers like Broadcom, Cerebras, and AWS Trainium.

OpenAI is pursuing a diversified hardware supply chain, signing multibillion-dollar agreements with Cerebras and leveraging Amazon’s Trainium, while developing its own processors through a Broadcom partnership; Nvidia remains central for training.

Katti’s leadership focuses on guiding OpenAI’s infrastructure and compute strategy, shaping how the company builds and scales its systems.

The hire of Katti reflects OpenAI’s broader efforts to optimize and scale infrastructure to support future deployments.

The push to scale infrastructure faces constraints like power-grid limits, memory shortages, geopolitical risks, and community resistance to large data centers.

No personal anecdotes—only professional responsibilities and role-related details are noted.