Running an AI product demands significant computational resources. As the tech sector competes to harness AI model capabilities, a parallel race is occurring to develop the infrastructure supporting them. Nvidia CEO Jensen Huang recently estimated during an earnings call that AI infrastructure spending will reach $3 to $4 trillion by decade’s end, largely funded by AI firms. This surge places considerable pressure on power networks and stretches the industry’s construction capacity to its limits.
Below, we detail the key AI infrastructure projects, including substantial investments from Meta, Oracle, Microsoft, Google, and OpenAI. We will update this information as the boom progresses and figures rise.
Microsoft’s $1 billion investment in OpenAI
This deal arguably initiated the current AI boom: In 2019, Microsoft invested $1 billion in OpenAI, a non-profit known for its association with Elon Musk. Crucially, Microsoft became OpenAI’s exclusive cloud provider. As model training demands increased, Microsoft’s investment shifted towards Azure cloud credit instead of cash.
It benefited both: Microsoft reported increased Azure sales, while OpenAI gained funds for its major expense. Over subsequent years, Microsoft increased its investment to nearly $14 billion, a move poised for significant returns when OpenAI becomes a for-profit entity.
The partnership has recently evolved. In January, OpenAI announced it would no longer exclusively use Microsoft’s cloud, granting Microsoft first refusal on infrastructure needs but exploring alternatives if Azure couldn’t meet them. More recently, Microsoft started exploring other foundation models for its AI products, further distancing itself from the AI leader.
OpenAI’s successful arrangement with Microsoft has led to AI services commonly partnering with specific cloud providers. Anthropic has received $8 billion from Amazon, making kernel-level hardware adjustments for better AI training. Google Cloud has also enlisted smaller AI companies like Lovable and Windsurf as “primary computing partners,” without investment. OpenAI also secured a $100 billion investment from Nvidia in September, enabling further GPU purchases.
The rise of Oracle
On June 30, 2025, Oracle disclosed in an SEC filing a $30 billion cloud services agreement with an unnamed partner, exceeding its previous fiscal year’s cloud revenues. OpenAI was later identified as the partner, positioning Oracle alongside Google as an OpenAI hosting partner post-Microsoft. Consequently, Oracle’s stock surged.
Techcrunch event
San Francisco
|
October 27-29, 2025
Months later, on September 10, Oracle announced a five-year, $300 billion compute power deal starting in 2027. Oracle’s stock rose further, briefly making founder Larry Ellison the wealthiest individual globally. The deal’s scale is immense: OpenAI lacks $300 billion in spending capacity, suggesting substantial growth for both and considerable confidence.
Regardless of actual spending, the deal has established Oracle as a premier AI infrastructure provider and a significant financial entity.
Nvidia’s investment spree
As AI labs rush to develop infrastructure, they predominantly purchase GPUs from Nvidia. This demand has enriched Nvidia, leading to unconventional reinvestments. In September 2025, Nvidia acquired a four percent stake in Intel for $5 billion. More surprisingly, it struck deals with its customers. A week after the Intel deal, it announced a $100 billion investment in OpenAI, paid in GPUs for OpenAI’s data center projects. Nvidia has since announced a similar arrangement with Elon Musk’s xAI, and OpenAI launched a separate GPU-for-stock arrangement with AMD.
This arrangement is circular. Nvidia’s GPUs are valuable due to scarcity. By trading them into a growing data center scheme, Nvidia ensures scarcity. Similarly, OpenAI’s privately held shares are valuable because they are inaccessible via public markets. Currently, OpenAI and Nvidia are successful, but scrutiny may increase if momentum slows.
Building tomorrow’s hyperscale data centers
For companies like Meta with existing infrastructure, the situation is complex but equally costly. Mark Zuckerberg stated Meta intends to spend $600 billion on U.S. infrastructure by 2028.
In the first half of 2025 alone, the company spent $30 billion more than the previous year, primarily due to its increasing AI ambitions. Some spending goes towards cloud contracts, such as a recent $10 billion deal with Google Cloud, but significant resources are being invested in two new massive data centers.
A new 2,250-acre site in Louisiana, named Hyperion, is projected to cost $10 billion and provide 5 gigawatts of compute power, including an arrangement with a local nuclear plant to manage the energy load. A smaller site in Ohio, called Prometheus, is planned to launch in 2026, powered by natural gas.
This buildout has environmental costs. Elon Musk’s xAI built a hybrid data center and power-generation plant in South Memphis, Tennessee, which has quickly become a major emitter of smog-producing chemicals due to natural gas turbines that experts believe violate the Clean Air Act.
The Stargate moonshot
Two days post his second inauguration, President Trump announced a joint venture between SoftBank, OpenAI, and Oracle to invest $500 billion in U.S. AI infrastructure. Named “Stargate” after the 1994 film, the project was heavily promoted, with Trump calling it “the largest AI infrastructure project in history.” Sam Altman concurred, calling it “the most important project of this era.”
SoftBank would provide funding, Oracle would handle the buildout with OpenAI’s input, and Trump would remove regulatory obstacles. However, doubts arose, including from Elon Musk, who claimed insufficient funds.
As the hype diminished, the project lost momentum. In August, Bloomberg reported partner disagreements. Nonetheless, construction of eight data centers in Abilene, Texas, continues, with completion expected by the end of 2026.
This article was first published on September 22.
