The New American Dream Is Owning Your Own GPU
Summary
- The middle is underserved. AI infrastructure financing runs from investment-grade bank syndications down to compute leasing under $10 million, and the Compute Middle Market between them has few dedicated sources of capital. The Compute Middle Market tracker on gpuloans.com counts 41 publicly announced GPU-backed financings of $500 million or less since January 2023, and USD.AI accounts for 20 of them.
- Ownership is spreading. In Conor’s view, companies will own their GPUs to control their data, deal sizes will shrink, and AI capex will become programmatic within roughly 24 months.
- Liquidity and structure make it financeable. Onchain distribution gives capital a way to exit a new asset class, and residual value coverage where required addresses depreciation risk.
A GPU Now Costs as Much as a House
A GPU and a house are “literally the same price,” according to Conor Moore, co-founder and COO of USD.AI Inc. That comparison sits at the center of how he sees the next phase of AI infrastructure.
“The American dream going forward is going to be: do you own your own GPU or not? Not do you own your own house.”
Conor sat down with Bullish at its offices to talk about where AI capex is heading and who will finance it. The conversation followed Bullish’s $100 million stablecoin debt facility to USD.AI, which funds financings secured by GPU infrastructure. Bullish described the facility as its entry into middle-market AI infrastructure financing.
This post covers where that market sits today, why Conor expects it to grow, how onchain capital fits in, and what could prove the thesis wrong.
The Compute Middle Market Has Few Dedicated Lenders
AI infrastructure financing runs in three tiers, and the widest gap sits between sub-$10 million leasing and multi-billion-dollar facilities.
- Investment grade at the top. Banks syndicate the largest facilities, such as the roughly $38 billion of debt led by JPMorgan and MUFG in September 2025 for data centers tied to Oracle.
- Near investment grade below that. Operators like CoreWeave and Nebius raise multi-billion-dollar financings that sit just below investment grade, because their compute is rented to investment-grade customers.
- Leasing at the bottom. Compute leasing covers transactions under $10 million, mostly for startups.
Between leasing and the multi-billion-dollar players sits the Compute Middle Market: GPU cloud operators with contracted demand that are too small for bank syndicates and too large for equipment leasing. “Nobody services that market at all,” Conor said. USD.AI’s GPU financing is built for this segment, including neoclouds that own and operate their own clusters.
The Compute Middle Market tracker on gpuloans.com counts 41 publicly announced GPU-backed financings of $500 million or less since January 2023, totaling $4.71 billion across 20 named lenders as of November 2, 2026. The median deal is $65 million, and 32 of the 41 were announced in 2026. USD.AI accounts for 20 of the 41, and no other lender in the tracker has more than four.
GPU Ownership Is Moving Down the Stack
In Conor’s view, companies will own their compute within roughly 24 months, because control over data increasingly means control over hardware.
In that world, the neocloud as a standalone business model fades. “There will be no business where my job is to buy a box, stick it in a bigger box, and connect it to OpenRouter,” Conor said. GPU ownership spreads to every kind of company, the same way every company became a tech company.
You can already see the shift. According to Conor, six months ago RunPod, Shadeform, Modal and Vast.ai operated as compute marketplaces. Today they are signing bare metal leases and owning GPUs themselves. Inference providers like Baseten, Fireworks and Together, which turn bare metal compute into the tokens consumed in every ChatGPT query, are now buying GPUs too.
“Everyone is just working their way down further and further, because everyone’s realizing that that’s ultimately what they care about. They want to own the compute.”
The shift reaches beyond AI companies. Jane Street committed about $6 billion to CoreWeave’s AI cloud in April 2026. Conor sees a firm like Bullish running GPUs in its own office for internal agents as a realistic outcome within 24 months.
Smaller Deals Need Programmatic Financing
As GPU ownership spreads, deal sizes shrink, and bespoke private credit becomes a poor fit for a market of many smaller financings. Funds work deal by deal, so each deal has to deploy a lot of capital to be worth the time. “No private credit fund can spend their time doing tiny little deals at scale,” Conor said.
Our view is that AI capex will become programmatic, with every deal scored and structured the same way. Every financing through USD.AI is scored by Compute IQ, USD.AI’s agentic underwriting system, which scores each GPU deal on a 100-point scale across the operator, the customer and the contract. The score sets financeability, the pricing tier, and the loan’s size and structure.
USD.AI is built as a technology business, which shapes how it grows with customers. A borrower financed today stays for the term of the loan, which creates room to add payments in USDai and other financial products built on top of the financing. The goal, in Conor’s words, is to be “the primary financial partner for all these guys.”
Why Onchain: Liquidity for a New Asset Class
There is no ABS market for chips, which leaves capital in GPU-backed credit with few ways to exit. Conor contrasted this with CoreWeave’s bonds, which he said trade roughly $100 million a year on billions of dollars of notional value.
USD.AI’s digital assets work differently. On the capital side, these financings are funded through Earn: you stake USDai and receive sUSDai, whose value tracks how the loans perform. USD.AI’s tokens have traded $22 billion in secondary volume as of November 2026.
That liquidity changes how capital behaves. “They’re not pricing the credit more, they’re pricing how quickly can I get out of this position,” Conor said. Capital that can exit is more willing to fund an asset class the rest of the market is still learning, and Bullish is adding to that liquidity by listing sUSDai across multiple trading pairs with a dedicated market-making program.
Onchain distribution also lets a capital base form quickly. sUSDai deposits reached about $550 million in October 2026, about 14 months after launch. As deals get smaller and more numerous, that capital base is what funds them.
What Could Go Wrong
Conor sees two ways the thesis could miss, and he rates competition from banks as the more realistic one.
The first is that AI turns out to be a bubble and the hardware goes to zero, which he calls a very low-likelihood outcome. USD.AI has also structured for it, with residual value coverage on its financings where required, through its partnership with Barkr, to protect against GPU depreciation. Conor doesn’t expect GPUs to lose that much value over a three-year term, and the coverage is designed to reduce loss severity if the hardware sells below its insured value.
The second is that big banks move in and use their lower cost of capital to win every deal. Conor’s answer is that the advantage sits in technology and borrower relationships, and he expects banks that commit to the sector to participate in financings originated through USD.AI, onchain or not.
Tech-led platforms in consumer lending show the economics. Rocket reported a 2.54% total gain on sale margin for the third quarter of 2026, and Figure reported a 3.6% net take rate for the second quarter of 2026.
“Even though it’s a different outcome, a different possible story, it could still be a success story.”
Financing the New American Dream
If the next 24 months go the way Conor expects, compute ownership becomes as common as home ownership. Every company that wants control of its data will want its own GPUs, and most of them will need financing to get there.
That market will run on many smaller deals, scored programmatically and funded by capital that can trade in and out. Financing through USD.AI is built for it, from the Compute IQ score to the capital that funds each loan. If you are planning a GPU deployment in the Compute Middle Market, the next step is a conversation.
FAQ
What is the Compute Middle Market?
The Compute Middle Market is the GPU cloud operators between startups and hyperscalers. Their deals are too large for compute leasing, which runs under $10 million, and too small for the bank syndicates behind multi-billion-dollar facilities.
How many GPU-backed financings are in the Compute Middle Market?
As of November 2, 2026, the Compute Middle Market tracker on gpuloans.com counts 41 publicly announced GPU-backed financings of $500 million or less since January 2023, totaling $4.71 billion. The median deal is $65 million, and USD.AI accounts for 20 of the 41.
Why does Conor expect GPU deals to get smaller?
Companies increasingly want control over their data, and owning the hardware is the most direct way to get it. As ownership spreads from specialist operators to every kind of company, each deployment and each financing gets smaller.
How does USD.AI address GPU depreciation risk?
Where required, financings through USD.AI carry residual value coverage through USD.AI’s partnership with Barkr. The coverage is designed to reduce loss severity if the GPUs sell below their insured value.
