OpenAI Raises 2030 Computing Expenditure to $750 Billion and Launches Georgia Data Center
OpenAI has raised its expected computing resource expenditure for 2030 from $600 billion to approximately $750 billion, while launching the Camellia data center project in Georgia, USA.
The initial investment for this project is $20 billion, with an expected power supply of 3.2 gigawatts from 2028 to 2032, and total costs may exceed $30 billion; the company has signed a 6 gigawatt agreement with Oracle, expanded its AWS collaboration to $13.8 billion, and plans to invest an additional $250 billion in Microsoft Azure.
Market mechanisms indicate that AI training demand drives capital-intensive investments from hyperscalers and chip manufacturers, benefiting power and computing suppliers. OpenAI diversifies risks through multiple suppliers, but if revenue growth does not keep pace with expenditures, it will face cash flow pressure, while latecomers like AMD are entering the supply chain through partnerships with Anthropic.
Source: Public Information
ABAB AI Insight
OpenAI previously achieved early computing power explosion through significant financing from Microsoft and binding with Azure. This further diversification to Oracle and AWS, along with an additional $250 billion investment, continues its "burning money for scale" strategy, similar to the early infrastructure phase of Amazon AWS.
In terms of capital strategy, OpenAI locks in power and chip supply through long-term agreements, while AMD invests up to $5 billion in exchange for procurement of Anthropic's MI450 series, creating an interlocking resource delivery system that accelerates expansion in training clusters for both parties.
Similar to Google's TPU self-research + multi-cloud strategy or Amazon's early heavy investment in data centers, OpenAI and Anthropic are currently in a transition phase of AI infrastructure from reliance on Nvidia to a diversified supply chain.
Essentially, this represents capital concentration and industrial chain restructuring: large prepayment agreements concentrate computing resources among a few tech giants, while power, chip, and cloud vendors secure pricing power through binding agreements, simultaneously pushing the entire AI supply chain from a single GPU dependency to a multi-architecture coexistence structure.
ABAB News · Cognitive Law
The higher the expected computing expenditure, the greater the pressure to match revenue.
When burning money for scale, multiple suppliers are the true moat.
In capital-intensive tracks, securing resources is more sustainable than chasing technological leadership.