It’s been an eventful September for OpenAI. In the space of a fortnight, OpenAI made two megadeals: a $300 billion, five-year cloud contract with Oracle and a strategic partnership with Nvidia worth up to $100 billion.
OpenAI’s leadership argues that to reach artificial general intelligence (AGI) and to keep pace with demand for ChatGPT and other models, the world needs a new layer of AI infrastructure: gigawatt-scale data centres, vast energy supplies, and purpose-built chips. Something that can cost hundreds of billions, if not trillions, of dollars.
In this article, we take a closer look at the two deals and the AI super-infrastructure they are enabling.
OpenAI’s $300 Billion Deal with Oracle…
On September 10, OpenAI signed a five-year agreement with Oracle to purchase $300 billion worth of cloud computing capacity, making it one of the largest deals in the history of the cloud industry. The deal, starting in 2027, will provide OpenAI with 4.5 gigawatts of computing power under the Stargate umbrella. For context, 4.5 gigawatts is roughly the electricity needed to power about four million US homes.
The announcement helped Oracle propel its stock up by more than 40% in a single week, and briefly made its founder, Larry Ellison, the richest man in the world. Oracle’s remaining performance obligations (RPO)—contracted but not yet delivered revenue—soared to $455 billion, with most of that tied to OpenAI. While RPO is not cash recognised today, the backlog has excited markets.
For OpenAI, the deal represents another step away from exclusive reliance on Microsoft Azure, its main cloud provider since 2019. By partnering with Oracle, OpenAI hopes to secure the compute it needs to scale ChatGPT and other models and to mitigate the supply bottlenecks that have hampered previous product launches.
… and $100 Billion Deal with Nvidia
On September 22, OpenAI announced a strategic partnership with Nvidia to deploy 10 gigawatts of next-generation AI datacentres built around millions of Nvidia’s GPUs. But if you read closely into the announcement, you’ll notice that what was announced was a letter of intent, under which Nvidia will provide up to $100 billion in funding. Crucially, this will be released in $10 billion tranches for each gigawatt of capacity that comes online. The first phase, about 1 GW, is scheduled for the second half of 2026 on Nvidia’s new Vera Rubin platform.
The deal was reportedly hammered out in last-minute negotiations between Altman and Nvidia founder Jensen Huang, finalised during President Trump’s state visit to the UK—hours before Altman flew to Texas to unveil OpenAI’s Stargate plans at Abilene.
It is worth stressing that this is only a letter of intent; no money or GPUs have yet been exchanged. Nevertheless, news of the deal was enough to raise Nvidia’s valuation by $177 billion.
Economics Enter the Chat
The megadeals with Oracle and Nvidia highlight a problem OpenAI has to solve, and that is its infrastructure ambitions far outstrip its current earnings.
OpenAI is not profitable. The company is burning billions of dollars on keeping its infrastructure up and running and has warned investors it expects to lose $44 billion before reaching breakeven. But its revenue is growing. In 2024, the company made $3.7 billion in revenue. This year, the revenue is projected to triple, reaching $12.7 billion. In 2026, the company might reach $30 billion in revenue. By 2029, OpenAI expects to reach $125 billion in revenue, and that is the point at which it is projected for OpenAI to become profitable.
We do not have access to OpenAI’s internal data, but it is reasonable to assume the company sees a future point where projected revenue and the cost of serving its models balance out. That point, however, remains distant, and OpenAI needs billions now to stay afloat and invest in its AI infrastructure.
For OpenAI, reducing infrastructure costs is key to reaching profitability. Currently, the most effective cost reductions are likely to come not from software, but from hardware—through massive AI-focused data centres and custom AI chips.
Custom AI chips, designed specifically to handle the types of workloads used by OpenAI, will certainly help. OpenAI is already making steps here by announcing a $10 billion deal with Broadcom to develop custom AI chips. However, the vast majority of improvements are expected to come from building massive, AI-focused data centres.
Stargate and Building the US AI Super-Infrastructure for OpenAI
The key to OpenAI’s future is Stargate, its overarching, multi-gigawatt AI infrastructure project. It is a network of next-gen data centres designed to deliver roughly 10 GW of compute and hundreds of billions of dollars of investment to train and run future models. OpenAI estimates that the Stargate Project will cost $500 billion over the next four years.
The flagship Stargate complex in Abilene, Texas, is already under construction. It is being built in partnership with Oracle and developed by Crusoe Energy. The campus is planned to provide around 1.1–1.2 GW at full build-out, with eight huge data-hall buildings, and ultimately housing over 400,000 GPUs. The Abilene facility is already partially running on Oracle Cloud Infrastructure, and it is being used by OpenAI for training and inference workloads.
But that is just the beginning, as OpenAI announced five additional Stargate sites in the US. Three sites, in Shackelford County, Texas, Doña Ana County, New Mexico, and an undisclosed Midwest location, will be developed with Oracle. Two sites, in Lordstown, Ohio, and Milam County, Texas, will be developed in partnership with SoftBank’s SB Energy. Additionally, there are also overseas initiatives: Stargate UAE, Stargate Norway, and the recently announced Stargate UK.
Altogether, the current US pipeline brings announced Stargate capacity close to 7 GW. For an estimated $50 billion per GW, the current price tag for Stargate is $350 billion.
However, that might not be enough for OpenAI, with one executive saying the company will need more than 20 gigawatts, for a total of $1 trillion, to meet the demand for ChatGPT. The Wall Street Journal quotes an OpenAI executive who said that the demand is likely to reach 100 gigawatts, which would cost $5 trillion to achieve.
Scepticism Enters the Chat
I want to pause for a moment and let those numbers sink in. We are talking about investments amounting to hundreds of billions—if not trillions—of dollars. The energy required to power these massive, gigawatt-scale AI data centres is equally staggering. One gigawatt is roughly the output of a large nuclear reactor, and OpenAI executives are discussing power needs of 7 GW, 10 GW, 20 GW or even 100 GW.
All of these resources and efforts are devoted to serving just one company—OpenAI.
Such massive, audacious deals and projects inevitably brought scepticism and proof that the AI Bubble is real (whenever we are in an AI bubble is a story for another time).
The $300 billion Oracle contract—worth many times OpenAI’s current annual revenue—was described by AI critic Gary Marcus as “peak bubble”, while technology commentator Ed Zitron called it “a grotesque attempt to mislead investors”. Both questioned whether OpenAI could possibly fund such a contract, and whether Oracle could even deliver the required capacity on time.
Sceptics point out that the Oracle contract represents future obligations rather than guaranteed income. In fact, much of the AI sector’s expected growth is based on forecasts of demand, not yet on proven, profitable usage.
Oracle, despite its surging stock, will likely have to take on debt to finance the purchase of GPUs and the construction of datacentres. Additionally, it is putting a large chunk of its future revenue on one customer—OpenAI. This could pay off massively if everything goes right, but it can also spell massive problems if it does not.
OpenAI, meanwhile, needs this enormous computing power to push itself into profitability. Before that happens, it runs mostly on future projections and hype. If the roadmap created by OpenAI’s leadership slips, the entire vision will crumble. Additionally, the deal with Nvidia is conditional on construction milestones. Any delays in building datacentres could slow the entire rollout, leaving OpenAI’s ambitions constrained by the physical limits of infrastructure.
The Dream of Abundant Intelligence
A couple of days ago, Sam Altman published on his blog a post titled “Abundant Intelligence,” where he briefly explained the vision of the world where 10 gigawatts of compute can help AI figure out a cure for cancer or give every student on Earth their own personal AI tutor.
Like him or not, but Sam Altman needs to be acknowledged for his ability to sell the dream of abundant intelligence to some of the most powerful people, companies and institutions in the world.
People like Altman, the Silicon Valley overlords, believe that AGI is the culmination of all of humanity’s work. It will be our greatest invention, opening unlimited possibilities and economic growth never before seen in history. They believe this future is just around the corner and merely requires more computing power to realise.
If you come from this perspective, then putting every resource into building AGI is worth it. The short-term issues, like increased emissions or the need to build new power plants to provide gigawatts of power for these new AI datacenters, mean nothing, as the payout will be infinite. After all, AGI is going to solve all our problems, from climate change to fusion power, and make everyone live longer, happier lives.
If, however, we take a cynical perspective, then those recent megadeals are just to prolong the AI hype a little bit longer, to buy time to figure out a profitable business model before the hype fizzles out.
OpenAI—alongside Nvidia, Oracle, Google, Microsoft, Anthropic and others—is making AI an industry too big to fail. AI startups comprise about a third of all startups funded, both globally and in the US. They received 53% of all global venture capital dollars invested in the first half of 2025. And, as Deutsche Bank reports, AI is the only thing keeping the US from recession.
Additionally, AI has become one of the focus points in geopolitics and a key area of competition in the New Cold War between the US and China. Initiatives like the Stargate Project become not just a strategic business project but also a matter of national security.
The vast resources being poured into AI will, sooner or later, have to justify the investment. If everything unfolds as OpenAI plans, we may enter a new golden age for humanity—and some people will become extremely rich along the way. Otherwise, the outcome is unlikely to be pleasant for anyone involved.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!
My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"