Cango (CANG) Q4 2025 Earnings Call Transcript

Cango reported a $622.0M net loss attributable to shareholders for 2025, driven by $338.3M of mining-machine impairments and sizable fair-value losses; full-year revenue was $688.1M with adjusted EBITDA of $24.5M. The company mined 6,594.6 BTC in 2025 (1,718.3 BTC in Q4), reached a peak hashrate of 50 EH/s, and sold 4,451 BTC in Feb 2026 to repay loans; year-end cash was $41.2M versus $557.6M of long-term debt. Management is pivoting toward AI with EcoHash and a 1–2 MW Georgia pilot (~$20M per MW capex) and will prioritize fleet efficiency over hashrate expansion, but execution and financing risks remain until pledged equity funding (~$65M) completes.

Analysis

Management’s shift from single-asset accumulation toward running a flexible compute platform creates a structural optionality play: power contracts, colocation rights and site-level permitting can be retasked from ASIC-dominated workloads to GPU inference, converting otherwise stranded energy exposure into multiple payors. That optionality has value only if the company can close the loop on three things simultaneously — obtain low-cost capital to fund GPU-heavy retrofits, secure long-term demand for low-latency inference, and preserve liquidity while decommissioning legacy equipment — otherwise the conversion becomes a capital sink with impaired returns.

A material second-order effect is on the capital markets for collateralized lending and secondary hardware markets. GPUs and cloud-equivalent racks are more fungible collateral than specialized ASICs, enabling asset-backed financing at tighter spreads, while accelerated ASIC disposals will depress used-miner prices and compress replacement costs for competitors, tightening industry consolidation dynamics. Separately, demand from miners entering AI will compete with hyperscalers for GPU supply, creating a near-term squeeze on procurement but a longer-term bifurcation where niche, latency-sensitive inference nodes can command higher ASPs than commodity cloud cycles.

Near-term catalysts to watch are (1) pilot commercialization data showing utilization and gross margins for inference nodes, (2) the outcome and timing of committed capital closings, and (3) any reduction in fixed-cost drag via renegotiated power contracts. Tail risks include a macro-led crypto rebound that leaves monetization decisions regretful, a failed pilot that forces asset write-downs, or credit events that trigger forced disposals; each would compress equity value far faster than a slow, controlled pivot would reverse it.

Key Decisions for Investors

• Short CANG equity via puts (6–12 month expiries): buy defined-risk put spreads to express downside from funding/liquidity and execution risk. Position size: tactical (1–3% notional). R/R: limited premium risk vs asymmetric downside if pilot and financing stumble.
• Pair trade — short CANG / long NVDA call spread (9–15 month): short equity to capture structural execution risk while buying NVDA call spreads to hedge and capture secular GPU tailwinds. Entry: after next liquidity milestone or pilot readout. R/R: balanced — downside protection from defined short leg with uncapped but hedged upside in NVDA spread.
• Directional long NVDA (12-month calls or buy-and-hold): buy exposure to GPU pricing upside driven by new entrants and infrastructure demand. Size as hedge to CANG short exposure. R/R: premium paid vs high-conviction capture of secular AI capex cycle.
• Event-monitor & convertible-credit watchlist: set alerts for announced equity closings, pilot go-live metrics, and any covenant waivers — use these triggers to move from options-based protection into directional credit or opportunistic long-equity if financing completes on favorable terms. R/R: requires active catalyst monitoring; capital deployment conditional on tightened balance-sheet signals.