A new bottleneck is emerging in artificial intelligence. It is not GPUs or high-bandwidth memory, but the capital required to build the data centers that will house them — projects increasingly measured in tens or even hundreds of billions of dollars. Nvidia is working with some of Wall Street’s biggest financial institutions, including Apollo Global Management, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs and KKR, on a financing platform that could channel more than $500 billion into AI infrastructure. The institutions are exploring separate pools of capital that could help Nvidia customers finance chips, data centers and the power infrastructure needed to run them.
No final financing agreement has been completed, and Nvidia and the financial institutions are still operating under memorandums of understanding. But the scale is significant: more than half a trillion dollars could ultimately be mobilized around the physical infrastructure of AI. More important is how Nvidia’s role is changing. Its core business has been designing and selling AI processors; it is now moving further downstream, helping connect customers with the capital needed to build the facilities where those chips will operate.
To Sell More AI Chips, Someone Has to Finance the Data Centers
Nvidia’s financial performance hardly suggests a company struggling to sell chips. Revenue for fiscal 2026 reached $215.9 billion, up 65% from the previous year, while fourth-quarter data center revenue alone climbed 75% to $62.3 billion. Demand for AI computing remains strong, but the problem is the sheer amount of infrastructure required to sustain that growth.
Morgan Stanley estimates that hyperscale cloud companies could spend about $3.5 trillion on AI infrastructure between 2026 and 2028, while longer-term estimates put the total infrastructure buildout at more than $8 trillion. Amazon plans roughly $200 billion in capital expenditures this year, Alphabet expects $175 billion to $185 billion, and Meta projects $115 billion to $135 billion. Microsoft is also maintaining annual investment well above $100 billion. AI is increasingly taking on the economics of capital-intensive industries such as power, telecommunications and heavy infrastructure rather than traditional software.
That creates a structural constraint. Nvidia can manufacture more GPUs, but if customers cannot secure the capital and electricity required to build data centers, there will be nowhere to deploy them. Nvidia’s growth therefore increasingly depends on the financing capacity of its customers.
The massive data center financing under discussion with OpenAI illustrates the shift. Nvidia has been discussing a potential guarantee of roughly $250 billion connected to a large OpenAI data center project in Ohio, while the total project could cost more than $500 billion. Nvidia would not necessarily pay the construction bill itself; the emerging model instead uses its financial strength alongside Wall Street capital to help customers raise enormous amounts of debt. If that structure takes hold, Nvidia will occupy an unusual position: it will not only supply the chips but also help finance the market that buys them.
The New AI Risk: Circular Financing
Chipmakers, data center operators, AI model companies and financial institutions are becoming increasingly intertwined. Nvidia supplies GPUs, data center operators use assets and long-term contracts to raise capital, and AI companies sign multiyear computing agreements. Banks and private-credit firms then lend against those assets and expected cash flows, creating an increasingly interconnected financing structure around AI infrastructure.
CoreWeave is one of the clearest examples. The AI cloud company has expanded rapidly by acquiring large quantities of Nvidia GPUs and supplying computing capacity to customers including Microsoft, Meta and OpenAI. Its projected revenue this year is around $12.6 billion, while its contract backlog stands at roughly $99 billion. CoreWeave has also raised more than $20 billion this year, including $8.5 billion in loans, while Nvidia is both a major GPU supplier and a significant shareholder.
There is nothing inherently problematic about this model. Capital-intensive industries such as power generation, aviation and telecommunications routinely finance large projects against physical assets and long-term customer contracts. AI presents a different risk, however: the underlying technology depreciates much faster.
A power plant or transmission network can remain productive for decades, while GPUs are replaced by more powerful generations at a far faster pace. Hardware purchased for billions of dollars today may not retain the same economic value several years from now. As GPU-backed borrowing expands, the mismatch between chip depreciation and debt maturities could become a new source of financial risk.
Demand is the bigger test. If AI services expand as rapidly as markets expect, the capital invested in GPUs and data centers could generate sufficient cash flow. If monetization takes longer, operators will have to absorb high electricity costs, depreciation and interest expenses at the same time. The central question is whether trillions of dollars of AI infrastructure can generate returns on the same scale; if investment begins to outrun monetization, the infrastructure boom could shift from a growth engine to a balance-sheet burden.
Power May Be Scarcer Than the $500 Billion
Even abundant financing cannot solve the next constraint: electricity. The International Energy Agency expects global data center electricity consumption to rise from about 485 terawatt-hours in 2025 to roughly 950 terawatt-hours by 2030, nearly doubling in five years. Electricity consumption from AI-focused data centers is expected to roughly triple over the same period.
Power generation required to serve data centers is also projected to rise from around 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours by 2030. The pace of AI data center construction will therefore depend not only on the ability to purchase GPUs but also on power plants, transmission networks, transformers, cooling systems and water. That is why Nvidia’s proposed $500 billion financing platform extends beyond chip purchases, with capital increasingly being structured around data centers and energy infrastructure as well as computing hardware.
Nvidia Chief Executive Jensen Huang has repeatedly described data centers as “AI factories.” His argument is that traditional data centers stored and processed information, while AI infrastructure produces tokens and inference output as an industrial product. Viewed through the lens of finance, the analogy becomes even more relevant: if an AI factory is truly a factory, semiconductors alone are not enough. It also needs land, electricity, construction capital, long-term financing and customers.
Nvidia’s Next Competitor May Not Be Another Chipmaker
The shift could ultimately change the nature of Nvidia’s competitive advantage. Measured purely by processor performance, Nvidia competes with semiconductor companies such as AMD and custom accelerators developed by Google and other hyperscalers. But the competitive landscape looks different if Nvidia can offer an ecosystem connecting chips, networking, software, data centers and finance.
For customers, buying the fastest GPU may eventually be easier than raising billions of dollars and securing enough electricity to bring a data center online. Connecting Wall Street’s financial capacity to Nvidia’s technology ecosystem could therefore become another competitive advantage. Nvidia would no longer compete solely through semiconductor performance but through its ability to mobilize the infrastructure surrounding those chips.
There is also a direct implication for South Korea. As more Nvidia GPUs are installed in hyperscale data centers, demand for high-bandwidth memory from SK hynix and Samsung Electronics should continue to grow. But if Korean companies remain primarily memory suppliers in the AI value chain, much of the additional value generated by data centers, electricity infrastructure and financing could accrue to U.S. hyperscalers, asset managers and private-capital firms.
South Korea’s expansion of semiconductor capacity in Yongin, Pyeongtaek and Cheongju should be viewed in the same context. Capital to build fabs is necessary, but it is no longer sufficient. Power, water, advanced packaging, data centers and long-term financing increasingly need to operate as parts of the same industrial system.
That is why the most important number in Nvidia’s Wall Street initiative may not ultimately be $500 billion. The AI industry can no longer run on better chips alone: it needs capital to buy those chips, data centers to house them, electricity to run them and financial institutions willing to carry the risk over many years. Nvidia’s move to bring Wall Street deeper into its ecosystem reflects that reality. The next phase of AI competition will depend not only on who builds the best GPU, but on who can integrate technology, capital, power and data centers into a functioning production system. reality. The next phase of AI competition will depend not only on who builds the best GPU, but on who can integrate technology, capital, power and data centers into a functioning production system.
