Companies retool infra
Firms are experimenting with new build strategies for AI capacity: Oracle is increasing AI spending and expanding energy partnerships, while Amazon is reportedly pursuing modular, factory‑built data‑centre designs to speed construction. Those moves are framed as responses to staffing, timing and energy constraints in the AI infrastructure race. (indiatoday.in) (fiduciarytech.com)
Oracle and Amazon are changing how artificial-intelligence data centers get built, with one locking in power first and the other trying to build server halls in factories. (bloomenergy.com) (datacenterdynamics.com) On April 13, Bloom Energy said Oracle plans to procure up to 2.8 gigawatts of fuel-cell capacity, with 1.2 gigawatts already contracted for United States projects and deployment underway into 2027. Oracle Cloud Infrastructure executive vice president Mahesh Thiagarajan said the systems are being used to meet customer demand across the country. (bloomenergy.com) Oracle’s spending push is tied to a much larger cloud expansion. In its March 10 fiscal third-quarter report, Oracle said remaining performance obligations reached $553 billion, cloud revenue rose 44% to $8.9 billion, and cloud infrastructure revenue rose 84% to $4.9 billion. (oracle.com) CNBC reported on April 14 that Oracle has raised more than $100 billion in debt to support data-center scaling, and Oracle shares rose more than 4% that day after the expanded Bloom deal. Bloom shares climbed 22% on the same session. (cnbc.com) Amazon Web Services is attacking a different bottleneck: construction itself. Business Insider reported, and Data Center Dynamics matched, that an internal effort called Project Houdini would shift more data-center assembly off-site into factories so capacity can come online faster. (businessinsider.com) (datacenterdynamics.com) Under that plan, Amazon would replace more traditional site-built work with prefabricated modules, sometimes called skids, built in controlled factory settings and then installed on-site. Mint, citing the Business Insider report, said the approach could cut some build phases from 15 weeks to two or three weeks and reduce labor hours by the tens of thousands. (livemint.com) (businessinsider.com) The common problem is that artificial-intelligence computing needs both buildings and electricity at a pace utilities and contractors often cannot match. Bloom said its fuel cells can be deployed faster than traditional power connections, while Amazon’s reported factory-build plan is meant to bypass on-site labor and scheduling delays. (bloomenergy.com) (datacenterdynamics.com) Oracle and Bloom have been moving in this direction for months. Bloomberg coverage carried by Yahoo Finance said the initial 1.2 gigawatts will be used this year and in 2027, and Bloom said it delivered a fully operational fuel-cell system to Oracle in 55 days last year, ahead of a 90-day schedule. (finance.yahoo.com) (bloomenergy.com) Amazon has not publicly laid out Project Houdini in its own filings, but an Amazon Web Services spokesperson told Business Insider that its construction changes are aimed at delivering artificial-intelligence infrastructure faster and at lower cost. That leaves the race looking less like a pure chip contest and more like a scramble for megawatts, modules, and crews. (businessinsider.com)
