GSL — Global Ship Lease, Inc.

Deep-value analysis — Global Ship Lease, Inc. (NYSE: GSL) · REFRESH

Analysis date: September 7, 2026 · Reference price: … (close of September 4, 2026, tracker indicators 06.09.2026) · Market cap: … on basic shares (… on H1 2026 diluted average) · Shares: 36,035,434 Class A outstanding at 06/30/2026; 36,652,419 H1 2026 diluted average

Regime: REFRESH over the thesis from August 16, 2026. The mechanical triage deep_delta.py (06.09.2026) returned REFRESH for a single reason: one of the thesis’ five falsifiers — free cash flow yield — could not be calculated mechanically, it wasn’t breached. The other four (net debt/EBITDA, P/B, P/E, ROE) all hold. There is no new SEC filing after August 6, 2026: the last document is the 6-K with the H1 2026 interim report, which was already the main source for the August analysis. This report therefore explicitly inherits the operational facts and re-derives only (a) the broken falsifier, verified manually from filings, (b) the entire valuation at the new price, (c) three things I found wrong or incomplete in the previous report.

Primary sources: 20-F FY2025 filed 16.03.2026 (PwC-audited statements S.A.); 6-K from 06.08.2026 (H1 2026 interim report, unaudited); 20-F FY2022/FY2023/FY2024 for historical buyback and share-capital series; MSI market data cited in the 20-F. Secondary, marked as such: research brief 07.09.2026, GBL tracker (indicators 06.09.2026), data pack 07.09.2026, Monte Carlo simulation mc-GSL-20260907.json. Figures are in thousand USD unless otherwise noted; “mil.” = million USD.

Executive summary

What I inherit, without re-derivation. The business model (mid-size containership lessor, 71 operating vessels, 423,020 TEU, TEU-weighted average age 18.4 years at 06/30/2026 — 6-K, note 3), the external management structure (Technomar technical, Conchart commercial, both controlled by the Executive Chairman), the earnings-quality conclusion (HIGH, with three reservations), the six accounting red flags, the three pre-mortem scenarios, and — most importantly — the free cash flow bridge and the … owner-earnings base. None of these elements has a new filing behind it; rewriting them would mean inventing movement where there is none.

What I re-derived. The price rose from … (the August report’s reference) to … …; versus the tracker quote at the thesis date (…) it’s …. The price move isn’t the reason for the revisit and I don’t build the thesis around it, but it mechanically consumes the safety margin of every valuation model, so all five models were recalculated at … and the Monte Carlo simulation was rerun with the same assumptions and the new price. Comparable multiples were refreshed from the tracker (06.09.2026). The broken falsifier was verified manually from the cash flow statement.

What I found wrong in my own August report — three corrections. First: the buyback history was truncated. The August report started from 2023 and counted …; the reality, from the 20-F FY2022 and FY2025, is that the program bought 2,555,075 shares for about … from March 2022 through 2024 (1,060,640 at … in 2022; 1,242,663 at … in 2023; 251,772 at … in 2024), with … remaining authorized of the … The research brief was right, my report wasn’t. Second, and more important: placed alongside the incentive-plan issuances (747,604 shares in 2021, 586,819 in 2022, 440,698 in 2023, 483,713 in 2024, 466,258 in 2025 — 6-K, note 10), the … buyback did not reduce the share count; it rose from 35,990,288 (Dec. 2022) to 36,035,434 (June 2026). Third: the ATM program was drawn once — 27,106 shares at … in 2024, under the August 2024 program (20-F FY2025, note 16). The August report said no ATM had been drawn; the correct statement was that the programs from September 2025 hadn’t been drawn.

Estimated value. Triangulation across the same five models, with the safety margin recalculated against …: DCF FCFE bear (partial rate reversion plus full renewal reserve) …; comparable multiples on normalized profit …; base-case DCF FCFE, the Monte Carlo median across 20,000 scenarios, … (median IV …); NAV with the fleet at 1.25× book value …; EPV Greenwald with economic depreciation …. Intrinsic value range: …-69/share, centered around ~… — unchanged, because the value hasn’t moved; the price has. The probability of undervaluation fell from …% to …%, and the pessimistic tail of the distribution (P10) moved from … to …. In other words: now one in ten scenarios loses over 8% of value, versus none three weeks ago.

Verdict. BUY with a moderate position, halved from August, and no adding above … The substantive arguments are intact: net debt of … (0.05× EBITDA), 21 unmortgaged vessels, average cost of debt 4.43%, dividend of 5.39% covered over four times, a proven history of selling old hulls at 2.0-2.8× book value. What’s worsened is purely the price-to-value ratio: at … the comparables model said “fairly priced” (…) and the bear model said …; at … they say … and …. Two of the five methodologies are now below price. And the correction on buybacks versus share issuance worsens the capital-allocation grade, not improves it: the company spent … to leave its share count unchanged. This remains a value position with a fortress balance sheet, not a conviction position.