How Is the AI Infrastructure Buildout Being Financed?
SubscribeSign in
How Is the AI Infrastructure Buildout Being Financed?<br>The emerging world of compute foundries, shadow borrowing, and infrastructure-scale AI finance
Under the Claim<br>May 21, 2026
Share
Elon Musk@elonmusk
Colossus II, the world’s first Gigawatt AI training cluster
X Freeze @XFreeze
xAI’s Colossus 2: The World’s First Gigawatt-Scale Datacenter
2:56 PM · Sep 17, 2025 · 13.2M Views
5.05K Replies · 7.52K Reposts · 82.6K Likes
Artificial intelligence is rapidly becoming one of the largest infrastructure buildouts in modern technological history. Hyperscalers and AI companies are spending tens of billions of dollars building data centers, acquiring GPUs, securing electricity, and expanding cloud capacity fast enough to keep up with exploding AI demand, demand which i dont know how they measure. Amazon alone expects AI-related investments to continue pressuring its free cash flow which came down to near 1 billion as of recent, while companies like Microsoft, Google, Oracle, and Meta are collectively pushing annual capital expenditures toward levels once associated more with national infrastructure projects than software companies.Oracle alone issued $43.0 Billion in senior notes to fund its technical property expansions, even though they dont even generate that much operating cash flow on their own by the way. At the same time, the physical scale of these projects has become enormous: OpenAI and Anthropic are now discussing multi-gigawatt data center campuses, and the largest AI facilities under development may soon consume as much electricity as small cities.AI having its own electric city. Yet despite this unprecedented spending boom, many of these companies continue to maintain surprisingly positive balance sheets. Debt levels have certainly risen, but not currently in their papers.
This enormous expansion of AI infrastructure is being financed through a rapidly evolving web of debt markets, private credit, leasing arrangements, joint ventures,all of which we can revolves around credit, and other forms of off-balance-sheet financing. Building frontier AI systems no longer just means hiring more expensive ML engineers or Researchers like Meta did or buying servers - it increasingly resembles financing large-scale industrial infrastructure, an Industrial Revolution. Training clusters require billions of dollars in GPUs, data centers consume vast amounts of electricity, and hyperscalers are now entering into long-term commitments for land, power, networking equipment, and compute capacity. Yet much of the financial burden created by these arrangements does not always appear on their paper directly on their corporate balance sheets in the form of traditional debt. Instead, many companies are increasingly relying on financial setups that allow them to continue expanding aggressively while the red flags on their paper relatively manageable, a dynamic that the Bank for International Settlements recently described as a form of “shadow borrowing”.
The Hidden Balance Sheet
At its simplest, off-balance-sheet financing refers to financial obligations that function economically like debt, but do not always appear directly as traditional liabilities on a company’s balance sheet. Companies have long used various forms of off-balance-sheet arrangements - leases, joint ventures, supplier financing agreements, special purpose vehicles, and long-term contractual commitments - to fund expensive projects without taking on large amounts of conventional debt. In many industries these arrangements are relatively mundane. Airlines lease aircraft instead of buying them outright, energy companies finance pipelines through joint ventures, and real estate developers routinely separate assets into standalone entities. But the enormous capital requirements of the AI infrastructure revolution are increasingly pushing hyperscalers and AI companies deeper into these financing setups, blurring the line between technology companies and large industrial borrowers.
What actually appears on a balance sheet can also be somewhat misleading. A company that borrows money directly by issuing bonds will record that borrowing clearly as debt. But many long-term financial obligations are structured differently. A hyperscaler might sign a multi-decade lease for a data center instead of financing and owning it directly, commit to enormous long-term power purchase agreements to secure electricity for future AI clusters, or enter into joint ventures where infrastructure assets and liabilities sit partly outside the parent company itself. Economically, these arrangements can create obligations very similar to debt, a new kind of debt in tech companies? : future cash flows are effectively committed long before these future cash flow even exist or can be realized from these AI Cities. But because these commitments are distributed across leases, partnerships, suppliers, and...