CGAU — Centerra Gold Inc.

Deep-value analysis — CGAU (Centerra Gold Inc.)

Date: September 4, 2026 · Price: … (NYSE American) · Market cap: … · Shares: 196,138,619 (June 30, 2026, note 13 FS Q2) · Reporting currency = price currency: USD

Regime: REFRESH. The baseline analysis is from July 27, 2026 (verdict SPECULATIVE). The mechanical deep_delta.py triage triggered the resumption on two broken falsifiers — pe (6.96 < threshold 7) and roe (29.4% > threshold 25%). Delta pack: rapoarte/deep/delta-CGAU-20260903.md.

What is explicitly carried over from the baseline report (not re-derived, because nothing in the new filing moves it): the Kumtor history and the definitive closure of the Kyrgyz file; the structure of the Royal Gold streaming agreement; the Mount Milligan PFS parameters (life extended to 2045, Sept. 2025); the Kemess PEA parameters from January 19, 2026 (initial capex …M, NPV ~…, IRR 16% at …/oz); the composition of the management team and appointment dates; the NCIB structure and TSX ceilings; the timeline of the 2025 impairment reversals (…M Goldfield in Q3, …M Kemess in Q4).

What has been re-derived from scratch, because the new filing moved it: the entire owner-earnings base (the old report used a normalized gold price of …/oz, incompatible with the company’s own …/oz deck published July 28, 2026); the FCF bridge on a trailing-four-quarter basis from the consolidated cash flow statement; the balance sheet line by line, June 2026 vs December 2025; the Quality of Earnings chapter (entirely missing from the old report); the Thorndike CEO profile (entirely missing); all five valuation models and the Monte Carlo simulation. Three data gaps flagged in the old report are now closed — management ownership, the outcome of the H.R.S. dispute, and the 2023-2025 profitability history (see the respective chapters).

The price move is not the thesis. The price rose from … (July 27) to … (…). Nothing here is built around that fact; it changes the margin of safety, not the business.

Executive summary

Centerra Gold is a mid-tier gold-copper producer with two operating mines — Mount Milligan (British Columbia, 100%, gold-copper, mine life extended to 2045 via the September 2025 PFS) and Öksüt (Türkiye, 100%, heap-leach, reserves depleting ~2029) — plus a molybdenum business under reconstruction: Thompson Creek (Idaho, restart decided September 2024, first production mid-2027) and the Langeloth refinery (Pennsylvania). The last two quarters have consolidated a third growth leg: the Goldfield project (Nevada) under construction, with first production at end-2028.

The thesis, in one sentence: at today’s price, the market is paying for CGAU as if gold stayed perpetually at …/oz and all three growth projects delivered on budget — and the business has no structural protection if gold doesn’t stay there. This isn’t a thesis about bad accounting or incompetent management; it’s a thesis about extreme operating leverage in a business with no moat.

The figures supporting this, all from the Q2 2026 filing (6-K from July 28, 2026, exhibits 99.1 and 99.2, accession 0001628280-26-050229, downloaded today from EDGAR):

• The company’s own 2026 deck is gold …/oz and copper …/lb (Q2 MD&A, note 2 to the guidance table). The average market gold price in Q2 2026 was …/oz, and in the first half …/oz. Consolidated realized price: …/oz in Q2, …/oz in H1 — the difference is the Royal Gold stream, which takes 35% of Mount Milligan’s gold at a fixed …/oz.
• Consolidated AISC by-product: …/oz in H1 2026, guidance …-1,750/oz — Mount Milligan …/oz (cheap, thanks to copper credits at …/lb), Öksüt …/oz (expensive). The gross margin per ounce today is ~…; at a gold price of …/oz it compresses to ~…/oz. A 25% drop in the gold price cuts owner earnings by ~58%.
• The Turkish royalty at Öksüt is …-700/ounce at a gold price of …/oz (Q2 MD&A, Cost Profile section) — i.e. the Turkish state collects ~15% of the market price, on a sliding scale that rises with price. Government royalties paid rose from …M (H1 2025) to …M (H1 2026). The price upside at Öksüt is shared with Ankara.
• The capital program consumes all the cash flow. 2026 guidance: …-230M gold/copper capex + …-220M Thompson Creek + …-5M Langeloth + …-10M Kemess = …-465M. Thompson Creek alone has swallowed …M since the restart decision and still needs ~…-195M (…-450M total to first production). The result: FCF of …M in H1 2026 and a …M deficit in Q2 — at the best gold price in history.
• Book earnings are inflated, but less than they appear. TTM net income of …M still contains …M of non-cash impairment reversals from Q3/Q4 2025. TTM normalized earnings (the sum of quarterly “Normalized Income” from the data pack) is …M, i.e. diluted normalized EPS of ~… and normalized P/E of 14.3x — not 7.1x as the screen shows. Consensus (12 analysts, stockanalysis.com, August 19, 2026) gives 2026 EPS of … and 2027 of …; at … that’s a forward P/E of 14.1x and 11.9x respectively. This is no longer a “cheap stock.”

Estimated value (triangulation, 5 models — see the dedicated chapter): range … – …/share, with the models’ median at … and the central case (owner earnings normalized at gold …/oz) at …. Monte Carlo over 20,000 scenarios on my assumptions gives a median intrinsic value of … and a probability of undervaluation of 0.3%.

Verdict: SPECULATIVE — avoid at the current price. The label doesn’t change from July 27, but the margin of safety is markedly worse. Then, at … the price sat within the old report’s …-18 base case. Today, at … the price is above the top end of all my conservative models and ~20% below the sole bull-type scenario. The position becomes interesting only below …-13/share, or if Thompson Creek publishes a project economics (annual production, cash cost) that justifies the …-450M of capital already committed — today those figures simply aren’t published, which is the file’s largest data gap.