A multi-billion-dollar AI infrastructure project unveiled last month is set to drive a new wave of US AI innovation
On February 14, OpenAI formally rejected Elon Musk’s USD97bn takeover bid. The proposal was viewed as a potential obstacle to OpenAI’s ambitions of becoming a fully for-profit firm, along with its participation in Stargate — a private consortium planning to invest USD500bn in US artificial intelligence (AI) infrastructure, including massive data centres nationwide. Stargate’s endorsement by President Donald Trump last month has made it a symbol of his push for US dominance in AI.
What next
Subsidiary Impacts
- Stargate may be best regarded as a nascent AI version of a private-sector defence contractor such as Lockheed Martin or RTX.
- States will compete for the giant USD7-12bn facilities but will also face local opposition, despite the promise of new jobs.
- Hyperscale data centres’ energy demands could boost low-cost wind and solar power supply, regardless of Trump’s support for oil and gas.
Analysis
Stargate is a private consortium led by Oracle, OpenAI and Japan’s SoftBank — leading groups in global AI investment and development. The project aims to build both the physical and virtual infrastructure necessary to reinforce the US global leadership in AI. Initially investing USD100bn, Stargate plans to increase this to USD500bn by 2029.
MGX, an Abu Dhabi state investment fund, is an equity partner. Technology partners include Microsoft, Nvidia and ARM. The companies in the consortium know each other well. SoftBank’s Masayoshi Son is regarded as the project’s architect, although OpenAI’s Altman has become its prime mover.
Stargate was established and had already begun work on its first data centre (in Abilene, Texas, for Oracle) before Trump’s re-election. However, it was strategically beneficial for both the president and the consortium for him to announce the plan on January 21, just one day after his inauguration.
Trump secured an immediate flagship project for his policy of positioning private companies at the forefront of US AI dominance over China, earning significant credit for his support. The consortium, in turn, gained a head start over potential rivals, such as Musk’s AI venture xAI, as well as platform tech giants including Meta and Google, which are investing billions in data centres for their AI operations while also vying for Trump’s favour.
Data centres
US AI will have a significant demand for advanced semiconductors, electricity and water — all of which converge at massive data centres. Stargate plans to build 20 such centres with a capacity of over 1 gigawatt, starting with an initial roll-out of ten. These colossal facilities are known as hyperscale data centres.
The consortium is now looking for sites in 16 states. In early February, OpenAI put out requests for proposals (RFPs) for land, electricity, engineers and architects. The RFPs stressed the need for proximity to utility-scale power and water.
US data centre construction is already growing rapidly. In December, it was running at an annualised rate of USD31.5bn — three times the rate of December 2021.
Prior to the Stargate announcement, forecasts projected an 8-10% compound annual growth rate for data centre construction through at least the end of the decade. Despite Stargate’s vast ambitions however, its impact on overall data centre construction spending may be limited, as more than half of the investment is expected to be allocated to chip-making, power generation and water supply.
As such, Stargate will compete with other data centre projects and manufacturing facilities for key resources, including energy, skilled labour and construction workers. Additionally, both Stargate and the utility-scale power plants it requires will vie for investment — though SoftBank’s involvement helps mitigate that risk.
Stargate will compete with other data centre projects and manufacturing facilities for key resources
Even if China’s new AI model, DeepSeek, signals a shift towards less power-hungry AI, and data centre AIs learn to reduce electricity consumption through innovations such as shifting workloads between sites — known in the industry as ‘curtailment-enabled headroom’ — power demands will still be significant (see INT: DeepSeek launch flags AI power demand uncertainty – January 31, 2025). As a result, the growing trend of building data centres alongside power plants is likely to become standard practice.
The president has promised federal government assistance in obtaining adequate land to build Stargate’s data centres and the electricity to run them. This will mean cutting red tape, such as environmental reviews and speeding up construction permits and licensing for nuclear and hydrogen technologies.
Deployment
Stargate has provided little detail on its plans for the data centres once they are built. However, given the companies involved, the facilities are likely to support research and development for their frontier AI models and serve as the backbone of an operational AI cloud. What remains unclear is the balance between in-house use and commercial availability.
Some capacity will likely be made available to the US military — an enthusiastic adopter of AI — and US intelligence services, either as rented raw computational power or as contracted development services (see UNITED STATES: Trump AI policy – December 24, 2024).
Stargate’s scale and presidential endorsement will also give it a market advantage over other private companies eyeing lucrative AI defence contracts such as Google and Musk’s xAI.
Musk
Musk has a long-running feud with Altman that began with a boardroom rivalry over who should run OpenAI, which both men helped found. As soon as Stargate was announced, Musk denigrated it on his social media platform, X.
On 10 February, Musk made a hostile takeover bid of USD97.4bn for the non-profit OpenAI. Four days later, OpenAI’s board formally rejected the offer. While there are no precedents for such a large hostile bid targeting a charity, Musk’s offer was likely an attempt to disrupt Altman’s vision for OpenAI’s future — and potentially undermine Stargate as well.
China
The emergence of the open-source, energy-efficient Chinese model DeepSeek has posed a new challenge to Stargate, raising concerns over its massive energy consumption and financial outlay.
Despite claims that DeepSeek matches ChatGPT at a lower cost — claims which OpenAI denies — US investors, already uneasy about the long-term returns on billions invested in AI, have been rattled. However, building the infrastructure for AI — such as massive data centres, along with the chips, power and water to support them — is not the same as developing the AI models, such as DeepSeek, that will run on that infrastructure. Therefore, considerations of the impact of DeepSeek on Stargate and any refiguring of the calculus of AI investment demand caution.
Considerations of the impact of DeepSeek on Stargate demand caution
US export controls
DeepSeek and other Chinese models, such as Alibaba’s Owen, are reigniting the debate over US semiconductor export controls and their effect on the advancement of China’s domestic chip production (see INT: DeepSeek may prompt tighter US restrictions – February 5, 2025). However, the need for Chinese companies to access advanced chips at scale in order to compete with US firms suggests that the Trump administration is likely to tighten restrictions, while continuing to support substantial investments in domestic AI infrastructure projects.
