Nvidia announced a massive data center partnership with OpenAI, building a 1.5 million GPU AI factory in Portsmouth, Ohio, expected to generate $150-200 billion in revenue. Nvidia will provide partial guarantees to help OpenAI secure financing, locking in a 20-year chip order pipeline.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Nvidia has announced partnerships with investment firms like KKR and Apollo to provide partial guarantees for a 1.5 million GPU data center for OpenAI in Ohio, locking in chip revenue for up to 20 years. This model extends Nvidia's 'lock-in methodology' from chips to land, power, and shell (LPS), and the SEC's recent loosening of risk-retention rules last month paved the way for this $500 billion deal.
Nvidia CEO Jensen Huang announced a move seen by outsiders as 'betraying ancestral teachings': Nvidia will personally help customers secure the land, power supply, and building shell needed to build 'AI factories,' packaging these into a new concept called 'LPS' (Land, Power, Shell). Nvidia has signed agreements with large investment firms like KKR and Apollo, which will fund $500 billion to build data centers, with Nvidia providing partial guarantees: if tenants (like OpenAI) cannot pay rent, Nvidia commits to covering a portion of those costs.
This model targets frontier AI labs like OpenAI: they have enormous and rapidly growing computing needs, but short histories, few assets, and insufficient bank credit ratings to independently build large-scale data centers. In contrast, cloud computing giants like Amazon, Microsoft, and Google do not need Nvidia's involvement; they can handle infrastructure themselves, and Nvidia simply sells them chips, a model that will remain the core of its business.
Huang announced the first concrete project: located at the PORTS-Pike technology park in Portsmouth, Ohio, with OpenAI as the sole tenant, and a power supply of 4.25 gigawatts, equivalent to the electricity consumption of three to four million households. OpenAI will build and operate an AI factory there, using Nvidia's DSX full-stack AI factory platform, encompassing GPUs, CPUs, networking, and infrastructure software. The initial deployment is expected to provide 4.25 gigawatts of AI factory capacity.
Huang calculated: each generation of Nvidia AI factory systems deployed at PORTS-Pike corresponds to about 1.5 million GPUs and $150 billion to $200 billion in Nvidia revenue. The site's design life is 20 years, during which chips will be replaced several times; each replacement means Nvidia sells a new batch of chips, and the next generation will only be more expensive than $200 billion. If all of OpenAI's committed projects are combined, they correspond to approximately $600 billion in Nvidia chip purchases by 2030.
Huang clarified several points about the guarantee: First, the guarantee is limited; Nvidia is not covering all of OpenAI's bills, only a portion of rent and electricity costs, plus a specific residual value commitment—if the equipment is no longer used in the future, Nvidia acknowledges it retains a certain value. The guarantee will take effect in phases as the data center becomes operational between 2028 and 2030, and as OpenAI begins paying rent and capacity comes online, Nvidia's remaining risk exposure will decrease accordingly.
Second, addressing concerns about circular financing—Nvidia guarantees → OpenAI gets money → OpenAI buys Nvidia chips—Huang responded that rent is paid by OpenAI itself, not by Nvidia. Third, in case OpenAI leases only half and stops using it, Huang stated that Nvidia's chips are industry-standard and can be used by another tenant; qualified lessees include cloud service providers, enterprises, AI labs, and startups.
Huang traces the starting point of this logic chain to CUDA: CUDA provides a common platform for developers and Nvidia engineers to continuously improve installed systems, thereby making computing power universal. Universality brings interchangeability, interchangeability boosts utilization and durability, ultimately making Nvidia's computing power a rentable, financeable productive asset.
A precondition for this $500 billion deal was settled last month. After the 2008 financial crisis, the U.S. passed the Dodd-Frank Act, which requires that when loans are packaged into financial products and sold, the originator must retain a portion of the risk—the 'risk-retention rule.' Data center financing also aims to package future rental income into financial products to sell to investors, but if the risk-retention rule applies, the originator would have to tie up a large amount of capital, reducing financing efficiency.
Last month, top law firm Latham & Watkins filed an application with the U.S. Securities and Exchange Commission (SEC), arguing that data center rental income is not the same as mortgage loans and should not be subject to the same rule. The SEC essentially agreed. This means that when financing data centers, originators are not forced to 'retain a piece of the risk,' allowing for more flexible, lower-barrier, and more efficient financing structures. However, this is only an opinion from the regulator and has not yet become formal regulation, but it effectively opens a door: more funds will be willing and able to flow into data center construction.
Nvidia has historically locked in semiconductor resources through its scale, predictability of long-term demand, and supply chain partnerships. Now it aims to extend the same methodology to the LPS domain. Huang stated that when Nvidia can foresee customer demand and doing so ensures long-term production capacity, it will secure the supply of key inputs.
As for how much LPS Nvidia plans to lock in, Huang's answer was vague: 'strategic and disciplined.' That is, Nvidia does not intend to do this everywhere; most data centers will still be built by customers themselves, and Nvidia will only provide guarantees on a very few projects with 'good locations and clear demand.'
The information in this article primarily comes from reports by QbitAI, which are second-hand accounts. Huang's statements and specific figures (such as 1.5 million GPUs, $200 billion, 4.25 gigawatts) are based on his public statements or media reports and have not been independently verified against official Nvidia documents. The SEC's regulatory stance is based on a law firm application and reports, and has not yet become formal regulation.
Nvidia's involvement in data center construction through limited guarantees is essentially using its credit backing to lock in long-term chip orders, and regulatory easing has provided the conditions for this model. However, risk exposure, circular financing concerns, and regulatory uncertainty still warrant attention.
Primary report
Primary source