Lenders committed $48.4 billion of new debt secured on GPUs in public financings in 2026 through October 10, against $6.6 billion in all of 2025, according to American Compute’s GPU Financing Tracker. By our arithmetic, that is about 7.3 times last year’s total in a little over nine months.
The tracker puts GPU-backed debt at 47% of the $103 billion of AI infrastructure debt it counts for 2026, up from 8% of $79 billion in 2025. Social Capital drew attention to the figures in a post on X on November 7.
Business Pill · Credit Spreads
An evergreen Business Engineer explainer on why two borrowers pay different rates. It teaches the general idea only and does not discuss the figures in this piece.
The key insight: American Compute’s tracker counts $48.4 billion of new GPU-backed debt in 2026 through October 10, against $6.6 billion in all of 2025, and puts it at 47% of the AI infrastructure debt it tracks this year, up from 8%; by our arithmetic, that is about 7.3 times last year’s total.
American Compute says its tracker covers “$48.4B across 21 public financings to October 10, 2026, out of $103B of GPU and AI data center debt”. It sorts each financing by what secures it. GPU-backed debt is “Secured on GPUs or financed compute equipment: GPU delayed draw term loans, GPU-backed facilities, equipment leases, sale-leasebacks, vendor financing” (American Compute).
The other two groups are data center and campus debt, secured on a campus, its lease or the project, and corporate debt, which is “Not ring-fenced on GPUs or a campus”. In 2026 to October 10, the tracker counts $22.4 billion of data center and campus debt across 11 deals and $32.1 billion of corporate debt across 17.
New AI infrastructure debt by what secures it: 2025, $6.6B GPU-backed, $46.1B data center and campus, $26.4B corporate ($79B); 2026 to 10 Oct, $48.4B GPU-backed, $22.4B data center and campus, $32.1B corporate ($103B). GPU-backed share 8% vs 47%. Data: American Compute, GPU Financing Tracker, as published.
What Was Posted on X
Lenders have committed $48.4B to GPU-backed loans this year, up from $6.6B in all of 2025.
According to American Compute, these loans now make up ~47% of AI infrastructure debt, up from 8% in 2025.
The loans are secured by the GPUs and repaid from the rental income they generate.
So lenders are exposed to falling rental prices on two fronts: the value of the collateral and the cash flow that repays the loans once current contracts end.
A post by Social Capital at 20:03 UTC on 7 November citing American Compute: it says lenders committed $48.4 billion to GPU-backed loans this year, against $6.6 billion in all of 2025, now about 47% of AI infrastructure debt, up from 8%. The post adds Social Capital’s view that lenders are exposed to falling rental prices through both the value of the collateral and the cash flow that repays the loans once current contracts end. The figures above come from American Compute’s own tracker page.
How the Mix Changed
The tracker says last year’s money “went mostly to campuses”: $46.1 billion of data center debt in 2025, “led by $27.3B for Meta Hyperion JV (Blue Owl) and $7.1B for Crusoe Abilene JV (Blue Owl, Primary Digital)”. In 2026 to date, data center and campus debt is $22.4 billion, while GPU-backed debt is $48.4 billion.
By quarter, the tracker shows $23.4 billion of GPU-backed debt in the first quarter of 2026, $9.3 billion in the second, $14.4 billion in the third and $1.4 billion in the fourth quarter to October 10. By our arithmetic, the first quarter of 2026 alone is about 3.5 times all of 2025.
By instrument, the largest GPU-backed category in 2026 is delayed draw term loans (DDTLs), at $21.5 billion across nine deals, followed by GPU-backed facilities at $14 billion across six. The tracker also counts one $3.5 billion sale-leaseback, $2.5 billion of vendor financing and a $725 million equipment lease.
Our Number: 72% Since 2023
The tracker counts $197 billion of new money across 101 financings since 2023, of which $66.8 billion is GPU-backed. By our arithmetic, the $48.4 billion raised in 2026 to October 10 is about 72% of all the GPU-backed debt the tracker has recorded since 2023.
Why Lenders Watch Rental Prices
Social Capital’s post says the loans “are secured by the GPUs and repaid from the rental income they generate”, and that “lenders are exposed to falling rental prices on two fronts: the value of the collateral and the cash flow that repays the loans once current contracts end”. That is Social Capital’s description; American Compute’s page reports the amounts.
As we read the tracker, the change is in what lenders take as security: in 2025 most of the money it counts was secured on campuses, and in 2026 most of it is secured on the chips.
As we read the quarterly table, the jump came early: the tracker shows $23.4 billion of GPU-backed debt in the first quarter of 2026, which by our arithmetic is about 3.5 times all of 2025.
As we read Social Capital’s post, its point is about repayment: it says the loans are repaid from the rental income the GPUs generate.
American Compute, GPU Financing Tracker
“Private deals we see are not in it”
Three Implications
FOR LENDERS Social Capital’s post says lenders are exposed to falling rental prices through both the collateral value and the cash flow that repays the loans.
FOR BORROWERS The tracker’s largest GPU-backed category in 2026 is delayed draw term loans, at $21.5 billion across nine deals.
FOR READERS OF THE TOTALS The tracker counts public financings and new money only, to October 10, 2026, so later or private deals are not in its figures.
The analysis says: “The job of any operator, investor, or builder is to know which layer they are operating inside and which layer the binding constraint is rotating toward next.”
As we read it, the tracker shows where lenders now take their security in that stack: on the chips rather than the campus. The analysis does not discuss American Compute’s tracker.
What Is Not Established
The tracker counts only public financings with a dated press release, filing or report; American Compute says “Private deals we see are not in it”. It counts new money only, so upsizes add their increase and reported-only deals add nothing. Its figures run to October 10, 2026, so they do not include later announcements, such as the $40 billion of GPU debt Elon Musk wrote on October 25 that SpaceX had closed, which we reported. This piece does not say whether any of the loans are at risk.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
American Compute’s tracker counts $48.4 billion of new GPU-backed debt in 2026 through October 10, against $6.6 billion in all of 2025, and puts it at 47% of the AI infrastructure debt it tracks this year, up from 8%. By our arithmetic, that is about 7.3 times last year’s total and about 72% of the GPU-backed debt the tracker has recorded since 2023.
A note on sourcing. The debt amounts, deal counts, shares and categories are from American Compute’s GPU Financing Tracker summary page, updated October 10, 2026, as published; it counts public financings only. The view on lender exposure is Social Capital’s. The multiples and the 72% figure are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
