The Stargate Project: A Manhattan Project for the AI era

President Trump Announces Joint Venture with OpenAI, SoftBank and Oracle to Build $500B Worth of AI-dedicated Data Centers

On his second day in office, President Trump approved funding for the Stargate Project, a joint venture with OpenAI, SoftBank and Oracle to initially build a $100 billion data center in Texas. Over the next four years, the project aims to expand into additional large-scale data centers, with a total of $500 billion in funding, making it the largest centralized data center investment in history. The funding includes significant financial backing from the U.S. government  with contributions from SoftBank, a firm known for its long-term investment strategies. OpenAI, SoftBank, Oracle and MGX will be the initial equity investors, while Arm, Microsoft, NVIDIA and OpenAI have been named as technology partners and will have some involvement in Stargate.
 
The $500 billion expenditure is unprecedented. Most Tier 1 hyperscaler data center projects are valued in the single-digit billions, making Stargate’s Phase 1 cost more than 50 times higher than its closest comparable. This venture will rely heavily on U.S. government funding, as SoftBank’s Vision Fund, which manages assets worth less than $200 billion, cannot shoulder the full burden. This positions Stargate as a “Manhattan Project” for the AI era, as it represents one of the largest technological undertakings in modern times. The project is poised to reshape global dynamics if it can navigate the significant hurdles that lay ahead. Regardless, OpenAI, equipped with the world’s largest AI cluster, will pursue its goal of artificial general intelligence (AGI) while enjoying unparalleled access to the compute infrastructure needed to push parameter counts higher.

 

 

What Does This Mean for Cloud Vendors and Model Developers?

OpenAI Gains a Powerful Competitive Advantage Over Other Model Developers

Being the largest single technology endeavor in recent history, the Stargate Project will have a notable influence on the budding AI market. In TBR’s opinion, no company stands to benefit more than OpenAI. The company’s access to dedicated compute resources will be unmatched in the model developer industry, enabling OpenAI to push parameter counts further and faster compared to peers and supporting the company’s objective of differentiating based on large language model (LLM) performance. OpenAI’s ultimate goal is to reach AGI, which the company defines as “highly autonomous systems that outperform humans at most economically valuable work.” While estimates about when OpenAI will reach this goal vary widely, the Stargate Project fulfills a critical requirement in the pursuit of AGI, making OpenAI a top contender to reach AGI before other firms.

Another Win for OCI

Oracle’s reentry into the already highly saturated IaaS market with Gen2 OCI (Oracle Cloud Infrastructure) has been widely successful. Though initially designed to drive stickiness with enterprises already rooted in the Oracle ecosystem, OCI appears to be gaining traction among digital natives and ISVs for AI use cases. In many ways, this is a testament to Oracle’s decision to cozy up to NVIDIA for GPUs early on in the emergence of generative AI (GenAI) by hosting the company’s DGX software in its data centers, helping NVIDIA position as a software company and avoid becoming another piece of commoditized hardware locked into the hyperscalers’ stacks. Now, riding a wave of over $70 billion in cloud backlog, OCI is Oracle’s fastest-growing business and will soon become its largest.
 
Aside from the GPUs, the other factor fueling OCI growth and granting Oracle its status as one of the largest data center operators in the world is the company’s ability to bring new data centers online extremely quickly. This is because Oracle has generally adopted a strategy of building more but smaller-scale (approximately 145-kW) data centers, with a focus on ensuring that, outside of scale, all Oracle data centers are identical.
 
This scale can vary significantly, though, and with the current AI wave, we have seen Oracle prioritize larger data centers, some spanning 800MWs for AI customers, including OpenAI. There was perhaps no greater testimonial for OCI as a cloud credible enough for AI applications than OpenAI’s mid-2024 decision to leverage OCI for AI training via Azure.
 
To be fair, the existing Azure-Oracle relationship influenced that decision, but OpenAI has made it clear that the company not only needs IaaS services to push the boundaries of its models but also needs them quickly, regardless of who provides them. The Stargate Project would only advance the OCI-OpenAI connection, bringing new workloads to OCI and sending a message to the market that OCI is also in the game with Amazon Web Services (AWS), Microsoft Azure and Google Cloud when it comes to AI workloads.

Oracle Sets Its Sights on Healthcare as the Company Pursues AI Opportunities for Cerner

Oracle CTO Larry Ellison’s remarks at the White House on Jan. 21 were brief but telling of where Stargate could go, with the Oracle co-founder highlighting AI’s role in modern electronic health record (EHR) management and healthcare transformation at large. Of course, Oracle entered the healthcare market over two years ago with its acquisition of Cerner and has since modernized Cerner Millenium on OCI in hopes of delivering a new cloud-based system that will challenge decades-old EHR systems. This includes an EHR system that can support disease-specific AI models that, importantly, are developed not by Oracle but by medical professionals with expertise in said diseases. The details and timeline around Stargate are still vague but stand to advance Oracle’s push in AI, including within the healthcare vertical, which perhaps has among the most to gain from AI’s potential.
 
When the Cerner deal closed, Ellison was very clear about plans to have a standardized database that unifies fragmented data so medical professionals can instantly access EHR data, regardless of what type of EHR system it is, within a single database. At the time, we wrote about the obvious roadblocks to overcome, including the security & compliance concerns and need to obtain legislative backing. Since Ellison’s initial remarks, we have not heard much of an update on this vision, but Stargate and what seems to be Oracle’s rising role in the new administration (stay tuned as we track any potential Oracle-TikTok developments) could help move this vision along.

While Microsoft Has Been a Close Ally of OpenAI, the Bond That Was Forged in the First Year of GenAI’s Time in the Spotlight Has Weakened

So, how did we get here? Rumors of the Stargate Project date back to March 2024, when OpenAI CEO Sam Altman outlined ambitions for a $100 billion data center in partnership with Microsoft. At the time, the partnership seemed logical, given the companies’ long-standing relationship and Microsoft’s significant equity stake in OpenAI. However, the dynamics have shifted. In October 2024 Altman publicly criticized Microsoft for its slow progress in building AI-dedicated infrastructure, an issue corroborated by reports of persistent infrastructure shortages from Microsoft management. OpenAI’s latest announcement reflects the outcome of this strained relationship, as Azure’s exclusivity agreement with OpenAI has been officially amended, granting OpenAI the right to seek alternative delivery agreements if Microsoft fails to meet its compute demands. Oracle is the first cloud provider OpenAI has turned to, leveraging Oracle’s substantial capacity for AI workloads and an increasingly strategic relationship with Microsoft.
 
OpenAI’s shift toward Oracle is a setback for Microsoft but does not entirely diminish the hyperscaler’s leadership in AI. OpenAI remains a close partner, and Microsoft is well positioned to grow its AI-related IaaS revenue as the company continues expanding its infrastructure. Furthermore, Microsoft’s SaaS portfolio serves as a key delivery mechanism for OpenAI’s models, and the company retains a significant equity stake in OpenAI.
 
These factors are likely to sustain the strategic partnership between the two entities for the foreseeable future. Although Microsoft is not a member of the Stargate joint venture, it is listed as a strategic technology partner, and TBR expects its platforms and software to play a role in the project.
 
Additionally, while Microsoft may have less influence over OpenAI’s hosting decisions, Oracle and Azure remain deeply interconnected. For instance, Oracle now uses Azure data centers to house its database hardware through the Oracle Database@Azure Service. This setup could theoretically integrate Azure OpenAI into AI development as customers bring enterprise data from Oracle into the Azure cloud.

Stargate’s First Phase Includes the Construction of a Massive $100B Data Center, the Largest GPU Cluster in the World

Why Build an AI Megacluster?

OpenAI’s primary motivation for increasing computational resources is to meet the exponential demands of training models with higher parameter counts. Scaling up these parameters allows models to process larger quantities of data, often referred to as context windows. As a context window expands, model outputs improve in quality and accuracy. The prevailing belief is that pushing parameter counts far enough will enable models to exhibit the capabilities defined in OpenAI’s vision of AGI. With full financial backing from the U.S. government, OpenAI’s pursuit of AGI appears more achievable. The result would likely be a versatile GenAI back-end architecture capable of transforming process automation in SaaS workflows. However, in the short term, OpenAI’s focus on parameter scaling keeps its AI strategy centered on general-purpose LLMs, rather than more specialized small language models (SLMs). This approach makes OpenAI’s models particularly suited for productivity tools and customer service applications, while specialized models may dominate in niche workflow tasks.

Possible Stargate Constraints

While unparalleled access to GPUs and compute infrastructure is a major advantage for OpenAI’s model training strategy, there are still several factors that need to be addressed alongside the data center construction. First, the Stargate Project initially intends to absorb Oracle’s privately funded, in-progress data center in Abilene, Texas, yet TBR believes this could heighten power supply challenges. While power transmission constraints are widespread, Texas’ power grid has had issues in the past, such as in the winter of 2021, due to the fact that Texas’ power grid, unlike the other 47 states in the continental U.S., does not connect to the two major national grids. This prevents access to backup power generated outside the state and poses a risk of a significant outage. To mitigate these power concerns, TBR believes developing alternative power sources, namely nuclear power, will be a priority early in the Stargate Project.
 
Second, having access to effective training data remains a persistent need within the model developer market. While OpenAI has been forthcoming in striking deals with internet platforms and media sources, some speculate that the corpus of data available to train LLMs and large multimodal models (LMMs) could soon be completely used up. The use of synthetic data has often been proposed to overcome this hurdle, yet this path brings separate issues like greater AI hallucinations and model drift. Altogether, while securing project financing is the first step, working around these constraints will challenge OpenAI as it pursues AGI, and the innovations created in the Stargate Project will need to reach beyond simply building the largest AI-dedicated data centers in the world.
 
Financing could also prove to be a risk. In response to Trump’s executive action, Elon Musk, a Trump insider and co-head of the new Department of Government Efficiency, publicly shared his belief that the U.S. government does not have enough money to fund the project. Of course, Musk has a bias as the founder of startup xAI, but nevertheless, the Stargate Project does have a staggering price tag. Still, with the Republican Party’s control of the legislative and executive branches, there will be fewer political barriers to financing the Stargate Project based on the assumption that AI supremacy is of greater strategic importance.

Conclusion

The Stargate Project marks a significant development in AI infrastructure, with OpenAI, SoftBank, Oracle and the U.S. government collaborating to create a network of AI-dedicated data centers. With a planned $500 billion investment, this initiative seeks to address the increasing computational demands of AI model development, positioning OpenAI to advance toward its goal of achieving AGI.
 
Oracle’s involvement, bolstered by partnerships with NVIDIA and its advancements in cloud infrastructure, highlights its growing role in the AI ecosystem and could advance Oracle’s Cerner ambitions. However, the project faces notable challenges, including substantial energy requirements and concerns over the availability of high-quality training data, which will require innovative solutions to address.
 
As one of the largest technology projects in recent history, Stargate reflects the evolving priorities in AI development and the broader strategic implications for technological leadership.