Report: OpenAI projects $750 billion in computing power spending by 2030! The company still claims "computing power is severely insufficient."

OpenAI has raised its computing power spending to $750 billion, with supply gaps pushing the AI infrastructure sector to the brink of stress testing.

On September 9, according to a report by DealBook of The New York Times, citing an unnamed source familiar with the data, OpenAI still feels that its computing power is "very insufficient".

OpenAI expects its computing power spending to reach approximately $750 billion by 2030, an upward revision of more than 25% from its previous forecast of approximately $600 billion.

This statement implies that the actual supply of chips, data centers, power, and network resources has not kept pace with the growing demand for the development and deployment of leading models.

For companies like Oracle and CoreWeave, which have staked massive infrastructure expansion on AI demand, this is both a demand signal and a direct source of balance sheet pressure.

The $750 billion plan reflects an upgrade in procurement scale.

As Wall Street Insights noted , on July 22, The Wall Street Journal reported that as the demand for AI computing power continues to rise, OpenAI has raised its future cloud infrastructure spending forecast to $750 billion, indicating that it continues to increase its long-term capital investment.

OpenAI has previously signed capacity agreements with several partners. Among them, the data center capacity contract signed with Oracle totals 6 gigawatts, most of which is still under development.

OpenAI has also extended its multi-year cloud computing agreement with Amazon Web Services to eight years and $138 billion, which includes 2 gigawatts of computing power based on Trainium chips.

In addition, OpenAI previously pledged to increase its cloud spending on Microsoft Azure by $250 billion, but did not disclose the specific timeline for this spending. The combination of multiple large contracts indicates that OpenAI is securing long-term computing resources through a multi-vendor model.

However, the expansion of computing power procurement commitments also means that OpenAI will need to continue to coordinate the relationship between demand growth, infrastructure construction and funding payments in the coming years.

Order growth coexists with financial pressure

OpenAI’s computing power needs provide long-term order support for cloud service providers and infrastructure supply chains, but expansion costs also put pressure on corporate balance sheets.

For example, Oracle's remaining obligations reached $638 billion, a year-on-year increase of 363%. Of this, $75 billion relates to prepayments or arrangements for providing GPUs to customers. Oracle Cloud Infrastructure revenue reached $5.79 billion, a year-on-year increase of 93%.

However, the order backlog is accompanied by high capital expenditures. Oracle's free cash flow for the fourth fiscal quarter was negative $23.69 billion, capital expenditures were $55.66 billion, and it projects net cash outflows of approximately $70 billion for fiscal year 2027. The company plans to raise approximately $40 billion through debt and equity financing, including a $20 billion plan to issue shares at market price.

CoreWeave faces a similar situation. The company reported negative free cash flow of $5.743 billion in the second quarter, an order backlog of approximately $104 billion as of June 30, and a target of exceeding 8 gigawatts of active power capacity by 2030. Its expansion is primarily supported by debt, upfront payments, and equity financing.

Supply constraints remain concentrated in chips, power, and data centers.

On the chip supply side, Nvidia CEO Jensen Huang recently stated that the supply chain is under strain, with current supply only able to meet about 70% of demand.

Broadcom is advancing the deployment of Jalapeno, OpenAI's first-generation custom accelerator, with plans to deploy 1.3 gigawatts by fiscal year 2027. However, the company also points out that land, power, and data center building shells will determine the actual timeline for capacity rollout.

This means that the bottleneck in AI infrastructure is not limited to GPU supply. Wafers, high-bandwidth memory, substrates, power, land, and data center construction capabilities can all affect the final delivery of computing power.

OpenAI's latest statement regarding the continued shortage of computing power reinforces the demand signal for AI infrastructure and presents the market with a more realistic test: whether huge orders can be converted into usable gigawatt-level computing power as planned, and whether related companies can maintain a balance between financing and cash flow during a period of high capital expenditure.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.