OpenAI to Increase Computing Power from 10GW to 30GW by 2030
In its latest public statement, OpenAI reviewed its commitment made in January 2025 to build a 10GW computing power infrastructure. Currently, it has implemented over 8GW of data center and hardware deployment capacity, stemming from the Stargate program and collaborations with partners like NVIDIA and Broadcom. The company disclosed that since 2023, its internally available computing power has increased from about 0.2GW to nearly 2GW, growing significantly faster than the traditional cloud infrastructure expansion pace. Based on this, OpenAI is raising its target to achieve approximately 30GW of computing power by 2030, aligning with investment bank and media reports of "investing hundreds of billions to over a trillion dollars by the end of this century for computing power and data center construction."
According to information released by OpenAI Global Affairs, this 30GW target is seen as a key leverage for gaining long-term computing power advantages amid an "exponential growth in demand for intelligent systems." The company states that computing power is the third major structural moat alongside data and algorithms. By securing electricity, land, chips, and cooling resources in advance, it aims to maintain a continuous lead in future competitions involving large models and intelligent systems. External analysis indicates that a 30GW level data center's total load is equivalent to the peak electricity consumption of multiple large cities, which will create cross-cycle demand for the global semiconductor, energy, and infrastructure industries. It will also elevate issues surrounding AI's energy consumption, carbon emissions, and grid security to new policy heights.
Source: Public Information
ABAB AI Insight
OpenAI's goal of increasing from 10GW to 30GW essentially attempts to build a closed loop of "scale advantage → cost advantage → model advantage" at the computing power level. For large models, the marginal cost of training and inference heavily depends on the scale of proprietary computing power and procurement bargaining power. Once several critical capacity thresholds are crossed, the amortized cost per unit of computing power will significantly decrease, creating long-term cost suppression against competitors. In this structure, computing power is not just a production factor but is evolving into an infrastructure with network externalities, similar to "railroads + electricity."
From a global industrial division perspective, a 30GW level AI data center cluster means that multiple industry chain segments, such as semiconductors, packaging and testing, optical modules, power equipment, and cooling systems, will welcome "ultra-long-term order clusters" centered around OpenAI as the core demand side. This will strengthen the deep binding between leading chip and cloud infrastructure manufacturers and OpenAI, sharing profits through long-term supply agreements and joint investments, dispersing the financial risks of computing power expansion to a broader capital market and industrial partners.
On a deeper level, the geopolitical implications of computing power expansion are also significant. An AI infrastructure targeting 30GW equates to building a highly concentrated "smart industrial belt" on electricity, land, cooling water, and network nodes, with energy consumption and land scale comparable to traditional heavy industries or large urban clusters. This will force governments to reassess the relationship between AI infrastructure and grid planning, as well as the layout of nuclear and renewable energy—electricity will no longer just serve existing industries but will preemptively lay out capacity for the yet-to-be-fully-formed "smart economy."
From a long-term historical perspective, humanity has experienced multiple energy and computing power leaps over the past few centuries, from "steam power—electricity—information computing." Today, OpenAI's pursuit of 30GW is not just an expansion plan for a single company but resembles a "new infrastructure movement" led by private entities. If intelligent systems penetrate production and life as expected, whoever controls such ultra-large-scale computing power clusters will hold higher levels of pricing power and agenda-setting authority in future knowledge production, automated decision-making, and economic coordination.