1. Purpose of This Tool
This is LEAPEF's internal analytical platform for GPU-backed structured finance.
It is used to build conviction on asset pricing, stress-test assumptions before investor calls,
and generate the analytical basis for term sheets and investor presentations.
This Methodology section is the portion shared externally when investors ask
"how did you arrive at this number?" —
it explains the approach, the evidence base, and the limitations,
without disclosing the proprietary parameters underlying the model.
2. The Question This Model Answers
"We are financing a purchase of X GPU units at today's price.
What is a credible estimate of secondary market resale value in 12, 24, and 36 months —
and what is the absolute floor we can rely on?"
The answer requires a model that accounts for technological depreciation,
supply chain dynamics, generation transition effects, and structural market constraints.
A straight-line depreciation assumption is not sufficient.
LEAPEF has built a multi-factor quantitative model calibrated from empirical market data
spanning a decade of GPU secondary market observations.
3. Model Architecture — Three Layers
LAYER 1 — DEPRECIATION CURVE
Exponential decay model with asymptotic floor. Decay rate fitted separately per GPU generation from observed secondary market transactions. Floor anchored to replacement cost economics — not an arbitrary assumption.
LAYER 2 — SUPPLY CHAIN FLOOR
Six observable supply chain factors determine the floor price. Each factor's weight reflects its share of GPU bill-of-materials and substitutability. Sources updated quarterly from published earnings data. Full derivation shown in the Rental tab.
LAYER 3 — MONTE CARLO
1,000 simulation paths produce a probability distribution of outcomes at each future date. P10–P90 confidence bands reflect genuine market uncertainty that grows over time. No false precision — the model shows a range, not a single number.
4. Empirical Calibration — The Evidence Base
The model is calibrated from a proprietary internal database compiled by LEAPEF's quantitative analysis team. The database draws on four evidence sources:
GPU secondary market price history (2016–2026) —
Multiple GPU generations across consumer and data centre segments, with price observations at standardised intervals from launch through floor formation. Sources include major secondary market platforms, enterprise resale markets, and LEAPEF's own transaction data.
Generation transition analysis —
For each GPU generation transition since 2018, LEAPEF tracks how the prior generation's price responded before, at, and after the new generation's launch. This produces empirically fitted transition coefficients incorporated into the model's time-varying decay rate.
Cross-asset validation —
GPU depreciation curves are validated against consumer electronics (flagship smartphone resale curves, 2020–2026) and automotive residual value data (premium and mainstream segments, 2021–2026). These comparable assets confirm that exponential decay with an asymptotic floor is the correct structural form, and that floor retention of 30–50% of launch price is consistent with observed behaviour across technology-intensive durable assets.
Supply chain quarterly data (2020–2026) —
Key input cost indicators tracked quarterly from public earnings sources: advanced packaging capacity, memory ASP, lithography equipment shipments, and specialty gas markets. Used to validate the supply chain factor weights in the floor derivation.
5. Key Empirical Findings
DATA CENTRE vs CONSUMER — STRUCTURALLY DIFFERENT
Data centre GPUs retain materially more value at equivalent age than consumer GPUs. The addressable workload pool is larger, stickier, and more price-inelastic than the consumer gaming market. B300 is modelled on data centre precedent — not consumer precedent. This distinction is material for investor analysis.
GENERATION TRANSITIONS — DECLINING IMPACT EACH CYCLE
Each successive GPU generation transition has caused progressively less depreciation pressure on the prior generation. The inference workload market has grown faster than supply with each cycle, absorbing displaced hardware into still-productive lower-tier use cases. This trend is incorporated into the model's time-varying decay rate.
FLOOR IS COST-ANCHORED — NOT ARBITRARY
The asymptotic floor is anchored to the economics of the lowest-cost workload the GPU can still serve — below which renting new capacity becomes more economical. Supply chain constraints keep replacement cost structurally elevated. Data centre GPU floors have been rising as a percentage of launch price with each generation — the opposite of what a simple obsolescence model would predict.
INFRASTRUCTURE CONSTRAINT — THE HONEST BEAR CASE
B300 requires liquid cooling and high-density power infrastructure not present in the majority of existing data centres. This reduces the addressable secondary buyer pool relative to H100. LEAPEF explicitly models this as a floor discount — it is the most significant bear case factor and is applied transparently in the floor derivation.
6. Model Validation — Back-Test Across 7 GPU Generations
The model was back-tested against all GPU generations for which sufficient secondary market data exists. Solid lines = real market observations. Dotted lines = model output using the same formula applied to B300. Close tracking across multiple generations and price scales confirms the model is well-calibrated before being applied to forward projections.
Solid lines = observed secondary market prices (real transactions).
Dotted lines = model prediction using the same methodology.
Data centre GPUs on left axis. Consumer GPUs on right axis.
7. Generation Transition Analysis
All GPU generations normalised to 100 at launch. Vertical markers show next-gen availability date. The declining impact of each transition on prior-gen price trajectory is the empirical basis for the time-varying decay rate.
EARLY ERA — STRONG ACCELERATION
First generation transitions caused the sharpest prior-gen price drops. The AI inference market was nascent and could not absorb displaced compute at scale.
MID ERA — MODERATING IMPACT
As inference workloads scaled, each new generation launch caused progressively less disruption. The secondary market deepened and became more liquid.
CURRENT ERA — MINIMAL IMPACT
The most recent transition saw the prior generation's long-term contract rate rise after the new generation launched. Demand growth now outpaces generational supply shifts.
8. Confidence Framework
LEAPEF applies an explicit confidence rating to each GPU's projection, shown in the resale table:
| Rating | Criteria | How to Use | Current GPUs |
|---|
| HIGH CONF. | 7+ observed data points. Strong R². Floor confirmed in real transactions. | Present with full confidence. Numbers are market-tested. | A100 · GTX 1080 · RTX 2080 · RTX 3090 |
| VALIDATED | 24+ months observed. Floor forming. Model tracking actual prices. | Use as primary reference. Floor directionally confirmed. | H100 |
| EARLY DATA | 6–18 months observed. Some data points. Model partially tested. | Use with caveat. Direction reliable, magnitude has uncertainty. | H200 · RTX 4090 |
| PROJECTED | Under 18 months. Extrapolated from validated precedent. No floor confirmed. | Present as model output — clearly label as projection, not observation. | B300 · B200 |
9. What This Model Does Not Do
In the interest of transparency, LEAPEF states explicitly what this model cannot claim:
It does not predict the exact secondary market price at a future date. It produces a probability distribution — P10 through P90 — that reflects genuine uncertainty.
It does not account for black swan events: geopolitical supply shocks, sudden regulatory changes, or unprecedented demand collapse. These are tail risks no model can price precisely.
It does not incorporate real-time market data automatically. Supply chain parameters are updated quarterly from published earnings. Secondary prices are refreshed via the data tool on the GPU Market Value tab.
It does not replace independent valuation. For formal credit analysis or investment decisions, the model output is one input among several — not a standalone valuation.
10. LEAPEF Deal Flow Index
LEAPEF AB operates as a licensed AIFM (Finansinspektionen) active in GPU-backed structured finance.
Our deal origination provides direct market observation of pricing not fully visible in public sources.
The Deal Flow Index (DFI) incorporates active transaction data as a proprietary signal —
adjusting the projected floor based on whether LEAPEF's observed deal prices are
above or below public market rates.
Individual transaction data is never disclosed externally.
The DFI value and anonymised deal markers are the only external outputs.
The underlying contracts are PIN-gated and session-only — never stored or transmitted.
Important Disclaimer
This tool is produced by LEAPEF AB for internal analytical purposes only. It does not constitute investment advice, a credit rating, a formal asset valuation, or a recommendation to buy or sell any asset. All projections are probabilistic estimates subject to material uncertainty. Past GPU price behaviour does not guarantee future results. GPU markets are subject to rapid technological change, supply disruptions, and demand shifts that may not be captured by historical calibration.
LEAPEF AB is a registered Alternative Investment Fund Manager (AIFM) supervised by Finansinspektionen, Sweden, operating under applicable EU and Swedish regulatory frameworks. March 2026.