From MinerFi to ComputeFi: CoreWeave’s $8.5B AI Loan Rewrites the Playbook
This article first appeared in Miner Weekly, a weekly newsletter by BlocksBridge Consulting curating the latest news in energy, compute, infrastructure, and data analysis from TheEnergyMag. Subscribe to receive in your inbox once a week.
Featured EIF speaker: Congressman Earl “Buddy” Carter (R-GA), a member of the House Energy and Commerce Committee and the House Budget Committee, where energy reliability, infrastructure investment, and technology policy collide.
EIF is a high-signal, curated conference convening leaders across energy, compute infrastructure, and capital to map the next decade of power and data center buildout.
Register to attend today or explore sponsorships.
If 2021 was the year of ASIC-backed loans, 2026 may go down as the moment AI compute financing grew up.
CoreWeave’s $8.5 billion delayed draw term loan—secured primarily by GPU infrastructure—has already made headlines as the first and largest deal of its kind. The headline numbers alone are notable: the facility allows CoreWeave to draw capital through June 2027, with final maturity in March 2032. It is split between floating-rate debt priced at roughly SOFR + 2.25% and fixed-rate tranches around the high-5% range, putting the overall cost of capital in the mid-single digits—remarkably low for an asset class that didn’t meaningfully exist a few years ago.
But buried inside the credit agreement is something far more intricate: a blueprint for how Wall Street now wants to finance compute. This isn’t just “GPU-backed debt.” It’s compute treated as infrastructure.
And in many ways, it’s what MinerFi tried—and failed—to become.
The lesson from 2021: hardware alone isn’t enough
Back in 2021, bitcoin miners tapped a wave of ASIC-backed loans to expand aggressively. The premise was simple: machines generate bitcoin, bitcoin repays debt.
What that model underestimated was how volatile and fragile both sides of the equation were.
When bitcoin’s price fell in 2022, the hashrate went up relentlessly as a result of the MinerFi boom, deteriorating the hashprice even further. At the same time, the resale value of ASIC machines declined. So lenders were left with collateral that was worth less and generating less cash.
That double exposure—asset value risk and cash flow risk—was fatal.
The real collateral isn’t GPUs
CoreWeave’s credit agreement emphasizes the Data Center (DC) Funding Conditions, which are really the spine of the whole deal.
Simply put, CoreWeave cannot simply point to a purchase order for GPUs and borrow against it. The financing is tied to whether those machines are actually becoming usable compute inside a functioning data center, for a customer contract with the customer’s acceptance and satisfaction.
That matters even more because the advance rate is high.
The agreement defines the borrowing base for a funding date as the “Funding Date GPU Amount,” which equals 90% of Funding Date Capital Expenditures, plus certain fees and expenses tied to the transaction. Those capital expenditures include the cost of the infrastructure itself and installation costs.
In practical terms, lenders are saying they are willing to fund almost the entire build cost of the GPU deployment. But they only do that because the deal does not treat a GPU as collateral the moment it is bought.
Instead, the agreement requires the financed infrastructure to satisfy the DC Funding Conditions. Those conditions tie funding to real-world deployment: the equipment has to be physically associated with the relevant data center site, prior loans must already have been used properly, and for the applicable infrastructure, the DC Funding Conditions must be satisfied at the time of the borrowing. The contract also includes language elsewhere that accepted GPU clusters count toward Funding Date Capital Expenditures only when the customer has accepted them in writing and the borrower certifies testing has been completed for the relevant data halls.
So the key idea is this:
CoreWeave is getting very high leverage — roughly 90 cents of debt for every dollar of qualifying deployment cost — but only after proving that the machines are not just purchased, but installed, tested, accepted, and on the path to generating contracted revenue.
That is a notable difference from the MinerFi era.
In 2021, ASIC-backed lenders often financed miners largely on the assumption that once the machines were delivered, the economics would take care of themselves. Here, the lenders are willing to go bigger precisely because they are financing not a box of chips, but a controlled, trackable, revenue-linked compute system. That is what makes this feel less like equipment lending and more like infrastructure finance.
Following the machines down to the rack level
Once you start from that premise—only funding “ready-to-earn” assets—the rest of the structure builds around visibility and control.
Every GPU financed under the facility is tracked:
- It must be located at an approved site
- It must be tied to a specific deployment
- Lenders must have contractual access to the facility
This is very different from early ASIC lending, where collateral could be loosely defined and, in some cases, difficult to locate or control in distress scenarios.
Here, lenders know not just what they are financing, but where it is and how it is being used.
Cash doesn’t flow freely—it follows a strict order
Once those GPU clusters start generating revenue, the cash doesn’t simply go to CoreWeave.
It flows through a controlled system—known as a cash waterfall.
Here’s the idea: Every dollar that comes in is a pre-assigned job.
- First, pay the costs to keep the system running (power, operations)
- Then, pay lenders (interest and principal)
- Only after that can any remaining cash go back to the company
Why it matters: It ensures that debt repayment is prioritized automatically, rather than relying on management discretion.
In MinerFi, that level of control was often opaque.
And the agreement goes one step further with what’s called a cash trap.
If the project starts underperforming—for example, if it’s not generating enough cushion above its debt obligations—then excess cash stops flowing back to the company.
Instead, it stays within the system to prevent situations where value leaks out of the structure just as risks are rising.
Even the electric bill is part of the credit story
One of the more telling details is how seriously the agreement treats power costs.
Running GPUs at scale is energy-intensive, just like running bitcoin miners, and electricity is one of the largest—and most volatile—expenses.
The structure requires mechanisms to manage that risk, including hedging or other arrangements to stabilize power pricing.
In simple terms: the loan isn’t just underwriting compute—it’s underwriting the cost of the electricity that makes that compute possible. And that variable is actively controlled rather than left exposed.
Borrowing only happens when the math works
All of this feeds into a final layer of discipline: the project must prove it can pay for itself.
That’s where the debt service coverage ratio (DSCR) comes in. DSCR compares how much cash a project brings in versus how much it needs to pay lenders (principal+interest).
- If a project generates $120 and owes $100 in debt payments, its DSCR is 1.2x
- That extra $20 is the cushion—it allows for things to go slightly wrong without immediately causing stress
In CoreWeave’s case:
- To take on new debt, the projected DSCR needs to be around 1.20x
- To stay in compliance, it needs to maintain roughly 1.15x
That buffer is what gives lenders confidence that even if things don’t go perfectly, the system can still hold.
The bigger shift: from machines to systems
Put all of this together, and the contrast with 2021 becomes clear.
MinerFi was built around machines: Finance the hardware and expect the revenue to follow.
This deal is built around systems:
- Verify the infrastructure is live
- Confirm the cash flow is real with customer contracts
- Control how that cash is used
It’s a fundamentally different way of risk management.
At $8.5 billion, backed by major banks and structured to investment-grade standards, this isn’t just a large deal. It’s a signal.
Compute—specifically AI compute—is being redefined as infrastructure that can be financed with the same discipline as power plants or toll roads.
Bitcoin mining showed that energy can be turned into digital output at scale.
CoreWeave’s financing shows how that output can be turned into something lenders are willing to underwrite at massive scale.
Of course, structure can’t eliminate risk. Both MinerFi and this new wave of compute financing ultimately depend on demand holding up.
The difference is where the risk sits.
- In mining, it was concentrated in network and market-driven revenue (hashprice)
- Here, it’s embedded in contract durability and utilization
Whether that shift makes the system more resilient—or simply changes how stress emerges—remains to be seen.
But one thing is already clear: Compute financing has entered a new phase—and this time, it looks a lot more like Wall Street than crypto.
Hardware and Infrastructure News
- Soluna Buys Texas 150MW Wind Farm for $53M to Power AI, Bitcoin Data Centers
- Bitdeer Unveils 9.45 J/TH Miner as Bitcoin Hashprice Remain Under Pressure
Corporate News
- DeepSeek’s V4 model will run on Huawei chips, The Information reports
- CleanSpark Shutters 25MW Tennessee Bitcoin Mine Tied to GRIID Noise Dispute
