OpenAI has hired former Intel executive Sachin Katti to orchestrate its colossal AI infrastructure buildout.
This strategic move signals a new phase in OpenAI's quest for artificial general intelligence, where ambition now confronts the hard realities of physical supply chains. Sam Altman's vision of spending "trillions" on AI requires more than just capital; it needs a massive, coordinated effort to secure chips, memory, data center space, and, most critically, power. OpenAI is centralizing this monumental task under Katti, a seasoned engineer, to navigate the complex web of constraints that could stall its progress.
The first major hurdle is the supply chain for computing hardware. While Nvidia's GPUs are the stars of the show, the unsung hero—and current bottleneck—is HBM (High Bandwidth Memory). Memory producers like Micron and SK hynix have reported that HBM is effectively sold out well into 2026. This scarcity makes securing a steady supply of memory just as critical as getting the latest GPUs. Katti's role is to forge relationships and lock in multi-year contracts across a portfolio of suppliers to de-risk this dependency.
Second, there's the power problem. AI data centers are incredibly power-hungry. Recent reports from energy research institutes warn that data centers could consume up to 17% of all U.S. electricity by 2030. This staggering demand is straining local power grids, making it difficult to find suitable locations with adequate energy, water, and transmission infrastructure. Katti’s job is therefore not just about technology procurement, but also about energy strategy—negotiating massive Power Purchase Agreements (PPAs) and selecting sites across multiple states and countries to avoid bottlenecks.
To address these challenges, OpenAI is aggressively diversifying its infrastructure partners. For years, it relied heavily on Microsoft Azure. But now, it has committed tens of billions of dollars to secure capacity from Oracle, AWS, and specialized "neocloud" providers like CoreWeave. This "all-of-the-above" strategy provides crucial optionality and ensures that a single vendor's hiccup won't derail its entire roadmap. Katti's appointment is the final piece of the puzzle, creating a central command to manage this sprawling, multi-vendor ecosystem and turn a trillion-dollar vision into functional compute power.
- HBM (High Bandwidth Memory): A type of high-performance computer memory used alongside GPUs to quickly process large amounts of data, essential for training large AI models.
- PPA (Power Purchase Agreement): A long-term contract between an electricity generator and a customer (like a data center) to purchase energy at a pre-negotiated price.
- Foundry: A company that manufactures semiconductor chips for other companies that design them. For example, TSMC is a foundry that makes chips for Nvidia and Apple.