Cango pivots Bitcoin mining infrastructure toward AI compute hosting
Cango Inc. (NYSE: CANG), a Bitcoin miner with operations across North America, the Middle East, South America and East Africa, posted Q2 2026 revenues of $50.8 million and a net loss of $81.6 million, as it accelerated a strategic pivot from pure-play mining toward an integrated AI compute infrastructure platform.
The headline loss is largely a bookkeeping artefact. Of the $81.6 million deficit, $42.9 million came from non-cash impairment on ageing mining machines, and a further $8.5 million from disposal losses as the company retired lower-efficiency S19 series rigs. Strip those out alongside depreciation and share-based charges, and the adjusted EBITDA loss narrows to $10.7 million, down sharply from a $154.1 million adjusted loss in Q1 2026.
From hash-rate to GPU rack
The more consequential story is what Cango is building in the space vacated by those retired miners. Its Georgia site completed conversion to AI compute in early July 2026, with capacity to support up to three megawatts of GPU infrastructure. Container units are installed, and GPU hardware is arriving in staged batches. The company says it expects to begin recognising AI-hosting revenue in Q3 2026. Test nodes are also operating in Texas and on the West Coast to serve customers requiring proximity-based deployment.
CEO Paul Yu framed the dual-track ambition plainly: "In the future, we intend to pursue two business models. The first is bare-metal GPU hosting, using our infrastructure to offer a standardised deployment environment. The second is colocation, intended to enhance overall infrastructure utilisation."
On the mining side, total operating hashrate stood at 27.58 EH/s at quarter end, comprising 19.84 EH/s of self-mining capacity and 7.74 EH/s leased. The company mined 656 Bitcoins during the quarter at an average cash cost of $73,313 per coin, a roughly 5% sequential improvement. It held 1,056 BTC in treasury as of 30 June, alongside $10.1 million in cash. A newly launched Bitcoin hedging programme, described by CFO Simon Tang as "strictly a risk management tool, not for speculative purposes", is designed to smooth cash-flow predictability as price volatility persists.
The broader infrastructure convergence
Cango's repositioning sits within a pattern that cross-sector investors are tracking closely: the structural overlap between high-density power infrastructure and AI compute demand. Bitcoin mining operations, built around cheap, dispatchable power and high thermal tolerance, offer a ready-made physical platform for GPU workloads once the economics of mining compress. That compression is real: Cango's Q2 revenue fell approximately 50% sequentially, reflecting both deliberate hashrate reduction and broader mining-industry headwinds following the April 2024 halving cycle.
The company is not alone in attempting this migration. Several North American miners have signalled analogous pivots, creating a new competitive dynamic in the wholesale GPU hosting market where they now jostle with hyperscaler-adjacent colocation providers and purpose-built AI data centre developers. For capital allocators, the critical question is timing and scale: three megawatts in Georgia is a proof-of-concept, not a platform. The company's balance sheet, with total assets having contracted from $1.13 billion at year-end 2025 to $294 million by 30 June 2026, leaves limited headroom for aggressive capital deployment without additional fundraising.
Geographically, the global spread of Cango's mining footprint across the GCC, East Africa and the Americas gives it optionality to route AI workloads to jurisdictions where power costs, regulatory posture and latency requirements align. That geographic arbitrage is increasingly a differentiator as sovereign and enterprise AI buyers grow more sensitive to data residency and energy provenance. Whether Cango can execute the transition at meaningful scale before its cash position tightens further will be the defining question for Q3 and beyond.