Industries / Finance and business services / Finance and banking / Equipment leasing and factoring
Pricing and residual values
This is the work of setting what a lease costs, which rests on a guess about the residual value: what the equipment will be worth when it comes back at the end of the term.
5
Milestones achieved, from 1979 to 2023.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1979
A computer bet collapses
Itel, a San Francisco lessor founded in 1967 by Peter Redfield and Gary Friedman, leased IBM and IBM-compatible computers and had Lloyd's of London underwriters insure it, from 1975 to 1978, against customers handing computers back before their leases ended, in effect insuring against the machines going out of date. Talk of IBM's new 4300 series in late 1978 and early 1979 brought a wave of cancellations. Itel lost $443.3 million in 1979, by the Christian Science Monitor's figure ($433.3 million in Computerwoche's), cut its reported 1978 profit from $47.2 million to $21.5 million because of lower residual values on its computers, and sought bankruptcy protection under Chapter 11 in 1981.
LessonOne new model from the maker can wipe out the residual values of a lessor's technology equipment, and insuring that bet only moves the loss to the insurer.
Credit
Itel; Lloyd's of London underwriters
1981
Tax breaks sold by lease
The US Economic Recovery Tax Act of August 1981 let companies with too little taxable income sell their investment tax credits and depreciation to profitable ones through "safe harbor" leases, in which the profitable firm owned the equipment only on paper and passed the tax saving back through lower rent. The Federal Reserve Bank of Cleveland wrote that such firms had faced a cost of capital 10% to 30% higher than profitable ones. Companies leased more than $37 billion of property this way in 1981 and 1982, by the count of the Internal Revenue Service's Statistics of Income division. The Tax Equity and Fiscal Responsibility Act (TEFRA) of 1982 restricted the leases, which the Equipment Leasing and Finance Association (ELFA) records as their repeal.
LessonA lease that exists only to move tax benefits lasts only as long as lawmakers accept its cost in lost revenue.
Credit
US Congress
1992
An aircraft lessor's flotation fails
On 18 June 1992 GPA (Guinness Peat Aviation), Tony Ryan's aircraft lessor at Shannon, Ireland, planned the largest share offering in the history of the Irish Stock Exchange and abandoned it the same day, after fund managers in the US and Japan stayed away. A former executive's diary, reviewed in the Irish Times, records that GPA had placed orders worth billions of dollars for new aircraft as airlines headed into recession, and was left with aircraft it could not place. GE Capital then took over the company.
LessonA lessor's order book is a bet on future demand for its aircraft, and the bet comes due when airlines fail.
Credit
GPA (Tony Ryan, chief executive Maurice Foley)
2019
Truck residuals cut
Ryder, a US truck-leasing company, sets the residual values in its lease prices on a long-term view of used-truck prices. In the third quarter of 2019, with the used market softening and more of its own trucks waiting to be sold, Ryder concluded that its residual estimates were above what its tractors and trucks would fetch and lowered them. The change added about $297 million to the year's depreciation in its fleet business, and total depreciation rose to $1.9 billion from $1.4 billion in 2018.
LessonResiduals set on a long-term view lag a falling used market, and the gap shows up at once as extra depreciation.
Credit
Ryder System
2023
Used truck prices fall
J.D. Power Valuation Services reported in January 2024 that the average used sleeper tractor sold at retail in 2023 brought 33.5% less than in 2022, and that late-model sleepers sold at auction brought 40.8% less, losing an average of 3.9% of their value each month. At retail, late-model sleepers were back to roughly 11% below late 2018, the last strong period before the pandemic, in dollars of the day, or about 28% below after inflation.
Credit
J.D. Power Valuation Services
Watch for
Six-year lives for AI chips
CoreWeave, which rents out Nvidia GPUs and borrows against them, has depreciated its data-center computing equipment over six years, up from five, since 1 January 2023, and repeated the policy in its annual report filed in March 2026. Amazon moved the other way: from 1 January 2025 it cut the life of a subset of its servers and networking equipment from six years to five, citing the "increased pace of technology development, particularly in the area of artificial intelligence". American Compute's GPU Resale Index, the resale price of a fixed basket of used data-center GPUs aged 2.5 to 8 years, stood at 75 in August 2026 (first half of 2025 = 100), 14% below normal depreciation by its count, while used servers built on Nvidia's H100 held their price.
Achieved whenpublished prices from real sales of GPUs five to six years old, in data an outside party can check, show them worth at least what a six-year straight-line schedule leaves on the books.
Claimed by
CoreWeave, March 2026
Open
01
Residual rules, replayed on trucks
A public replay of how a lease book of sleeper tractors would have fared from 2018 to 2026 under different ways of setting residual values: a fixed depreciation curve, a curve from the trailing five-year average of used prices, the current used price with a fixed cut, and a rule that resets every quarter. Each simulated lease is priced with its rule and settled against the used price actually recorded when its term ends.
The briefCredit · unclaimed02
GPU loans, stress-tested
An open model of loans secured on GPUs, of the kind data-center operators now raise, that shows how far used GPU prices could fall before the chips no longer cover the debt. It models three made-up structures shaped like published ones (debt repaid in step with depreciation, a customer contract shorter than the chips' book life, and a re-lease after the first contract ends) and runs each against price paths drawn from published resale indexes. It names no borrower and judges no real loan.
The briefCredit · unclaimed
Open 01 · the brief
Residual rules, replayed on trucks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public replay of how a lease book of sleeper tractors would have fared from 2018 to 2026 under different ways of setting residual values: a fixed depreciation curve, a curve from the trailing five-year average of used prices, the current used price with a fixed cut, and a rule that resets every quarter. Each simulated lease is priced with its rule and settled against the used price actually recorded when its term ends.
Why it is still open
Ryder's 2019 cut (see above) shows what a slow reset costs, and J.D. Power publishes monthly used-truck prices in free PDF reports, but the figures sit in one PDF per month. This research found no public comparison of residual-setting rules against them; a GitHub search found models that estimate a machine's value, such as a 2023 ensemble-learning project for heavy construction equipment, not tests of how lessors set residuals.
Built with what exists
J.D. Power's monthly Commercial Truck Guidelines PDFs, used within J.D. Power's terms, with every figure cited to its report and page; pdftotext or pdfplumber to pull the numbers; Python with pandas for the simulation; an AI model such as Claude Fable 5.2 (see the Note "Three working devices") to draft the extraction code, checked by hand.
First step
Download the 12 monthly reports for 2023, extract the average retail price of the benchmark sleeper, and check the yearly fall against J.D. Power's stated 33.5%.
Done looks like
A monthly price series from at least 60 reports, each value linked to its page; a second person rebuilding the series from 10 randomly chosen reports with no difference above $100; and a published table of each rule's gain or loss per truck at the end of 36-, 48- and 60-month leases.
Open 02 · the brief
GPU loans, stress-tested
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open model of loans secured on GPUs, of the kind data-center operators now raise, that shows how far used GPU prices could fall before the chips no longer cover the debt. It models three made-up structures shaped like published ones (debt repaid in step with depreciation, a customer contract shorter than the chips' book life, and a re-lease after the first contract ends) and runs each against price paths drawn from published resale indexes. It names no borrower and judges no real loan.
Why it is still open
Writing in May 2026, Quartz put it plainly: "On residual value, the data is thin." In September 2026 an analysis by the venture firm Wing reported that CoreWeave's newest GPU credit agreement on file never uses the words a lender uses to value hardware against an outside reference, such as appraisal, residual value or loan-to-value, and called these loans "aircraft leasing with a five-year asset" without the residual-value history aircraft have. Resale indexes now exist, such as American Compute's, but no open model ties them to loan structures.
Built with what exists
Python; American Compute's monthly GPU Resale Index and any other index whose method is published; depreciation policies from public filings, such as CoreWeave's six-year life; Nvidia's stated one-year rhythm for new chip generations as the main stress, since each new generation pushes down the price of the last.
First step
Build the first structure in a spreadsheet, a 1,024-GPU loan repaid over four years, and find the used price at which it breaks even.
Done looks like
A public notebook in which, for each of the three structures, the price fall at which the chips stop covering the debt is reported for 1,000 simulated price paths, with every input sourced, and two people outside the project rerunning it to the same results.